Public Kempax report

Daily Update Global Data Center and GPU capacity · 2026-10-06

Kempax Research

Loop report · 2026-10-06

Global data-center and GPU capacity — 6 October 2026

As of 6 October 2026 UTC, the report remains a bounded, source-qualified snapshot rather than a defensible global total. The 5 October evidence ledger contains 12 project or platform records and 7 GPU or hosted-service records across six regions; the 6 October premarket overlay adds no project or inventory rows. It retains Crusoe’s separate corporate capacity and financing disclosure, AWS capacity-block rental checkpoints, and July 2026 EIA industrial retail-revenue proxies. IT load, site facility compute capacity, grid or generation evidence, GPU state, company roles, investor or funder disclosures, and price or cost objects remain separate. No global MW, grid capacity, installed-GPU, invested-capital, or comparable purchase-price total is asserted or summed; ND means not disclosed or not verified, not zero.

Current-run update — 2 September 2026

This additive update keeps the accepted baseline intact and records the fresh evidence packet below. The nine data-center and power records and two accelerator catalog records are selected public signals, not a global census. ND means not disclosed or not verified in the reviewed evidence, not zero. IT load, facility or campus power, grid interconnection, generation PPA, cloud orderability, installed GPU inventory, and economic objects remain separate; no unlike or overlapping measures are summed.

Data-center plans, construction, power, and ownership

Record location Source date and state Capacity and electricity basis GPU inventory specification Company, developer, operator, investor or financing role Price economic object Direct evidence --- --- --- --- --- --- --- DC-01 · Cerebras Compute Nordic — Mikkeli, Finland 1 Sep 2026 release dateline; initial 50 MW phase reported under construction; contract not reported energized 165 MW contracted IT capacity delivered in phases of 50 MW, 80 MW, and 165 MW; seven-year service orders; grid interconnection, generation PPA, and annual-energy basis ND No GPU count or model; source describes Cerebras wafer-scale AI compute, not a GPU inventory Cerebras is long-term customer platform; Compute Nordic Finland is development partner; project equity, lender, and investor roles ND €1.0–1.7B indicative regional investment at full scale; estimate, not stated committed capex, rent, or GPU price Cerebras IR release — page index labels 31 Aug while release dateline says 1 Sep; discrepancy retained DC-02 · Bitdeer AI A102 — Malaysia 31 Aug page label 1 Sep release dateline; sold out under offtake commitments ahead of Q1 2027 energization ; not reported operating Approximately 9.5 MW data-center capacity ; IT boundary not stated. Separate Bitdeer target of up to 350 MW AI-cloud capacity by Q1 2028 is not additive; grid interconnection, generation PPA, and tariff ND Planned rack-scale NVIDIA GB300 NVL72 deployment; installed, delivered, ordered, and committed count ND Bitdeer AI, part of Bitdeer Technologies Group, is AI-cloud provider developer; customers unnamed. More than 50% of associated capex is expected from customer prepayments; other financing names ND More than US$800M total expected revenue over five-year offtake commitments; contract service revenue expected from Q1 2027, not capex, rent, or hardware price Bitdeer A102 release DC-03 · Bitdeer new parcel Rockdale — Milam County, Texas, US Fee-simple greenfield acquisition completed and announced 1 Sep 2026; no construction or energization stated 200 acres acquired. Company context cites about 563 MW interconnected grid capacity and plans expected to scale to 742 MW; post-transaction statement cites about 255 acres and 742 MW existing pipeline power in the county. These are portfolio context, not capacity assigned to the new parcel No GPU count, delivery state, or model; AI HPC development intent only Bitdeer is owner, developer, and infrastructure operator; paid in cash; external investor, lender, and project-finance roles ND Approximately US$100M cash land-acquisition consideration ; land transaction value, not data-center capex, lease value, tariff, or GPU price Bitdeer land-acquisition release DC-04 · Digital Vault Sulcis — Nuraxi Figus, Sardinia, Italy National-strategic designation announced 1 Sep 2026; accelerated administrative pathway; not a permit, construction, energization, or operating claim Initial Edge AI phase up to approximately 30 MW over 24 months , scaling toward 50–60 MW IT and up to 100 MW total site power ; renewable generation and storage integrated; utility import limit, generation nameplate, PPA, and annual energy ND GPU count, model, and allocation ND Energy Vault is developer and Build, Own & Operate platform; Italian government designation and Autonomous Region of Sardinia support; project equity, lender, and financing roles ND Project capex and financing ND; no price inferred from the designation BusinessWire-syndicated release · Energy Vault newsroom — direct BusinessWire page unavailable in the evidence packet DC-05 · Fervo Energy Google Cape Station — Utah, US PPA announced 1 Sep 2026; Cape Station EGS GeoCluster expected online in 2028; potential data-center plans remain subject to approvals and commercial conditions 396 MW geothermal PPA is generation offtake capacity, not data-center IT load. Google has an option for approximately 600 MW additional offtake, nearly 1 GW by June 2030; option is not committed current capacity. Grid interconnection and site IT load ND No GPU count or model; no operating data-center inventory claimed Fervo is geothermal developer generator; Google is offtaker and potential data-center anchor; investor, lender, and project-finance roles ND PPA price, project capex, tariff, and colocation cloud price ND Fervo IR release DC-06 · FuelCell Energy unnamed operator — Texas, US First capacity-reservation agreement reported 2 Sep 2026; planned project; no construction, energization, operator identity, or delivery date stated 75 MW planned project ; source does not specify whether this is IT load, generation, or total facility capacity. Grid interconnection, PPA, and tariff ND GPU count and model ND FuelCell Energy is supplier; major data-center operator is unnamed; investor, lender, and project-finance roles ND Upfront reservation payment exists, but amount and terms are ND; reservation signal, not project capex or GPU price FuelCell Energy Q3 2026 release DC-07 · FuelCell Energy Fit Energy — US data-center power supply Agreement reported 2 Sep 2026; four potential phases; not operating or energized 30 MW committed fuel-cell generation plus an option for up to 350 MW additional generation ; source describes up to 380 MW across both. Generation is not IT load; site allocation, grid, and PPA ND No GPU count or model; end data-center operator unnamed FuelCell Energy is fuel-cell system supplier; Fit Energy USA LP is purchaser option holder; lender and equity-investor roles ND US$1.3B committed backlog includes the 30 MW commitment; US$2.4B awarded-capacity backlog relates to the 350 MW option. Backlog is not project capex, rent, or hardware price FuelCell Energy Q3 2026 release DC-08 · Hyperscale Data Michigan facility — US Bitcoin mining ceased effective 1 Sep 2026 ; facility is in AI engineering and equipment-procurement preparation; AI energization and operation not reported Initial MSA provides 20 MW critical AI compute capacity and a right to an additional 32 MW . Company describes approximately 340 MW eventual facility power capacity ; this is not current IT load and is not added to the MSA figures GPU count, model, delivery state, and allocation ND; California neocloud customer unnamed Hyperscale Data is facility owner operator and AI-conversion developer; California neocloud is customer; investor, lender, and equipment-finance roles ND More than US$1.2B expected revenue for the initial 10-year MSA plus options; more than US$3.0B at maximum term and option scope. Expected service lease revenue, not capex or GPU price Hyperscale Data release DC-09 · KIDZ AI HUADING national cloud data center — Phnom Penh, Cambodia Company-distributed release dated 2 Sep 2026 reports the Tier III facility operational and energized; KIDZ transaction remains subject to customary closing conditions Pending transaction includes a three-year right of first refusal to lease up to 1 MW of data-center load ; this is contractual access, not current leased or installed load. Exact grid MW, MVA, generation, and PPA ND AI GPU deployments described generically; GPU count, model, and installed state ND KIDZ AI is proposed 2.0% minority equity holder and strategic partner; HUADING DEVELOPMENT CAMBODIA CO., LTD. is owner operator; Cambodian government national-cloud designation; consideration, lender, and other investor roles ND Acquisition consideration, lease rent, capex, tariff, and GPU price ND KIDZ AI company-distributed release

GPU inventory, model specifications, and current price checkpoint

Record Quantity state Model specification Company, vendor, investor or financing role Possible price with original unit, type, and date Direct evidence --- --- --- --- --- --- GPU-01 · NVIDIA GB300 NVL72 NVIDIA lists the platform as Available Now ; Bitdeer A102 identifies planned rack-scale deployments. No installed, delivered, ordered, committed, or allocated site count is disclosed Fully liquid-cooled rack-scale platform with 72 NVIDIA Blackwell Ultra GPUs and 36 Arm-based NVIDIA Grace CPUs ; NVIDIA also describes 800 Gb s networking per GPU. Vendor specification is not a project inventory count NVIDIA is OEM platform provider; Bitdeer A102 is intended deployment context; purchaser, investor, and financing roles ND OEM purchase price and rental price ND; no unit-price date verified NVIDIA GB300 NVL72 · Bitdeer A102 GPU-02 · Lambda on-demand catalog Current catalog offers 1x, 2x, 4x, and 8x GPU instances; 1-Click Clusters offer 16–2,000+ interconnected B200 or H100 GPUs . This is orderability offer capacity, not a deployed fleet count NVIDIA B200 SXM6: 180 GB VRAM GPU ; NVIDIA H100 SXM: 80 GB VRAM GPU ; catalog also lists A100 and V100 Lambda is cloud operator; NVIDIA is OEM; investor, lender, and project-allocation roles ND US$6.69 per B200 GPU-hour and US$3.99 per H100 GPU-hour ; hosted rental list rates, catalog captured 2 Sep 2026, no page publication date; applicable sales tax VAT GST may apply Lambda GPU instances · NVIDIA B200 · NVIDIA H100

The current packet does not verify an OEM purchase price or executed hardware transaction price for GB300, B200, H100, H200, Rubin, or MI350X. Cloud rates are not hardware prices, installed-inventory counts, electricity tariffs, or project capex.

Economic and price signals

Record Value currency Unit Economic type Date Boundary and non-additivity control Direct evidence --- --- --- --- --- --- --- Cerebras Compute Nordic Mikkeli €1.0–1.7B Total regional investment estimate at full scale Indicative investment estimate 1 Sep 2026 release dateline; captured 2 Sep Not stated committed project capex, rent, or GPU price Cerebras IR release Bitdeer A102 Malaysia More than US$800M Total expected revenue over five years Contracted AI-cloud offtake service revenue 1 Sep 2026 release dateline; revenue expected from Q1 2027 Not capex, colocation rent, tariff, or hardware price; not added to the separate 350 MW target Bitdeer A102 release Bitdeer Rockdale land acquisition Approximately US$100M Cash land-acquisition consideration Completed land transaction 1 Sep 2026 announcement Not data-center construction capex or GPU price; parcel-specific capacity ND Bitdeer land-acquisition release FuelCell Energy Fit Energy commitment US$1.3B Committed backlog associated with 30 MW generation Company backlog economic signal 2 Sep 2026 Q3 release Backlog is not project capex, rent, tariff, or hardware price FuelCell Energy Q3 2026 release FuelCell Energy Fit Energy option US$2.4B Awarded-capacity backlog associated with up to 350 MW optional generation Company backlog economic signal 2 Sep 2026 Q3 release Option is not committed generation and backlog is not project capex or GPU price FuelCell Energy Q3 2026 release Hyperscale Data Michigan MSA More than US$1.2B Expected revenue over 10-year MSA plus options Expected AI data-center service lease revenue 1 Sep 2026 release Not current revenue, facility capex, or GPU price; 32 MW option and 340 MW eventual power are not summed Hyperscale Data release Hyperscale Data maximum MSA scope More than US$3.0B Expected revenue including 32 MW option and maximum term Expected AI data-center service lease revenue 1 Sep 2026 release Not capex, realized revenue, or GPU price; it is an expanded contractual scenario Hyperscale Data release Lambda B200 catalog US$6.69 Per GPU-hour Hosted rental list rate Catalog capture 2 Sep 2026; page publication date ND Dynamic provider rate; taxes may apply; not hardware purchase price or project economics Lambda GPU instances Lambda H100 catalog US$3.99 Per GPU-hour Hosted rental list rate Catalog capture 2 Sep 2026; page publication date ND Dynamic provider rate; taxes may apply; not hardware purchase price or project economics Lambda GPU instances Digital Vault Sulcis ND Project capex and financing No accepted price verified 1 Sep 2026 designation No price inferred from an approval pathway BusinessWire-syndicated release Fervo Energy Google Cape Station ND PPA price, project capex, and tariff No accepted price verified 1 Sep 2026 PPA announcement 396 MW is generation offtake capacity, not an IT-load price basis Fervo IR release

Methodology and figure discipline

• The immutable accepted report baseline at generation 414 and commit e8de9024815ca6ac46576a5d56ab2ae04135358e was read before editing. Its complete kempax-ui block remains byte-for-byte unchanged; the current-run material is appended outside it.
• The current evidence packet was captured on 2 September 2026 UTC. Direct company, investor-relations, vendor, and distributed-release URLs are kept beside the relevant claims. The Energy Vault row identifies its syndicated-source limitation; the KIDZ row identifies its company-distributed release.
• Capacity is kept in the source-defined boundary: contracted IT, critical AI compute, data-center capacity, facility site power, grid interconnection, generation, and PPA are not substituted. No MVA-to-MW or energy-to-power conversion is made.
• GPU states remain distinct: available-now product, planned deployment, catalog orderability, installed, delivered, ordered, committed, allocated, and potential are not interchangeable. A rack specification is not a site inventory count.
• Economic objects remain distinct: investment estimate, land consideration, backlog, expected service revenue, reservation payment, PPA price, cloud rental, and hardware purchase price are not combined. Original currency, unit, type, and source date are retained; no FX conversion is made.
• Parent targets, child sites, options, and portfolio context are not added. Examples include Bitdeer’s 350 MW target, Milam County’s 742 MW context, Fervo’s additional 600 MW option, Hyperscale’s 340 MW eventual facility potential, and Fit Energy’s optional generation.
• ND means not disclosed or not verified in the reviewed evidence; it does not mean zero. Company claims remain source-qualified and are not promoted to independent audits.

Uncertainty and unresolved risks

Risk Retained treatment Evidence and implication --- --- --- Date-boundary conflict Cerebras and Bitdeer A102 retain both page index and release-dateline dates Cerebras and Bitdeer A102; do not silently backdate or forward-date milestones Status risk Announced, contracted, land-acquired, designated, PPA, and AI-preparation states are not upgraded to construction, energization, operation, or delivery Energy Vault, Fervo Google, and Hyperscale Data show why status is material Boundary risk 75 MW FuelCell project basis and 9.5 MW Bitdeer A102 IT boundary remain ND; generation PPA is not IT load FuelCell Energy, Bitdeer A102, and Fervo Google support separate treatment GPU inventory risk GB300 rack specifications and Lambda cluster offers do not establish installed, delivered, ordered, or globally available site stock NVIDIA GB300 and Lambda; no current project GPU count is manufactured Ownership and financing risk Customers, project investors, lenders, and counterparties remain unnamed where the source does not name them; supplier, government, and offtaker roles are not relabeled as equity ownership Bitdeer A102, FuelCell Energy, and KIDZ AI Source-access risk The Energy Vault claim uses a BusinessWire-labeled syndication because the direct page was unavailable; KIDZ uses a company-distributed Newswire release Syndicated Energy Vault release and KIDZ release; neither is treated as an independent facility audit Price risk Lambda rates are provider-listed rental snapshots; no comparable OEM hardware quote, project electricity tariff, PPA price, or GPU-cost-per-MW estimate is verified Lambda, Fervo Google, and NVIDIA GB300 Census risk The current-run records are a bounded selection and do not support a global total The report retains the selected rows and baseline context without adding unlike or overlapping measures

Milestone watchlist

Watch item Verification target Direct evidence --- --- --- Finland and Malaysia commissioning Confirm construction progress, energization, first service, and the date discrepancy for the 165 MW Finland phases and 9.5 MW A102 site Cerebras Compute Nordic · Bitdeer A102 Rockdale parcel Obtain parcel-specific permits, interconnection allocation, and construction milestone; do not reuse Milam County portfolio power as parcel capacity Bitdeer land acquisition Sulcis and Cape Station Check authorization, site-load definition, PPA economics, and actual generation data-center coupling Energy Vault · Fervo Google FuelCell agreements Verify the 75 MW boundary, named operator, reservation-payment terms, Fit Energy phase schedule, and delivered generation FuelCell Energy Q3 2026 release Michigan AI conversion Confirm MSA load actually energized, customer identity when disclosed, equipment delivery, and whether any of the 340 MW eventual potential is contracted Hyperscale Data release Cambodia transaction Check closing of the 2.0% minority stake and whether any portion of the 1 MW lease right is executed or energized KIDZ AI release GPU and price validation Recheck GB300 deployment counts and Lambda rates; obtain a named OEM or executed procurement quote before treating any price as hardware cost NVIDIA GB300 · Lambda

Current-run update - 3 September 2026

This additive update preserves the accepted baseline and appends the latest evidence packet. Four new data-center, power, or manufacturing records, nine accelerator and inventory records, normalized product specifications, and current provider-rate checkpoints are shown below. ND means not disclosed or not verified in the reviewed evidence, not zero. IT load, facility or campus power, grid or interconnection, generation or PPA, manufacturing throughput, cloud orderability, installed inventory, and economic objects remain separate. No unlike or overlapping measures are summed.

New data-center, power, and manufacturing signals

Record location Source date and state Capacity and electricity boundary Company, developer, operator, investor or financing role Price economic object Direct evidence --- --- --- --- --- --- DC-12 · Linea Energy Google Duffy Solar - Matagorda County, Texas, US Fifteen-year PPA announced May 2026; solar construction scheduled for Q3 2026; co-located battery storage reported under construction Site IT load ND. 500 MW solar PPA and 235 MWac Duffy BESS are generation and storage values, not site IT load or interconnection capacity Linea develops, owns, and operates clean-energy infrastructure; EnCap Investments and EnCap Energy Transition sponsor Linea; Google is offtaker; project financing ND PPA price and project capex ND Linea Google Duffy PPA DC-16 · Desert Technologies Swiss CTS Group CTS-DT - Saudi Arabia Joint venture announced 2 Sep 2026 at LEAP; no construction, energization, or operating site evidence 300 MW year initial manufacturing throughput for advanced AI data-center solutions, scaling toward 1.2 GW year . This is manufacturing output, not installed data-center capacity. Site IT, grid, generation, PPA, and tariff ND Desert Technologies and Swiss CTS Group establish CTS-DT; JV ownership split, investors, lenders, and customer or site allocation ND Capex, financing, contract value, and price ND MENAFN Mid-East Info · CTS-DT DC-17 · Kasi Cloud LOS1 Lekki campus - Lagos, Nigeria Commissioning and operational-readiness ceremony reported 19 May 2026; staged campus; full current operating state ND Approximately 100 MW critical IT capacity at full development ; LOS1 phase load ND. Grid, interconnection, generation, and PPA MW ND Kasi Cloud Datacenters develops and operates; government and NSIA participation is reported around commissioning; equity and lender allocation ND Project capex, financing, lease, tariff, and GPU price ND. The release’s US$850M annual offshore-cloud-spend figure is market context, not project capital Kasi commissioning release DC-18 · iXAfrica NBOX1 campus - Nairobi, Kenya NBOX1.1 reported live; NBOX1.2 is the planned full-site phase; page captured 3 Sep 2026 4.5 MW live IT power and 22.5 MW full-site IT power are nested. Initial feeders are 2 x 4 MVA; planned substation is 2 x 40 MVA; MVA is retained without conversion iXAfrica operates; Future-tech provided design expertise; Helios Investment Partners supplied earlier growth capital; site financing allocation ND US$50M growth-capital signal from 2022; not current site capex or GPU price iXAfrica campus specifications

Accelerator inventory, state, and company roles

Record Quantity and state Capacity or electricity boundary Company, operator, OEM, investor or financing role Price signal and control Direct evidence --- --- --- --- --- --- GI-01 · xAI Colossus - United States 200,000 H100 GPUs in one interconnected cluster; operating, first-party claim captured 3 Sep 2026. A one-million-GPU roadmap is future context, not installed stock MW and IT load ND on the current page xAI SpaceXAI operator; NVIDIA OEM; investor and financier roles ND on the cited page Price ND; do not treat the company claim as an independent audit xAI Colossus GI-02 · AWS Project Rainier Nearly 500,000 Trainium2 chips in service; operating, first-party AWS claim. Trainium2 is a custom accelerator, not an NVIDIA GPU Site power ND AWS Amazon operator and builder; Anthropic collaborator and user; investor and financier roles ND AWS cites 30-40% better price-performance than certain GPU instances, but no comparable dollar rate; exclude from NVIDIA GPU totals AWS Project Rainier GI-03 · Yotta D2 - Greater Noida, India 20,736 NVIDIA B300 GPUs ; committed or planned. Later release targets delivery and completion by 30 Sep 2026 ; completion was not verified at capture D2 60 MW , scalable to 250 MW ; separate Navi Mumbai scale signal is not assigned to D2 Yotta operator; NVIDIA DGX Cloud relationship; Gorilla commercial and infrastructure partner; Hiranandani Group backing or leadership relationship indicated by Yotta; equity split and financing allocation ND Approx. US$2.8B aggregate commercial framework, not a card price or proof of completion Yotta Gorilla GI-04 · Noetra national AI infrastructure - Japan 27,500 Rubin GPUs plus 13,750 Vera CPUs ; announced and planned, not verified installed 140 MW data-center capacity ; not stated as per-GPU IT load Noetra operator and establisher; NVIDIA OEM and architect; METI support; investor and financing amount ND OEM price ND; architecture quantity is not live inventory NVIDIA Japan GI-05 · DataSection AI data center - Thailand 4,696 B200 GPUs in 587 server systems ; acquired or contracted, with planned Jul-Aug 2026 delivery; live status ND Site power and capacity ND DataSection operator and buyer; ASRock Rack server OEM; unnamed US customer; customer advances, scheduled loans, and share-subscription rights are disclosed; external investor allocation not verified Estimated fixed-asset acquisition US$257.4M approximately ¥41.13B for server systems, not card-only price DataSection count · JapanIR disclosure mirror · BigGo disclosure mirror GI-06 · National supercomputer access program - Israel 1,000 NVIDIA B200 accelerators over coming years; allocated program capacity, not necessarily cards installed at one site Site power and capacity ND Nebius provider and operator selected by competitive tender; Israel Innovation Authority and Telem government program; investor role ND Discounted versus market is stated; dollar price ND Israel Innovation Authority GI-07 · Stargate Norway - Nscale Aker 100,000 NVIDIA GPUs targeted by end-2026 ; a separate agreement adds more than 30,000 Rubin GPUs for Microsoft in 2027. Targets and commitments are not added or treated as installed 230 MW initial planned capacity, with ambition for another 290 MW ; not labelled as IT load Nscale designer and builder; planned 50 50 Nscale-Aker JV subject to closing; OpenAI initial offtaker; Microsoft later customer; NVIDIA OEM Approximately US$1B initial 20 MW commitment including more than US$250M equity; project financing, not GPU price Nscale Norway · Nscale Microsoft GI-08 · Cassava AI Factories - South Africa and planned African expansion NVIDIA-powered AI factories confirmed in South Africa with planned expansion to Nigeria, Kenya, Egypt, and Morocco; model, count, delivery state, and site allocation ND MW and site capacity ND in the current official release Cassava Technologies operator; NVIDIA platform and OEM relationship; project investor and financing allocation ND Price ND; secondary quantity or investment claims are not promoted without first-party confirmation Cassava AI factory announcement GI-09 · Stargate UAE - G42 OpenAI Oracle NVIDIA Grace Blackwell GB300 systems; GPU count ND; announced and planned 1 GW cluster , with 200 MW expected live in 2026 ; source boundary is not an installed GPU or IT-load count G42 builds; OpenAI and Oracle operate; NVIDIA supplies systems; Cisco and SoftBank are named partners; financing amount ND Price ND; official OpenAI page body was unavailable in the current capture, so no GPU count is inferred G42 announcement · OpenAI Stargate UAE

Normalized accelerator and system specifications

Product specifications describe model or system capability. They do not establish how many units an operator owns, has installed, or can deliver to a named site.

Vendor model Published specification Form factor, power, or interconnect Availability and role Price signal Direct evidence --- --- --- --- --- --- NVIDIA H100 SXM PCIe Hopper; 80 GB HBM per GPU. SXM 3 TB s memory bandwidth; PCIe 2 TB s SXM max TDP 700 W ; PCIe max TDP 350 W . SXM NVLink 900 GB s ; PCIe Gen5 NVIDIA OEM; product availability is not physical inventory; operator and investor ND at product level OEM MSRP ND; hosted rental rates appear below H100 product page · H100 datasheet NVIDIA H200 Hopper; 141 GB HBM3e ; 4.8 TB s memory bandwidth; datasheet cites 4 PFLOPS FP8 NVIDIA describes the same power profile as H100; exact variant TDP ND in the retrieved source NVIDIA says available; availability is not a site inventory count OEM MSRP ND; RunPod rental signal below H200 product page · H200 datasheet NVIDIA DGX B200 Eight Blackwell GPUs; 1,440 GB total GPU memory; 64 TB s HBM3e bandwidth; 72 PFLOPS training and 144 PFLOPS inference Approximately 14.3 kW maximum system power; 14.4 TB s aggregate NVLink; 10 RU. These are system values NVIDIA OEM and system vendor; no physical deployment count OEM system price ND; cloud rental signals below DGX B200 · DGX B200 datasheet NVIDIA B300 GB300 Blackwell Ultra B300 GPU: 279 GB HBM3E , up to 20 PFLOPS FP4 sparse inference. GB300 NVL72: 72 GPUs plus 36 Grace CPUs , up to 37 TB high-speed memory and up to 1.44 exaFLOPS Fully liquid-cooled rack; 72-GPU NVLink domain; 800 Gb s networking per GPU. DGX B300 is 8 GPUs, 2.1 TB total memory, approximately 14 kW NVIDIA pages describe availability or shipping; not installed inventory OEM price ND; RunPod and project values below are different economic objects GB300 NVL72 · Blackwell Ultra datasheet · DGX B300 NVIDIA Vera Rubin DGX Vera Rubin NVL72 72 Rubin GPUs and 36 Vera CPUs , 75 TB total fast memory, up to 3,600 PFLOPS NVFP4 inference and 2,520 PFLOPS NVFP4 training 100% liquid-cooled rack; sixth-generation NVLink; 72 GPUs in one NVLink domain; per-GPU HBM and TDP ND NVIDIA production-ramp signal; Japan quantity is a planned deployment OEM price ND Vera Rubin product page · Vera Rubin datasheet AMD Instinct MI300X CDNA3; 304 CUs , 192 GB HBM3 , 5.3 TB s peak theoretical bandwidth OAM accelerator; exact board power ND on current page AMD OEM; no public physical-inventory count; operator and investor ND OEM MSRP ND AMD MI300 family AMD Instinct MI325X CDNA3; 304 CUs , 256 GB HBM3E , 6 TB s . Eight-module platform: 2 TB total HBM3E and 48 TB s aggregate bandwidth OAM platform specification; exact board TDP ND AMD product availability is not a deployment count; operator and investor ND OEM MSRP ND AMD MI300 family AMD Instinct MI350X CDNA4; 288 GB HBM3E , 8 TB s ; AMD cites 9.2 PFLOPS FP8 with sparsity and MXFP6 MXFP4 support OAM; exact board TDP ND AMD OEM; no public physical-inventory count; operator and investor ND OEM MSRP ND AMD MI350X AWS Trainium2 Trn2 UltraServer Custom AWS accelerator, not a GPU. Trn2 instance has 16 chips ; UltraServer has 64 chips , up to 6 TB HBM , 83.2 PFLOPS , and 185 TB s bandwidth NeuronLink across four Trn2 instances; AWS says UltraServers are available now AWS Amazon operator and designer; Anthropic is Project Rainier collaborator and user AWS cites relative price-performance, not a comparable dollar quote Trn2 documentation · Project Rainier Google Ironwood TPU7x Custom Google ASIC, not a GPU. Per chip: 2,307 TFLOPS BF16 , 4,614 TFLOPS FP8 , 192 GiB HBM , 7,380 GB s HBM bandwidth, 1,200 GB s bidirectional ICI Up to 9,216 chips per pod configuration; 4-chip VM Google operator and designer; documentation exposes configurations, not installed stock Cloud price ND TPU7x documentation · Ironwood architecture

Current price signals with original units and dates

Record Original value and unit Economic type Date boundary Direct evidence --- --- --- --- --- Lambda rate card H100 SXM 80 GB US$3.99 GPU-hour ; B200 SXM6 180 GB US$6.69 GPU-hour Hosted provider rental list rate Capture 3 Sep 2026; taxes may apply; not hardware purchase price Lambda pricing RunPod Pods rate card H100 PCIe US$2.89 GPU-hour ; H100 SXM US$3.29 GPU-hour ; H200 US$4.59 GPU-hour ; B200 US$6.79 GPU-hour ; B300 US$7.89 GPU-hour Hosted provider rental; community and secure-cloud plans differ Capture 3 Sep 2026; region, plan, and availability can change; not card purchase price RunPod pricing CoreWeave North America rate card HGX H100 US$49.24 , H200 US$50.44 , B200 US$68.80 per 8-GPU instance-hour Hosted provider rental in instance-hour units Capture 3 Sep 2026; not directly comparable with per-GPU-hour rows without explicit normalization CoreWeave pricing DataSection Thailand acquisition Estimated US$257.4M approximately ¥41.13B for 587 server systems containing 4,696 B200 GPUs Fixed-asset acquisition estimate for server systems 3-4 Jun 2026 disclosure; not B200 card-only price and live deployment ND DataSection · TDnet mirror Yotta Gorilla India framework Approximately US$2.8B Aggregate commercial framework or offtake value 30 Apr 2026; 20,736 B300 delivery target, not GPU price or proof of completion Yotta Gorilla OEM price check ND for H100, H200, B200, B300, GB300, Rubin, AMD, AWS, and Google products in the reviewed sources No accepted OEM MSRP or list price Capture 3 Sep 2026; do not back-solve from rental rates, financing, or project values NVIDIA · AMD

Evidence risks retained

Risk Retained treatment Evidence and implication --- --- --- Global census risk The current packet supports selected source-qualified records only. No global MW, electricity, installed-GPU, investment, or comparable-price total is asserted The quantity chart mixes operating claims, targets, allocations, and contracted or acquired records; bars are not additive State and timing risk xAI’s 200,000 H100 value is a current first-party claim; Yotta’s 20,736 B300 target remains unverified as complete at the 3 Sep capture; Nscale’s 100,000 figure remains a target xAI · Yotta · Nscale Boundary risk Manufacturing throughput, generation, storage, MVA, grid power, facility power, IT load, and custom-accelerator chip counts remain separate CTS-DT · Linea Google · iXAfrica Source-access risk CTS-DT capacity is retained as secondary-source evidence because the official landing page did not expose the announcement body. DataSection monetary terms use disclosure mirrors while its count is anchored to the official press page. UAE GPU count remains ND because the official OpenAI page body was unavailable CTS-DT · DataSection · OpenAI Stargate UAE Price risk Provider rates are dynamic hosted rentals; DataSection and Yotta values cover systems or commercial frameworks; no OEM hardware price was verified Lambda · RunPod · CoreWeave Custom accelerator risk AWS Trainium2 and Google Ironwood are included for context but excluded from NVIDIA GPU quantities and totals AWS Trn2 · Google TPU7x

Current-run update — 4 September 2026

This additive update preserves every preceding section and the established report presentation. It adds only the current-run evidence below. Source-defined figures are non-additive unless explicitly comparable: parent programs, phases, options, IT load, facility or campus power, grid interconnection, generation PPA, MVA, manufacturing output, financing, rental rates, and expected revenue are kept separate. ND means not disclosed or not verified in the reviewed source, not zero. A rack specification is not site inventory; cloud rental, backlog, transaction value, PPA, tariff, and expected service revenue are not hardware price or project capex.

Global data-center plans and construction register — current-run delta

The five current-run records are organized by region and stage. The fields intentionally distinguish possible or announced capacity, permitted status, construction, operation, IT load, facility or campus power, grid interconnection, generation PPA, and MVA.

Region record Stage and status date Possible announced capacity Permitted approved Under construction Operating energized IT load Facility campus power Grid interconnection Generation PPA MVA Developer, operator, company roles Explicit investor financing role Capital possible price GPU inventory Direct evidence --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- North America · DC-10 Hut 8 Beacon Point — Nueces County, Texas, US Second phase commercialized by lease in a 20 Jul 2026 release; Phase 1 construction commenced and was financed with investment-grade debt; Phase 2 data-hall delivery expected Q2 2028; no operating or energized state claimed in the July release 352 MW IT second phase; 704 MW contracted IT after Phase 1 and 2; 1,000 MW utility envelope . These are separate, non-additive boundaries ND as a separate permit record Phase 1 construction commenced per company; Phase 2 construction status ND in the July release ND; first-phase release stated initial energization target Q1 2027, with no current confirmation read 352 MW second phase and 704 MW contracted campus IT 1,000 MW campus utility capacity; second AI factory supported by about 500 MW utility capacity; not IT load 1,000 MW utility capacity secured under an interconnection agreement with AEP Texas ; current delivery confirmation ND ND; utility capacity is not generation ND Hut 8 developer operator and infrastructure platform; confidential high-investment-grade tenant; AEP Texas utility interconnection role; NVIDIA DSX design reference only Phase 1 investment-grade debt reported; lender, equity investor, and project-finance allocation ND Company-reported $9.8B 15-year lease for the second phase; $19.6B campus base-term value and potential $50.2B renewal value. Lease contract value, not capex, tariff, or GPU price Count, model allocation, delivery, and operation ND; NVIDIA DSX is architecture intent, not inventory Hut 8 second-phase release — source 20 Jul 2026, captured 4 Sep 2026; first-phase release — source 6 May 2026, captured 4 Sep 2026 North America · DC-11 Chevron Project Kilby — West Texas, US Planned co-located power facility for a Microsoft-operated data center; 22 Jun 2026 release describes a milestone toward FID expected by end-2026, subject to conditions, and first power anticipated in 2028; no construction or energization claim Approximately 2.67 GW generation capacity in phased modular delivery; data-center capacity and campus envelope ND FID not yet reported; permit and approval status ND No construction claim No operating or energized claim; first power anticipated 2028 ND Data-center facility and campus power ND; 2.67 GW is not relabeled as facility or IT power Interconnection agreement, grid import limit, and utility delivery capacity ND; co-location is intended to reduce regional-grid impact 20-year PPA between Energy Forge One LLC, a wholly owned Chevron subsidiary, and Microsoft for dedicated electricity; PPA price ND; GE Vernova and Solar Turbines supply generation equipment ND Energy Forge One Chevron power developer and supplier; Microsoft data-center operator and offtaker; Chevron and Engine No. 1 collaborators; GE Vernova and Solar Turbines equipment suppliers No named lender, equity investor, or project-finance structure; Chevron’s targeted mid-teen return is a company economic claim, not a financing term Mid-teen return target; capex, tariff, lease, and PPA price ND Count, model, state, and allocation ND Chevron Project Kilby release — source 22 Jun 2026, captured 4 Sep 2026 Africa · DC-12 Heka Data US consortium AI hub — Egypt Official Ministry of Communications and Information Technology release dated 3 Sep 2026 reports discussion of initial scope, scale, and implementation phases; further studies and procedures remain; exploratory proposal, not a build Initial scope, scale, phases, site, and possible capacity ND No permit or approval claim No construction claim No energization or operating claim ND IT, facility, and campus power ND MCIT references integrating digital infrastructure with needed power; grid connection, import limit, utility, and interconnection capacity ND Generation, PPA, storage, and tariff ND ND Heka Data named as developer; US consortium unnamed; MCIT, NTRA, and ITIDA officials are government regulatory participants; operator and customer ND No named investor, lender, equity commitment, or project-finance role; the US consortium is not treated as an investor absent disclosure Investment, capex, lease, service revenue, tariff, and hardware price ND Count, model, state, and allocation ND Egyptian MCIT press release — source 3 Sep 2026, captured 4 Sep 2026 Oceania · DC-13 Macquarie Data Centres IC3 Super West — Sydney, Australia Current official pages describe the facility as under construction , with end-state power secured and service advertised for 2026; pages are undated and captured 4 Sep 2026 Project page describes 65 MW potential campus IT load ; facility page body states 47 MW IT load while its header displays “45MW”; discrepancy retained State Significant Development Application in hand; construction certificates listed; approval evidence is not grid delivery Yes , per current project facility pages No operating or energized claim 47 MW IT load per body text; 45 MW header discrepancy; 65 MW potential campus IT load Entire end-state power secured, but MW, MVA, and annual-energy basis ND Utility, grid connection, and interconnection agreement ND; development approval and certificates do not establish power delivery Generation, PPA, storage, and tariff ND ND; page does not provide a convertible electrical rating Macquarie Data Centres developer operator; customers and offtakers ND; design supports hyperscale, cloud, government, enterprise, and AI use Investor, lender, and project-finance role ND on directly read pages Capex, land, lease, service price, tariff, and GPU price ND Count, model, delivery, and allocation ND; cooling modes are described but do not establish inventory IC3 Super West facility page and project page — current pages captured 4 Sep 2026 Latin America · DC-14 RT-One AI data center — Uberlândia, Minas Gerais, Brazil Municipality update dated 23 Feb 2026 says architectural review was the final local step before works and later operation; environmental and other licenses may remain; RT-One company page dated 11 Oct 2025; not reported constructed, energized, or operating 100 MW first phase , expansion up to 400 MW ; source boundary is not defined as IT, facility, campus, or generation Municipal architectural zoning review described; remaining licensing steps possible; no complete permit record verified No construction claim No operating or energized claim ND; 100 400 MW boundary not defined ND; the announced capacity is not relabeled as facility power Power discussions with Cemig ; no interconnection agreement, grid capacity, import limit, or tariff disclosed Company claims 100% renewable energy ; generation nameplate, PPA, storage, and annual energy ND ND RT-One developer; Uberlândia municipality planning zoning authority; Cemig power-discussion counterparty; DMAE water feasibility of 2.77 L s; operator, customer, and vendors ND No named external investor, lender, or project-finance structure More than R$6B announced project investment, original BRL, company claim; not a funding close, tariff, lease, or GPU price Count, model, state, and allocation ND RT-One project page — source 11 Oct 2025; Uberlândia municipal update — source 23 Feb 2026; both captured 4 Sep 2026

Current-run development-cost benchmark

Benchmark Status source date Possible or operating scope IT load power boundary Capital or price signal Control and direct evidence --- --- --- --- --- --- COST-01 · Cushman & Wakefield 2026 US Data Center Development Cost Guide 2026 guide; release date 3 Sep 2026 surfaced in the current evidence; official page captured 4 Sep 2026 US market benchmark, not a named project; pipeline approaching $2.3T , including $492B under construction and $1.8T planned precommitted No project-specific IT, facility, campus, grid, generation, PPA, or MVA value Construction costs up 21% per MW from the prior report; average greenfield cost $17.6M MW for the most modern facilities; original USD, per MW, 2026 market-average development-cost benchmark Do not apply the benchmark to a row, convert it to GPU capex, or treat it as an electricity tariff. Cushman & Wakefield guide

Accelerator, inventory, specification, company, investor, and price register

Quantity state remains separate from catalog orderability or product specification. No row below establishes a global installed-GPU total. Hosted rates are provider rental signals, not hardware purchase prices.

Record Quantity and state Model key specification Vendor, provider, and company role Explicit investor financing role Possible price — original unit, type, date Direct evidence --- --- --- --- --- --- --- GPU-01 · NVIDIA GB300 NVL72 Site-deployed quantity ND ; NVIDIA lists the platform Available Now catalog orderability ; Bitdeer A102 is planned rack-scale context; installed, delivered, ordered, committed, and allocated count ND Fully liquid-cooled rack with 72 NVIDIA Blackwell Ultra GPUs and 36 Arm-based Grace CPUs ; 800 Gb s network connectivity per GPU; rack specification is not site inventory NVIDIA OEM platform provider; Bitdeer intended deployment context; purchaser and operator allocation ND Investor, lender, and financing role ND OEM purchase and rental price ND ; no price inferred from site revenue or cloud rates NVIDIA GB300 NVL72 — vendor page captured 4 Sep 2026; related Bitdeer A102 release GPU-02 · Lambda on-demand catalog Catalog orderability: 1x, 2x, 4x, and 8x GPU instances; 1-Click Clusters offer 16–2,000+ interconnected B200 or H100 GPUs ; no deployed fleet count B200 SXM6 180 GB VRAM GPU ; H100 SXM 80 GB VRAM GPU ; A100 and V100 also listed Lambda cloud provider operator; NVIDIA OEM; inventory-finance and project-allocation roles ND Investor, lender, and inventory-finance role ND B200 $6.69 GPU-hour; H100 $3.99 GPU-hour , USD, pay-by-minute on-demand catalog rates, captured 4 Sep 2026; taxes may apply; cloud rental, not hardware price Lambda GPU instances — current catalog captured 4 Sep 2026 GPU-03 · Nebius NVIDIA HGX B300 Provider-deployed quantity ND ; operating provider-deployed claim because Nebius says HGX B300 is deployed in production on Nebius AI Cloud; no count disclosed 8× Blackwell Ultra GPUs; 270 GB HBM3e GPU , 2.1 TB total HBM3e , 62 TB s total memory bandwidth, 14.4 TB s total NVLink bandwidth, 1.6 Tb s networking; projected performance labels retained Nebius cloud provider operator; NVIDIA platform OEM Investor, lender, and equipment-finance role ND $4.30 GPU-hour preemptible; $7.85 GPU-hour on-demand , USD, provider rental rates captured 4 Sep 2026; not hardware price Nebius HGX B300 and Nebius GPU pricing — captured 4 Sep 2026 GPU-04 · Nebius current GPU catalog — B200, H200, H100 Provider fleet count ND ; catalog orderability ; VM offers are not installed inventory B200 180 GB GPU ; H200 141 GB HBM3e GPU, 4.8 TB s ; H100 80 GB GPU ; specifications do not imply inventory Nebius cloud provider operator; NVIDIA OEM platform provider Investor, lender, and financing roles ND Nebius page captured 4 Sep 2026: B200 $3.95 preemptible $7.15 on-demand; H200 $2.45 $4.50; H100 $2.15 $3.85 per GPU-hour , USD, hosted rental Nebius GPU pricing — current rates captured 4 Sep 2026; NVIDIA H200 — vendor specification captured 4 Sep 2026; NVIDIA DGX B200 and Lambda instances — platform specification references GPU-05 · Nebius NVIDIA GB300 NVL72 Provider-deployed racks GPUs ND ; operating provider-deployed claim because Nebius says the platform is deployed in production on Nebius AI Cloud; no fleet count or customer allocation Liquid-cooled rack with 72 Blackwell Ultra GPUs and 36 Grace CPUs , 37 TB total fast memory, 1,440 PFLOPS sparse FP4 performance, 576 TB s total memory bandwidth as listed by Nebius Nebius cloud provider operator; NVIDIA OEM platform provider Investor, lender, and financing role ND Nebius pricing page says Contact us ; numerical rental and purchase price ND , captured 4 Sep 2026 Nebius GB300 NVL72 and Nebius GPU pricing — provider pages captured 4 Sep 2026; related NVIDIA GB300 NVL72

Current-run analytical charts

The charts use only same-unit or record-count comparisons. Nested phases and different economic objects are shown with explicit non-additivity notes.

Current-run evidence controls

Risk Treatment in this update Direct evidence --- --- --- Boundary and non-additivity IT load, facility campus power, utility interconnection, generation PPA, MVA, manufacturing output, and project capital remain separate. No global total is asserted. Hut 8 · Chevron · Macquarie Status and timing Announced, discussion-stage, permitted, under-construction, operating, and energized states are not substituted. Hut 8 current energization, Chevron FID, Heka site scale, Macquarie 45 47 MW discrepancy, and RT-One licensing remain open. MCIT · Uberlândia GPU inventory and specification NVIDIA Nebius rack specifications, provider claims, and catalog orderability are not treated as deployed counts. No global installed-GPU total is inferred. NVIDIA GB300 · Nebius B300 Price type Provider rental rates are preserved in original USD GPU-hour units and dates. C&W is a market-average development-cost benchmark. No OEM hardware price was verified. Lambda · Nebius pricing · C&W guide

Current-run update — 7 September 2026

This additive update retains every prior section and the existing report presentation. It adds ten source-qualified project site rows across four regions, global electricity context, a power-financing framework, nine accelerator catalog records, and two charts. Figures remain source-defined and non-additive. ND means not disclosed or not verified in the reviewed evidence, not zero. IT load, facility campus power, utility interconnection, generation PPA, annual electricity, MVA, GPU inventory, cloud rental, investment, lease value, backlog, and transaction value remain separate economic or physical objects.

Current conclusion

The strongest current IT-load signals are Digital Realty's 288 MW Northern Virginia portfolio Digital Realty , Cerebras Compute Nordic's phased 165 MW contracted IT build Cerebras , AirTrunk's more than 280 MW JHB3 JHB4 plan AirTrunk , Hut 8's 352 MW second phase and 704 MW contracted campus boundary Hut 8 , and Macquarie's 47 MW facility figure Macquarie . They are not a sum. Power-only signals also remain separate: Chevron's approximately 2.67 GW generation plan under a 20-year PPA Chevron , Hut 8's 1,000 MW utility capacity under an AEP Texas interconnection agreement Hut 8 , and IEA's 415 TWh global data-center electricity consumption in 2024 IEA .

No reviewed project source discloses a verified installed, delivered, ordered, committed, or allocated GPU count. NVIDIA, AMD, AWS, Lambda, and Nebius pages below establish specifications, availability or provider claims, catalog orderability, or hosted rental offers only. No current OEM hardware purchase price was verified.

Cross-regional current-run project and construction table

The table is intentionally flat. Each row carries source publication date, capture as-of date, direct evidence, and an uncertainty note. Capacity boundaries are not interchangeable or additive.

ID Project and location Region Source publication; capture as-of Status Possible or announced capacity IT load Facility campus power Grid interconnection; generation PPA electricity signal Developer operator and company roles Verified investor or financing role Possible price or capital signal GPU inventory Direct source URL Uncertainty --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- DC-01 Hut 8 Beacon Point AI Data Center Campus, Nueces County, Texas, US Americas 2026-07-20; captured 2026-09-07 Under construction Phase 2 planned; initial Phase 2 data-hall delivery expected Q2 2028; no current energization claim in the read Second phase 352 MW IT ; campus contracted IT 704 MW after two leases; campus utility envelope 1,000 MW . Separate boundaries 352 MW second phase; 704 MW contracted campus IT 1,000 MW utility capacity; second AI factory supported by about 500 MW utility capacity, company-reported 1,000 MW secured under an AEP Texas interconnection agreement; generation, PPA price, MVA, and annual energy ND Hut 8 developer operator and energy-infrastructure platform; tenant confidential; NVIDIA DSX is a design reference; AEP Texas utility role Phase 1 investment-grade debt reported; lender, equity investor, and project-finance terms ND USD 9.8B 15-year Phase 2 lease; campus base-term value USD 19.6B and potential renewal value USD 50.2B ; lease value, not capex or hardware price GPU count model state ND Hut 8 release Utility, IT, lease, and delivery claims are company-reported; energization and lender identity remain open DC-02 Chevron Project Kilby co-located power facility for a Microsoft-operated data center, West Texas, US Americas 2026-06-22; captured 2026-09-07 Planned; FID expected by end-2026 subject to conditions; first power anticipated 2028 Approximately 2.67 GW generation capacity in phased modular delivery; data-center IT and facility load ND ND Generation facility approximately 2.67 GW; data-center facility campus power and MVA ND Energy Forge One LLC, a wholly owned Chevron subsidiary, signed a 20-year PPA with Microsoft; interconnection import limit ND; GE Vernova and Solar Turbines supply equipment Chevron Energy Forge One power developer and supplier; Engine No. 1 collaborator; Microsoft operator offtaker; GE Vernova and Solar Turbines suppliers No named lender, equity investor, or project-finance structure Chevron targets mid-teen returns; capex, PPA price, tariff, and lease price ND; target is not realized performance GPU count model state ND Chevron release 2.67 GW is generation, not IT load; FID, permits, grid arrangements, and PPA economics remain open DC-03 Digital Realty Blackstone three hyperscale data centers in Manassas and Sterling, Northern Virginia, US Americas 2026-06-29; captured 2026-09-07 Under construction development-stage; fully leased; two expected to stabilize H1 2027 and one H1 2028 288 MW total IT capacity , three facilities at 96 MW each 288 MW total IT; each facility 96 MW Facility campus power, generation, PPA, MVA, and annual energy ND Grid interconnection and utility capacity ND; source describes development and long-term leases, not power delivery Digital Realty operator developer; Blackstone-affiliated funds were JV equity seller; three investment-grade hyperscale customers are lessees Blackstone-affiliated funds verified as equity JV seller; Digital Realty agreed to buy a blended 64% equity interest; no lender named Gross asset value USD 7.8B including assumed debt and remaining capex; USD 3.5B consideration including USD 1.2B cash and USD 2.3B shares; acquisition asset value, not site capex or GPU price GPU count model state ND Digital Realty investor release Capacity is IT only; value includes debt and remaining capex; stabilization and closing are forward company conditions DC-04 RT-One AI data center, western Uberlândia, Minas Gerais, Brazil Americas RT-One 2025-10-11; municipal status 2026-02-23; captured 2026-09-07 Planned municipal and licensing steps; architectural review described as final municipal step before works; not energized or operating RT-One claims 400 MW installation ; municipality reports 100 MW first phase with expansion up to 400 MW; boundary undefined ND ND; no MVA or annual-energy basis RT-One says 100% renewable energy; electricity discussions with Cemig; no interconnection agreement, grid import limit, generation nameplate, PPA, or tariff verified RT-One developer; Uberlândia municipality planning zoning authority; Cemig electricity counterparty; operator, customer, and vendors ND RT-One claims investment above BRL 6B ; no named external investor, lender, or financing close BRL 6B+ company-reported project investment; not a funding close, tariff, lease, or GPU price GPU count model state ND RT-One statement · Uberlândia update Oldest capital statement is from 2025; capacity boundary, renewable claim, permits, financing, and interconnection require recheck DC-05 Cerebras Compute Nordic Finland AI data centre, Mikkeli, Finland Europe Body dateline 2026-09-01; page title index 2026-08-31; captured 2026-09-07 Under construction; initial 50 MW phase under way; phased delivery 50 MW, 80 MW, then 165 MW 165 MW contracted IT capacity delivered in phases; seven-year service orders 165 MW contracted IT; facility power, MVA, and annual energy ND Closed-loop cooling and heat reuse; facility power, MVA, and annual energy ND Grid interconnection, generation, PPA, and tariff ND Cerebras long-term customer platform; Compute Nordic Finland developer; partner network not fully named Project equity, lender, and investor roles ND Ramboll estimates EUR 1.0–1.7B indicative regional investment at full scale; estimate, not committed capex, lease, or GPU price GPU count model state ND; source describes wafer-scale AI compute Cerebras investor release Date title discrepancy; investment is an estimate cited by the company, not a financing close; power delivery unverified DC-06 Nebius Lappeenranta AI factory, South Karelia, Finland Europe 2026-03-31; captured 2026-09-07 Planned; construction announced; first capacity expected available to customers in 2027 Up to 310 MW facility capacity ; IT facility boundary undefined ND Multi-building campus on about 100 acres; predominantly low-carbon electricity mix; MVA and annual energy ND Grid interconnection, utility, generation, PPA, and tariff ND Nebius developer operator and AI cloud; City of Lappeenranta host partner; NVIDIA Blackwell and Vera Rubin platform roles Investor, lender, and project-finance roles ND Project capex, price, lease, tariff, and financing amount ND NVIDIA platform intent; GPU count model allocation and delivery state ND Nebius Lappeenranta release 310 MW is source-defined facility capacity without an IT power boundary; progress and financing not disclosed DC-07 Heka Data and unnamed US consortium integrated data-center and AI hub, Egypt Middle East Africa 2026-09-03; captured 2026-09-07 Announced exploratory; further studies and procedures remain Initial scope, scale, phases, and capacity ND ND Facility campus power, MVA, and annual energy ND MCIT discussion included integrating digital infrastructure with power supply; grid, interconnection, generation, PPA, and tariff ND Heka Data named developer; US consortium unnamed; MCIT, NTRA, ITIDA, and Telecom Egypt officials regulatory participants; operator customer ND No named investor, lender, equity commitment, or project-finance role Investment, capex, lease, service revenue, tariff, and hardware price ND GPU count model state ND Egypt MCIT release Discussion confirms neither site selection, permit, construction, energization, nor financing DC-08 AirTrunk JHB3 and JHB4 hyperscale campuses, Iskandar Puteri Johor Bahru, Malaysia APAC 2026-04-30; captured 2026-09-07 Planned; new campuses announced; existing JHB1 JHB2 almost 100% contracted per company New JHB3 JHB4 combined more than 280 MW IT load ; four-campus Malaysia platform more than 700 MW IT load; not additive More than 280 MW IT load for JHB3 JHB4 Facility campus power, MVA, grid, generation, PPA, and annual energy ND Electricity source and PPA ND; 100% recycled water is not an electricity signal AirTrunk developer operator; global cloud and AI customers intended committed users; local suppliers named without equipment inventory Investor, lender, and project-finance role ND New-campus investment MYR 12B USD 3B ; total Malaysia committed investment about MYR 27B USD 6.8B ; company-reported investment, not funding close or GPU price GPU count model state ND AirTrunk Malaysia release Company-reported capacity and investment; construction, grid, customer contract, and financing instrument not read DC-09 Macquarie Data Centres IC3 Super West, Macquarie Park campus, Sydney, Australia APAC Page undated current; captured 2026-09-07 Under construction; end-state power secured; ready for service in 2026 per operator page Facility heading 45 MW ; body 47 MW IT load ; campus page 65 MW potential; values retained, not reconciled or summed Body: 47 MW IT load; heading: 45 MW; campus: 65 MW Entire end-state power secured; MW, MVA, utility, grid agreement, and annual energy ND Generation, PPA, storage, tariff, and electricity source ND Macquarie Data Centres developer operator; sovereign AI cloud colocation facility; customer offtaker ND Investor, lender, and project-finance role ND Capex, lease, service price, tariff, and GPU price ND GPU count model delivery allocation ND; cooling supports air, liquid, and hybrid environments Macquarie IC3 Super West Operator page has a 45 47 MW inconsistency and no grid or investor detail; construction claim not independently audited DC-10 NTT DATA Phoenix Mesa multi-data-center campus, Arizona, US Americas 2025-05-13; captured 2026-09-07 Planned; 174 acres acquired; first data center scheduled to open FY2028 Seven planned facilities with anticipated 324 MW capacity; existing Mesa 240 MW and market total just under 600 MW are not assigned to the new campus IT-load boundary ND; capacity basis undefined Facility campus power, MVA, grid, generation, PPA, and annual energy ND Grid interconnection and electricity arrangements ND NTT DATA developer operator; parent company backing stated; Mesa Arizona land and development role Parent NTT DATA backing verified; site-specific external investor, lender, and project-finance terms ND NTT DATA previously announced USD 10B expansion investment through 2027; portfolio program, not Phoenix site capex GPU count model state ND NTT DATA expansion release 2025 portfolio announcement; current Phoenix construction, power, funding allocation, and IT boundary not revalidated

Regional current-run slices

These regional tables are navigation views of the cross-regional ledger; they do not create new totals or alter any row's boundary.

Americas

Record Status source-as-of Capacity and electricity boundary Company backing and capital signal Direct source Uncertainty --- --- --- --- --- --- DC-01 · Hut 8 Beacon Point, Texas Under construction Phase 2 planned; 20 Jul 2026 source, captured 7 Sep 352 MW second-phase IT; 704 MW contracted campus IT; 1,000 MW utility capacity; AEP Texas interconnection; generation PPA ND Hut 8 developer operator; Phase 1 investment-grade debt reported; USD 9.8B Phase 2 lease Hut 8 Energization, lender identity, and delivery remain open DC-02 · Chevron Project Kilby, West Texas Planned; 22 Jun 2026 source, captured 7 Sep 2.67 GW generation; 20-year Microsoft PPA; data-center IT, facility power, and interconnection ND Chevron Energy Forge One developer supplier; Microsoft operator offtaker; capex and PPA price ND Chevron Generation is not IT load; FID and permits remain open DC-03 · Digital Realty Blackstone, Northern Virginia Under construction development; 29 Jun 2026 source, captured 7 Sep 288 MW total IT; facility grid generation PPA ND Digital Realty operator developer; Blackstone-affiliated equity JV seller; USD 7.8B gross asset value and USD 3.5B consideration Digital Realty Closing and stabilization remain forward conditions DC-04 · RT-One, Uberlândia Planned licensing; 11 Oct 2025 and 23 Feb 2026 sources, captured 7 Sep 100 MW first phase and up to 400 MW expansion; boundary, grid, generation, and PPA ND RT-One developer; Cemig electricity counterparty; BRL 6B+ company-reported investment RT-One · Municipality Permits, financing, renewable claim, and capacity boundary require recheck DC-10 · NTT DATA Phoenix Mesa Planned; 13 May 2025 source, captured 7 Sep Seven facilities 324 MW source-defined capacity; IT, grid, and electricity boundary ND NTT DATA developer operator and parent backing; USD 10B portfolio expansion program, not site capex NTT DATA Current construction and site-level funding not revalidated

Europe

Record Status source-as-of Capacity and electricity boundary Company backing and capital signal Direct source Uncertainty --- --- --- --- --- --- DC-05 · Cerebras Compute Nordic, Mikkeli Under construction; 1 Sep 2026 body dateline 31 Aug page index, captured 7 Sep 165 MW contracted IT in 50 80 165 MW phases; grid, generation, PPA, and facility power ND Cerebras customer platform; Compute Nordic developer; EUR 1.0–1.7B indicative regional investment estimate Cerebras Date discrepancy and no power-delivery proof DC-06 · Nebius Lappeenranta Planned; 31 Mar 2026 source, captured 7 Sep Up to 310 MW facility capacity; IT, grid, generation, PPA, and tariff ND Nebius developer operator; City of Lappeenranta host partner; financing and project price ND Nebius Facility boundary and construction progress not disclosed

Middle East Africa

Record Status source-as-of Capacity and electricity boundary Company backing and capital signal Direct source Uncertainty --- --- --- --- --- --- DC-07 · Heka Data US consortium, Egypt Exploratory discussion; 3 Sep 2026 source, captured 7 Sep Scope, scale, phases, IT, facility, grid, generation, PPA, and tariff ND Heka Data named developer; US consortium unnamed; no investor or financing role verified Egypt MCIT Discussion is not a site, permit, construction, energization, or financing claim

APAC

Record Status source-as-of Capacity and electricity boundary Company backing and capital signal Direct source Uncertainty --- --- --- --- --- --- DC-08 · AirTrunk JHB3 JHB4, Malaysia Planned; 30 Apr 2026 source, captured 7 Sep More than 280 MW IT load; Malaysia platform more than 700 MW; grid, generation, PPA, and annual energy ND AirTrunk developer operator; MYR 12B USD 3B new-campus investment; financing role ND AirTrunk Lower-bound capacity and company-reported investment; build grid status not read DC-09 · Macquarie IC3 Super West, Sydney Under construction; current undated page, captured 7 Sep 47 MW body IT figure, 45 MW heading, 65 MW campus figure; end-state power secured; grid generation PPA ND Macquarie Data Centres developer operator; capex and financing ND Macquarie 45 47 MW inconsistency and no investor grid detail

Electricity, power-only, and global context

ID scope Electricity or power object Value and unit Relation to IT load project boundary Source publication; capture as-of Uncertainty Direct source --- --- --- --- --- --- DC-02 · Chevron Project Kilby Planned generation under a Microsoft PPA Approximately 2.67 GW generation ; 20-year PPA; first power anticipated 2028 Generation offtake is not data-center IT or facility load; grid import and PPA price ND 22 Jun 2026; captured 7 Sep 2026 FID, permits, interconnection, and economics remain open Chevron DC-01 · Hut 8 Beacon Point Utility interconnection capacity 1,000 MW utility capacity secured under AEP Texas interconnection agreement Utility capacity is separate from 352 MW second-phase IT and 704 MW contracted campus IT; generation and annual energy ND 20 Jul 2026; captured 7 Sep 2026 Company-reported agreement; current delivery energization not confirmed Hut 8 OUT-01 · IEA global context Annual electricity consumption About 415 TWh , or 1.5% of global electricity in 2024 Global annual energy is not a site load, generation nameplate, or capacity total Page date ND in retrieved content; captured 7 Sep 2026 Global context uses a different scope and method from project rows IEA Energy demand from AI OUT-01 · JLL global outlook Data-center capacity and investment outlook Capacity expected 103 GW to 200 GW by 2030 ; up to USD 3T investment over five years Outlook estimate is not a site ledger or additive total of rows above 6 Jan 2026; captured 7 Sep 2026 Estimate scenario; methods differ from IEA and project sources JLL outlook

Financing and power-program context

ID Program scope Status and boundary Company roles and verified financing role Capital possible price signal Direct source Uncertainty --- --- --- --- --- --- --- FIN-01 Brookfield–Bloom Energy strategic AI-infrastructure power partnership; site allocation ND Announced global framework; project MW, IT load, grid connection, generation nameplate, and PPA terms ND Bloom Energy onsite power-platform provider; Brookfield AI-infrastructure investor and capital operating partner USD 25B financing framework ; source also references a USD 100B target fund; framework, not committed site capex, tariff, lease, or GPU price Bloom Energy investor release Company-reported framework; no project allocation, draw, lender terms, MW, PPA price, or installed asset verified

Possible price and capital signals

Record Value currency Unit economic type Source as-of Boundary and non-additivity control Direct evidence --- --- --- --- --- --- Hut 8 Beacon Point Phase 2 USD 9.8B 15-year lease value with 3.0% annual base-rent escalator 20 Jul 2026; captured 7 Sep 2026 Lease value, not capex, tariff, financing close, or GPU price; campus base-term and renewal values are separate Hut 8 Digital Realty Blackstone Northern Virginia USD 7.8B gross asset value; USD 3.5B consideration Acquisition asset value including assumed debt and remaining capex; consideration includes USD 1.2B cash and USD 2.3B shares 29 Jun 2026; captured 7 Sep 2026 Not site capex, rent, tariff, or GPU price; closing stabilization remain forward conditions Digital Realty RT-One Uberlândia BRL 6B+ Company-reported project investment 11 Oct 2025; captured 7 Sep 2026 Not a completed funding close, tariff, lease, or GPU price; 100 400 MW boundary undefined RT-One Cerebras Compute Nordic Mikkeli EUR 1.0–1.7B Indicative regional investment estimate at full scale 1 Sep 2026 release dateline; captured 7 Sep 2026 Estimate, not committed project capex, rent, or GPU price Cerebras AirTrunk Malaysia MYR 12B USD 3B new campuses; MYR 27B USD 6.8B Malaysia platform Company-reported investment 30 Apr 2026; captured 7 Sep 2026 New-campus and platform figures are not additive; not a financing close, tariff, lease, or GPU price AirTrunk NTT DATA expansion program USD 10B Portfolio expansion investment through 2027 13 May 2025; captured 7 Sep 2026 Program amount, not Phoenix site capex or GPU price NTT DATA JLL global outlook Up to USD 3T Five-year total investment outlook; includes real-estate value creation and debt estimates 6 Jan 2026; captured 7 Sep 2026 Scenario outlook, not project capex or a sum of listed records JLL Brookfield–Bloom Energy USD 25B Financing framework for power projects 30 Jun 2026; captured 7 Sep 2026 Framework, not drawn capital, site capex, tariff, lease, or GPU price Bloom Energy

Accelerator inventory, specifications, company roles, and possible price

Product specifications are not site inventories. Provider catalog sizes are not installed fleet counts. All prices below are original-currency provider listings captured 7 September 2026; OEM purchase prices are ND unless explicitly stated otherwise.

ID Vendor model Source date; capture as-of Directly sourced specifications Availability orderability Inventory state Company operator and OEM roles Verified investor financing role Possible price: currency, unit, type, date Direct source URL Uncertainty --- --- --- --- --- --- --- --- --- --- --- GPU-01 NVIDIA GB300 NVL72 Product page undated; captured 2026-09-07 Fully liquid-cooled rack-scale platform; 72 NVIDIA Blackwell Ultra GPUs and 36 Grace CPUs ; 800 Gb s network connectivity per GPU NVIDIA page says Available Now Catalog; delivered, ordered, committed, allocated, and installed quantity ND NVIDIA OEM platform provider; no site operator or purchaser named ND OEM purchase price and cloud rental price ND NVIDIA GB300 NVL72 Availability is not a delivered or installed count GPU-02 NVIDIA DGX B200 Product page undated; captured 2026-09-07 8 Blackwell GPUs ; 1,440 GB total GPU memory; 64 TB s HBM3e bandwidth; approximately 14.3 kW maximum system power; 10 RU Product catalog page; explicit stock orderability ND Catalog specification; inventory quantity state ND NVIDIA OEM and integrated AI platform provider; operator purchaser ND ND OEM purchase and rental price ND NVIDIA DGX B200 Product specification is not a deployment or inventory claim GPU-03 NVIDIA H100 SXM Product page undated; captured 2026-09-07 Hopper architecture; fourth-generation Tensor Cores; Transformer Engine with FP8; 900 GB s GPU-to-GPU interconnect; Lambda catalog lists 80 GB H100 SXM NVIDIA product catalog; Lambda self-serve cloud is first-come; OEM stock ND Catalog cloud-orderable; provider fleet count ND NVIDIA OEM; Lambda cloud operator provider ND USD 3.99 GPU-hour , on-demand cloud rental, captured 2026-09-07; not hardware price NVIDIA H100 · Lambda instances Cloud rental does not establish ownership or hardware inventory GPU-04 NVIDIA H200 Product page undated; captured 2026-09-07 141 GB HBM3e at 4.8 TB s ; Hopper architecture; NVIDIA says Now available OEM page says Now available ; Nebius lists HGX H200 cloud instances Catalog orderable provider rental; installed count ND NVIDIA OEM; Nebius cloud provider for HGX H200 ND Nebius USD 2.45 preemptible and USD 4.50 on-demand GPU-hour , captured 2026-09-07 NVIDIA H200 · Nebius pricing Availability and rental offer are not a delivered or allocated count GPU-05 AMD Instinct MI350X Product page undated; captured 2026-09-07 Fourth-generation AMD CDNA; 288 GB HBM3E ; 8 TB s bandwidth ; MXFP6 and MXFP4 support AMD product catalog page; orderability and delivery timing ND; page references AMD Instinct Cloud Options Catalog specification; inventory quantity state ND AMD OEM platform provider; operator purchaser ND ND OEM purchase and rental price ND AMD Instinct MI350X Specification is not stock, delivery, commitment, or installed-count evidence GPU-06 NVIDIA HGX B300 on Nebius AI Cloud Product page undated; captured 2026-09-07 8 Blackwell Ultra GPUs ; 2.1 TB HBM3e total memory; 62 TB s total memory bandwidth; 14.4 TB s total NVLink bandwidth; 1.6 Tb s networking Nebius says deployed in production on Nebius AI Cloud Operating provider-deployed claim; fleet count, delivered quantity, and allocation ND Nebius cloud operator; NVIDIA OEM platform provider ND Nebius USD 4.30 preemptible and USD 7.85 on-demand GPU-hour , captured 2026-09-07 Nebius HGX B300 · Nebius pricing Provider-reported production claim is count-free; rental is not hardware price GPU-07 AWS EC2 Capacity Blocks P6-B300 Pricing page current; captured 2026-09-07 8x B300 accelerators per p6-b300.48xlarge ; listed regions include US, GovCloud, and APAC AWS publishes current Capacity Block reservation rates; next update stated as October 2026 Catalog capacity-block reservation offer; AWS fleet count and guaranteed inventory total ND AWS cloud operator provider; NVIDIA B300 OEM role implied by model ND USD 112.32 instance-hour or USD 14.04 accelerator-hour , reservation effective rate plus separate OS fee AWS EC2 Capacity Blocks pricing Reservation pricing is capacity access, not hardware purchase price or installed stock GPU-08 Lambda NVIDIA B200 Catalog page current; captured 2026-09-07 180 GB VRAM GPU ; 1, 2, 4, and 8 GPU instances; clusters of 16–2,000+ interconnected B200 H100 GPUs Self-serve, first-come cloud access Catalog cloud-orderable; cluster size is an offer range, not fleet inventory Lambda cloud operator; NVIDIA OEM ND USD 6.69 GPU-hour , on-demand cloud rental, captured 2026-09-07; tax may apply Lambda instances No installed fleet count GPU-09 Lambda NVIDIA H100 Catalog page current; captured 2026-09-07 80 GB VRAM GPU ; 1, 2, 4, and 8 GPU instances; clusters of 16–2,000+ interconnected H100 B200 GPUs Self-serve, first-come cloud access Catalog cloud-orderable; cluster size is an offer range, not fleet inventory Lambda cloud operator; NVIDIA OEM ND USD 3.99 GPU-hour , on-demand cloud rental, captured 2026-09-07; tax may apply Lambda instances No installed, delivered, or allocated count

The current packet does not verify an OEM purchase price or executed hardware transaction price for GB300, B200, H100, H200, or MI350X. Cloud rates are not hardware prices, installed-inventory counts, electricity tariffs, or project capex.

Current-run analytical charts

The charts use same-unit or same-object comparisons only. Values remain non-additive across projects, phases, providers, and billing terms.

Current-run evidence controls and unresolved risks

Risk Treatment in this update Direct evidence and implication --- --- --- Boundary and non-additivity IT load, facility campus power, utility interconnection, generation PPA, annual electricity, MVA, GPU specifications, and project capital remain separate. No global total is asserted. Hut 8 · Chevron · IEA Status and timing Announced, exploratory, planned, under-construction, operating provider-deployed, and energized states are not substituted. Hut 8 energization, Chevron FID, Heka site scale, Nebius construction progress, AirTrunk delivery, and NTT Phoenix status remain open. Egypt MCIT · Nebius · NTT DATA Conflicting or lower-bound figures Macquarie's 45 MW heading and 47 MW body figure remain unreconciled; AirTrunk's more-than-280 MW remains a lower bound; Cerebras retains a date title discrepancy; RT-One retains an undefined 100 400 MW boundary. Macquarie · AirTrunk · Cerebras GPU inventory and specification Vendor specifications, provider production claims, and cloud catalog orderability are not treated as installed counts. No global installed-GPU total is inferred. NVIDIA GB300 · AMD MI350X · Nebius B300 Price type Hosted rates retain original USD GPU-hour or accelerator-hour units and dates; capital rows retain original currency and economic type. No OEM purchase price was verified. Lambda · Nebius pricing · AWS Capacity Blocks Dashboard display contract A triggered run must refresh a flat dashboard table with source, as-of date, and uncertainty. This report-only session did not alter the dashboard. The current report preserves table-first display and the dashboard constraint; no dashboard write was performed in this session

Current-run update — 8 September 2026

This additive evidence refresh preserves every preceding section, table, chart, and structured-report element. It rechecks a bounded sample of 11 data-center projects or expansions and 9 accelerator, inventory, or provider-capacity records. Source-defined capacity, electricity, grid access, generation PPA, GPU state, company backing, financing, revenue, and hosted rental prices remain separate; ND means not publicly disclosed or not verified in the reviewed source, not zero. The sample is not a global census and no global capacity, electricity, installed-GPU, investment, or comparable-price total is asserted.

Current evidence screen — data-center plans, construction, power, company, backing, and price

ID record Status and timing IT compute capacity Electricity, grid, or power boundary Company, operator, and explicitly documented backing Possible price or capital signal Direct evidence --- --- --- --- --- --- --- DC-01 · Start Campus SINES — Sines, Portugal SIN01 operating since Q4 2024 and inaugurated 4 Apr 2025; later buildings phased toward 2030; current SIN02–06 state ND 1,200 MW secured IT capacity across six buildings; SIN01 26 MW IT ; nested REN allocation grid access reported secured; exact interconnection MW MVA ND; 100% renewable target and seawater cooling Start Campus developer operator; Davidson Kempner Capital Management and Pioneer Point Partners named shareholders backers; customer split ND €8.5B privately funded construction investment signal; not per-building capex, tariff, or GPU price Start Campus capacity · SIN01 DC-02 · Applied Digital Polaris Forge 1 — Ellendale, US Building 1 ready-for-service and fully energized 24 Nov 2025; later campus delivery ND 100 MW critical IT load in Building 1; 400 MW fully contracted deployment across campus; separate boundaries Building 1 energized at 100 MW; utility interconnection MW MVA ND Applied Digital developer operator; CoreWeave long-term lessee customer; site equity lender allocation ND About US$16B contracted revenue across Polaris Forge 1 and 2; revenue, not capex or GPU price Applied Digital DC-03 · ODATA QR03 — Querétaro, Mexico 200 MW energy phase energized Feb 2025; first building scheduled Q2 2025; current full-build status ND 300 MW IT capacity at full build-out; first building 72 MW IT ; nested 200 MW energized electrical infrastructure , expansion toward 400 MW ; 400 kV substation and transmission lines; electrical infrastructure is not IT load ODATA, an Aligned Data Centers company, developer operator; Ammper renewable-energy partner; investors lenders ND US$3.3B reported total investment; not GPU price ODATA QR03 DC-04 · Crusoe Lancium Abilene — Texas, US First phase reported live 30 Sep 2025; six additional buildings under construction in the cited expansion notice with mid-2026 target; post-target status ND 200+ MW initial phase and 1.2 GW total facility power ; neither is labelled IT load 1.2 GW grid interconnection ; battery, solar, and proximate wind described; generation MW ND Crusoe developer operator; Lancium clean-campus energy role; Oracle long-term collaboration customer context; project investors ND About US$1B direct indirect initial-phase economic impact over 20 years; project capex ND Campus expansion · Live phase DC-05 · Meta Hyperion — Richland Parish, Louisiana, US Under construction and expanding in Meta's 13 Jul 2026 update; completion and operating date ND 5 GW compute capacity ; source does not define this as IT load Entergy Louisiana energy agreement; site grid MW and generation boundary ND Meta owns operates; Entergy utility; DPR, Turner, and Mortenson contractors; external equity allocation ND More than US$50B regional investment signal; not treated as site-only capex or GPU price Meta Hyperion update DC-06 · Nebius AI Factory — Lappeenranta, Finland Construction announced 31 Mar 2026; first customer capacity expected in 2027; energization ND Up to 310 MW AI-factory capacity ; IT-versus-facility boundary ND Predominantly low-carbon electricity mix described; grid interconnection MW, MVA, generation, PPA, and tariff ND Nebius developer operator; customer allocation and project investors lenders ND Site capex, financing, tariff, lease, and hardware price ND Nebius Lappeenranta DC-07 · Nscale Aker OpenAI Stargate Norway — Narvik Kvandal Planned; mobilization announced 31 Jul 2025; Microsoft deployment described for 2027; energization ND 230 MW planned capacity plus 290 MW expansion ambition ; initial phase 20 MW ; basis not labelled IT Entirely renewable-powered claim with hydropower context; limited transmission capacity noted; separate interconnection rating ND Nscale designs builds manages; 2025 announcement planned a 50 50 Nscale–Aker JV subject to closing; OpenAI initial offtaker; 2026 statement says Narvik deployment solely managed by Nscale About US$1B initial 20 MW commitment, including up to US$250M equity ; separate US$790M financing and 115 MW accordion are not added Stargate Norway · Microsoft deployment · Norway financing DC-08 · Khazna multi-country expansion — MENA international Roadmap announced through 2030; site delivery and energization vary and are not consolidated More than 1 GW additional operational hyperscale capacity , including more than 400 MW international ; source-defined hyperscale capacity, not IT total Italy project described as using Eni “Blue Power”; site grid interconnection and generation MW ND Khazna G42 platform; UAE national-infrastructure role; ADCB and FAB lenders for platform facility; site equity allocation ND US$2.62B financing facility , announced 3 Sep 2025; platform growth financing, not a site or GPU price Khazna capacity · Khazna financing DC-09 · Cerebras Compute Nordic — Mikkeli, Finland Initial 50 MW phase under construction in 1 Sep 2026 announcement; not reported energized 165 MW contracted IT capacity , delivered as nested 50 80 165 MW phases; seven-year service orders Closed-loop cooling and heat reuse; grid interconnection MW, MVA, generation, PPA, and annual energy ND Cerebras long-term customer platform; Compute Nordic Finland developer; project equity lender investor roles ND €1.0–1.7B estimated regional investment at full scale; indicative estimate, not committed capex or GPU price Cerebras Compute Nordic DC-10 · Bitdeer AI A102 — Malaysia About 9.5 MW sold out under five-year offtake commitments ahead of Q1 2027 energization; not reported operating About 9.5 MW data-center capacity ; IT boundary ND; separate target up to 350 MW AI-cloud capacity by Q1 2028 is not additive Power availability cited as a siting rationale; grid interconnection, generation, PPA, and tariff ND Bitdeer AI, part of Bitdeer Technologies Group, provider developer; customers unnamed; customer prepayments, operating cash flow, and financing against contracted cash flows cited More than US$800M expected five-year AI-cloud revenue ; prepayments expected to cover over 50% of associated capex; revenue prepayment, not GPU price Bitdeer A102 DC-11 · Kasi Cloud LOS1 — Lagos, Nigeria Flag-off commissioning ceremony 19 May 2026 moved LOS1 into operational readiness; fully operating load ND About 100 MW critical IT capacity at full development ; LOS1 phase load ND Grid interconnection MW, generation, PPA, and tariff ND Kasi Cloud Datacenters developer operator; government and NSIA participation reported around ceremony; equity lender allocation ND Project capex, financing, lease, tariff, and GPU price ND; US$850M annual offshore-cloud spend is market rationale, not project capital Kasi Cloud commissioning

Current evidence screen — GPU inventory, sourced specifications, company backing, and price

ID record Quantity and state Model or sourced specification Company, operator, and explicitly documented backing Possible price and type Direct evidence --- --- --- --- --- --- GPU-01 · CoreWeave global fleet More than 250,000 GPUs across 32 data centers as of 31 Dec 2024 in the S-1; historical operating lower bound. FY2025 10-K reports 43 data centers, more than 850 MW active power , and about 3.1 GW contracted power , but no current GPU count Mixed accelerator fleet; model split and per-GPU specifications ND in the cited fleet disclosures CoreWeave operator cloud provider; NVIDIA supplier and disclosed stockholder in later filings; no per-GPU investor allocation US$7.5B debt facility led by Blackstone and Magnetar, with other participants; company financing, not GPU purchase price; fleet acquisition price ND CoreWeave S-1 · FY2025 10-K · Debt facility GPU-02 · xAI Colossus — Memphis, US Current first-party page states 200,000 H100 GPUs in one interconnected cluster and a roadmap to 1M; operating expanding claim, not an audited receipt ledger NVIDIA H100; Lambda lists 80 GB H100 SXM ; NVIDIA describes fourth-generation NVLink up to 900 GB s GPU-to-GPU interconnect xAI operator customer; NVIDIA OEM; xAI's 2024 Series B named Valor Equity Partners, Vy Capital, Andreessen Horowitz, Sequoia, Fidelity, Prince Alwaleed, and Kingdom Holding; company backers, not GPU purchase financiers GPU purchase price ND; US$6B Series B is company financing, not hardware price xAI Colossus · xAI Series B · NVIDIA H100 GPU-03 · DataSection Thailand 4,696 GPUs across 587 servers acquired procured for planned deployment; delivery, commissioning, and installed state ND; 8 GPUs server is derived, not a sourced node specification NVIDIA B200 Blackwell GPUs in ASRock Rack servers; B200 VRAM interconnect not published in the cited acquisition release DataSection operator developer; ASRock Rack supplier; customer unnamed; investor, lender, and customer-finance allocation ND Acquisition amount, unit price, and transaction terms ND DataSection GPU-04 · Yotta D2 — Greater Noida, India 20,736 Blackwell Ultra GPUs announced 18 Feb 2026, expected live by Aug 2026; as of 8 Sep 2026, installed operating state was not independently reverified; more than 10,000 committed to IndiaAI is not additive Liquid-cooled NVIDIA Blackwell Ultra HGX B300 ; 800 Gbps Quantum-X800 InfiniBand and more than 40 PB parallel storage; GPU VRAM ND in release Yotta operator; NVIDIA DGX Cloud engagement; NVIDIA technology customer partner, not named project equity investor; project investors lenders ND More than US$2B infrastructure investment; separate four-year NVIDIA engagement more than US$1B ; neither is per-GPU price Yotta D2 GPU-05 · Nscale Aker OpenAI Norway 100,000 NVIDIA GPUs targeted by end-2026 ; separate agreement adds more than 30,000 Rubin GPUs for Microsoft in 2027; targets are not added or treated as installed NVIDIA GPU family target; Rubin SKU, VRAM, rack, and interconnect details ND Nscale Aker OpenAI project roles; Microsoft customer deployment role; backing and financing are project-level, not GPU purchase price GPU purchase price ND; project commitment and financing are infrastructure economics, not hardware price Nscale Norway · Nscale Microsoft · Nscale financing GPU-06 · Bitdeer AI A102 — Malaysia GPU count, ordered quantity, delivery, and installed state ND; site sold out for 9.5 MW ahead of Q1 2027 energization Planned NVIDIA GB300 NVL72 : liquid-cooled rack with 72 Blackwell Ultra GPUs + 36 Grace CPUs , 800 Gb s networking per GPU Bitdeer AI provider developer; NVIDIA preferred Cloud Partner OEM; customer and investor allocation ND More than US$800M expected five-year AI-cloud revenue ; service revenue, not GB300 purchase price Bitdeer A102 · NVIDIA GB300 GPU-07 · Nebius AI Cloud HGX B300 deployed in production ; installed fleet count, delivered quantity, and allocation ND; 8-GPU figure is per HGX system, not fleet inventory 8 Blackwell Ultra GPUs , 2.1 TB HBM3e , 62 TB s total memory bandwidth, 14.4 TB s total NVLink bandwidth, 1.6 Tb s networking Nebius cloud operator; NVIDIA OEM platform; investors backers and procurement financing ND US$4.30 preemptible US$7.85 on-demand per GPU-hour , captured 8 Sep 2026; cloud access rate, not hardware price Nebius B300 · Nebius prices GPU-08 · Lambda Cloud catalog Self-serve catalog: 1 2 4 8-GPU instances; 1-Click Clusters offer 16–2,000+ interconnected GPUs ; orderable range, not deployed fleet count NVIDIA B200 SXM6 180 GB VRAM GPU and H100 SXM 80 GB VRAM GPU Lambda cloud operator; NVIDIA OEM; investors backers and physical inventory ND US$6.69 B200 GPU-hour and US$3.99 H100 GPU-hour , captured 8 Sep 2026; provider rental list rates, not purchase prices Lambda instances GPU-09 · AWS EC2 Capacity Blocks Capacity-block reservation offer; no AWS fleet count or guaranteed global inventory total 8 × B300 accelerators per p6-b300.48xlarge ; B300 VRAM interconnect not disclosed on pricing page AWS cloud operator; NVIDIA B300 OEM role; investors backers and procurement allocation ND US$112.32 per instance-hour US$14.04 per accelerator-hour in standard listed regions; GovCloud separately US$117 US$14.625 ; reservation effective rate, not hardware price AWS Capacity Blocks pricing

Current-run analytical charts

The charts are selected-sample views derived only from source-defined records in the 8 September evidence package. ND regions are omitted rather than plotted as zero, and GPU bars carry their source state so they are not additive.

Current-run evidence controls and unresolved risks

Risk Treatment in this update Direct evidence and implication --- --- --- Global-total risk The 11-project and 9-accelerator screen is representative and bounded; no global MW, TWh, installed-GPU, investment, or comparable-price total is asserted Start Campus · xAI Colossus Boundary risk IT load, compute capacity, facility power, grid interconnection, generation PPA, hyperscale capacity, manufacturing or service capacity, and MVA remain source-defined; nested phases are not added ODATA · Nscale Status risk SINES later buildings, Abilene expansion, QR03 full build, Lappeenranta, Narvik, Yotta D2, Bitdeer A102, and Kasi LOS1 still lack a complete current independent operating energization confirmation in this packet Nebius Lappeenranta · Yotta D2 · Kasi Cloud Financing and investor risk Named shareholders, lenders, debt facilities, platform financing, company financing, and customer offtake arrangements are retained only where explicitly documented; they do not prove GPU ownership or project funding allocation CoreWeave debt · Khazna financing Price risk Cloud rates are dated access prices; revenue, investment, financing, lease value, and system acquisition estimates are not OEM hardware prices. No accepted OEM purchase price was verified Nebius prices · Lambda instances · AWS Capacity Blocks Dashboard scope This child session updated only the canonical report candidate; the dashboard was not changed here The report remains table-first and retains the dashboard constraint; no dashboard write is claimed

Current-run update — 9 September 2026

This additive reconciliation preserves every preceding dated section, table, structured-report element, and existing quantitative chart slot. The accepted evidence review reconciles 24 canonical project, site, program, or platform records after deduplication and 10 bounded GPU or deployment records. It supports a source-qualified ledger, not a global census: ND means not disclosed or not verified in the reviewed evidence, not zero; parent programs, child phases, utility power, generation, IT load, GPU targets, allocations, financing, and hosted prices are not interchangeable.

Current conclusion

The current screen is strongest on role-separated signals rather than totals. Named capacity records range from SB Energy OpenAI's 1.2 GW Milam County site under construction SB Energy, 12 Jan 2026 to G42 Khazna's nested 5 GW campus, 1 GW cluster, and 200 MW first build G42, 16 Oct 2025 . Named GPU signals range from xAI's current-page claim of 200,000 H100s xAI, accessed 9 Sep 2026 and Yotta NM1's operator-reported 8,192 H100 plus 1,024 L40 live Yotta, 23 Jun 2026 to DataSection's 4,696 purchased B200s DataSection, 4 Jun 2026 . These states are not audited or additive. No global capacity, electricity, installed-GPU, investment, or comparable-price total is asserted.

Reconciled data-center plans, construction, power, and backing

The rows below are the current high-value cross-check from the 24-record registry. Capacity remains in the source-defined class; electricity fields show a utility, grid, generation, PPA, or annual-energy signal only when the source states one.

Record Status and source date Source-defined capacity Electricity grid boundary Company, operator, and documented backing Capital or possible-price signal --- --- --- --- --- --- P01 · SB Energy OpenAI Stargate, Milam County, US Under construction; service expected from 2026 SB Energy, 12 Jan 2026 1.2 GW data-center site lease; not a utility-interconnection total New generation is planned; MW, interconnection, and usable IT load ND SB Energy builds operates for OpenAI; OpenAI and SoftBank are platform investors; Ares preferred equity is a company-financing signal OpenAI and SoftBank each invested US$500M in SB Energy; Ares secured about US$800M of redeemable preferred equity. Not site capex or GPU price P03 · Meta Hyperion, Louisiana, US Under construction expanding; completion and operating date ND Meta, 13 Jul 2026 5 GW compute capacity ; not relabelled IT or utility MW Entergy agreement; site grid and generation MW ND Meta operates and leases facilities; Blue Owl-managed funds own 80% and Meta 20% of the stated JV Meta JV, 21 Oct 2025 About US$27B JV development cost and US$7B Blue Owl cash contribution; later US$50B regional frame is separate P11 · DataVolt NEOM Oxagon, Saudi Arabia Planned; first phase expected operational by 2028 NEOM, 10 Feb 2025 1.5 GW AI factory; phase-one MW and IT split ND Renewable net-zero design intent; grid and generation MW ND DataVolt developer, investor, and operator; NEOM Oxagon land and infrastructure host; other investors lenders ND DataVolt US$5B initial investment for phase one; no close, tariff, lease, or GPU price P12 · Stargate UAE Khazna-G42, Abu Dhabi, UAE Under construction; first 200 MW targeted for 2026 G42, 16 Oct 2025 Nested 5 GW UAE-US campus, 1 GW cluster, 200 MW first build; not additive First-build figure is not a separate grid allocation; utility and generation MW ND G42 leads; Khazna develops; OpenAI, Oracle, NVIDIA, Cisco, and SoftBank are named strategic technology partners, not automatically project investors Project capex, debt, equity allocation, PPA, and GPU price ND P14 · Yotta D2, Greater Noida, India Committed target; completion target not independently verified Yotta, 18 Feb 2026 60 MW site, scalable to 250 MW ; site basis is not split into IT facility grid Grid-interconnection MW ND Yotta operator; NVIDIA engagement; project investor lender allocation ND More than US$2B deployment investment; later about US$2.8B aggregate contract value; neither is GPU price P18 · AirTrunk MEL2, Melbourne, Australia Planned site acquired; completion date ND AirTrunk, 24 Dec 2025 354+ MW hyperscale campus; not stated as IT load Diverse 66 kV feed listed; supply interconnection MW ND AirTrunk developer operator; customer and project co-investor allocation ND More than A$5B direct MEL2 investment and more than A$7B Melbourne platform investment; scopes are separate P19 · NEXTDC S4, Western Sydney, Australia In development; current page accessed 9 Sep 2026 NEXTDC S4 350 MW IT capacity on current page; original 300 MW target is historical, not additive Near major substation; binding grid MW and energization date ND NEXTDC developer operator; portfolio funding disclosed, S4-specific investor allocation ND A$2.2B portfolio capital plan; not isolated S4 capex or GPU price P24 · BDx Indonesia CGK3 CGK4 CGK5 Portfolio power commitments; current site states vary BDx power release, 3 Jun 2026 1.2 GW reported secured-power portfolio; components 788 MVA , about 60 MVA , and up to 385 MVA ; no conversion Separate loan release uses 1.2 GVA ; GW GVA discrepancy remains open; no IT total BDx owner operator; I Squared Capital, Indosat, Lintasarta, and PLN roles retained without inferred site percentages US$320M loan closed 1 Apr 2026; portfolio debt, not site-only capex or GPU price

GPU inventory, model specifications, company roles, and state

The quantities below preserve their states—operator-reported live, dated deployment, purchased, target, planned, or allocated. Product specifications are reference anchors, not extra inventory.

Record Quantity and state Model specification anchor Company, backing, and power context Price or capital context --- --- --- --- --- G1 · xAI Colossus, Memphis, US 200,000 NVIDIA H100 GPUs , current-page operator claim; 1M is a roadmap and excluded xAI Colossus H100 reference: 80 GB HBM3 and about 3.35 TB s for SXM; project form factor ND NVIDIA H100 xAI operator; NVIDIA platform supplier; MLGW utility request 150 MW is utility evidence, not IT load MLGW update xAI US$6B Series C is company financing; Colossus GPU purchase price and site allocation ND G2 · CoreWeave MERLIN Edged, Barcelona, Spain 10,224 NVIDIA H200 GPUs , dated site deployment announced 13 May 2025; current 2026 stock ND CoreWeave Barcelona H200: 141 GB HBM3e 4.8 TB s NVIDIA H200 CoreWeave operator; MERLIN Properties and Edged Energy facility partners; site is a 15 MW source-defined facility Site procurement price ND; wider CoreWeave investment is not assigned to Barcelona G3 · DataSection, Bangkok area, Thailand 4,696 individual B200 GPUs across 587 purchased servers ; delivery commissioning ND DCD report DGX B200 reference: 8 GPUs, 1,440 GB aggregate memory, 64 TB s aggregate bandwidth, about 14.3 kW max system power; not assigned per card NVIDIA DGX B200 DataSection operator developer; ASRock Rack supplier; unnamed US customer; prior power plan 10 MW , basis ND Reported estimated US$257.4M package agreement; may include servers and related terms, not per-card MSRP G5 · Yotta NM1, Navi Mumbai, India 8,192 H100 + 1,024 L40 live , operator-reported 23 Jun 2026; derived subtotal 9,216 is not a separately reported count Yotta H100 reference as above; L40 model form factor detail ND Yotta operator; NM1 page lists 52 MW IT power , 7,000 racks, PUE 1.5 NM1 page GPU purchase price, customer allocation, investor, and lender ND G6 · Mistral AI Eclairion, France 13,800 NVIDIA GB300 GPUs reported for deployment; earlier 18,000 GB200 disclosure is not reconciled or added CNBC GB300 NVL72 reference: 72 Blackwell Ultra GPUs + 36 Grace CPUs ; project rack count ND NVIDIA GB300 Eclairion reported owner operator; Mistral customer developer; 44 MW powered capacity ; seven-bank lender group US$830M debt for infrastructure GPU acquisition; financing scope, not OEM price G7 · Nscale Aker Stargate Norway 100,000 NVIDIA GPUs targeted by end-2026 ; separate 30,000 Rubin GPUs for Microsoft in 2027; not additive Nscale Vera Rubin NVL72 reference: 72 Rubin GPUs + 36 Vera CPUs; 100,000 target model mix ND NVIDIA Vera Rubin Nscale designs builds manages; planned 50 50 Nscale-Aker JV versus later sole-management wording; 230 MW campus and 20 MW initial phase About US$1B initial phase and US$790M bank financing; neither is GPU price or proof of delivery G9 · Noetra NVIDIA national AI infrastructure, Japan 27,500 Rubin GPUs + 13,750 Vera CPUs , announced planned; not delivered NVIDIA Japan Vera Rubin NVL72 reference as above; source-defined 140 MW data-center capacity, IT grid basis ND Noetra, NVIDIA, and METI-supported FRONTia named; title, investors, lenders, and procurement allocation ND Price, lease, capex, and procurement value ND G10 · Israel national AI supercomputer Nebius 1,000 NVIDIA B200 accelerators allocated over coming years ; infrastructure stated live operational, allocation is not installed count Israel Innovation Authority DGX B200 reference as above; project server count and power ND Israel Innovation Authority selected Nebius; project equity debt allocation ND Discounted access is described; tariff and hardware price ND

Role-separated capital and possible-price context

Economic signal What is documented Correct interpretation --- --- --- CoreWeave GPU-backed facility US$8.5B DDTL 4.0 announced closed 31 Mar 2026; Blackstone Credit & Insurance anchor; fixed tranche about 5.9% , floating tranche SOFR + 2.25% CoreWeave financing Capital cost and asset-backed financing; not a GPU quote or installed-fleet valuation Nscale Aker Norway Planned 50 50 JV; about US$1B initial 20 MW phase including up to US$250M equity on a 100%-project basis; later US$790M financing announcement Nscale Aker Announced committed project economics; close, draw, title transition, and GPU delivery remain open YTL Johor Current location page says up to 600 MW park and 72 MW JDC1 operational ; earlier records state RM1.5B first phase and RM1.1B arranged financing YTL locations Lenders arrangers and Sea's tenant role are not silently treated as equity investors; dated scope figures are not additive Cloud rental snapshots RunPod on 9 Sep: H100 PCIe US$2.89 , H100 SXM US$3.49 , H200 US$4.59 , B200 US$6.79 , B300 US$7.89 per GPU-hour; Lambda: H100 SXM US$3.99 , B200 SXM6 US$6.69 RunPod, Lambda Dynamic hosted-service list prices; not hardware purchase price, site capex, electricity tariff, or inventory H100 purchase proxy RunPod page updated 27 Aug 2026 cites US$25,000–30,000 per H100 PCIe card RunPod H100 PCIe Secondary provider proxy, not NVIDIA MSRP or a signed quote; excludes server, network, cooling, software, support, tax, and availability Electricity benchmark EIA June 2026 Texas industrial average 6.58¢ kWh = US$65.80 MWh EIA Table 5.6.A Retail benchmark only; not a named campus tariff, PPA strike price, demand charge, or interconnection fee

Current-run analytical view

The report retains the established quantitative chart set: source-defined capacity bars and status-qualified GPU or hosted-rate bars. A new line or range chart is not supportable without a time series or a normalized, non-nested comparison. The baseline chart contract therefore remains unchanged rather than fabricating a trend, uncertainty band, or global range.

Current-run controls and unresolved risks

Risk Treatment in this update Evidence and implication --- --- --- Global roll-up No global MW, TWh, installed-GPU, invested-capital, or comparable-price total is asserted The 24-record registry is bounded; mixed capacity classes, nested phases, MVA GVA, and source dates block addition Commissioning and inventory Live, purchased, target, planned, allocated, and historical states remain distinct Yotta D2, Mistral, Nscale, HUMAIN, Noetra, DataSection, and several campus phases still lack complete current commissioning or delivery proof Conflicting quantities Yotta 20,736 vs 30,000 , Mistral 13,800 vs 18,000 , Nscale 100,000 vs 30,000 same-campus targets, and HUMAIN's ambiguous 18,000 GB300 wording are retained without addition Evidence reconciliation · Yotta · Mistral financing Ownership and financing Developer, operator, host, tenant offtaker, supplier, lender, strategic partner, and equity investor remain separate roles Legal title, SPV, draw, and site allocation are incomplete for most records; no role is inferred from proximity Primary-reader blockers DataSection's official filing PDF and the MLGW xAI utility PDF were not page-readable in the accepted evidence pass; claims remain qualified The blocked sources are not treated as negative evidence; readable official secondary evidence is cited with its strength Price boundary No OEM list GPU price, signed BOM, site tariff, PPA strike price, or normalized colocation rate was established Cloud rates, provider purchase proxy, financing, transaction proceeds, revenue, lease value, and derived arithmetic retain their original economic labels Dashboard scope This report-only session did not change the dashboard The table-first dashboard requirement remains for its separate session; no dashboard completion is claimed

Current-run update — 10 September 2026

This additive update records the accepted 10 September evidence pass. It adds 14 normalized project records, 6 GPU or procurement records, 4 manufacturer specification anchors, and dated capital or price findings. ND means not disclosed or not verified in the reviewed evidence, not zero. The records are representative and non-additive: IT load, source-defined facility or compute capacity, utility or grid capacity, generation or PPA, GPU quantities, and financial objects remain separate.

Current conclusion

The pass confirms a large, state-qualified pipeline rather than a defensible global total. The clearest capacity signals are Meta Hyperion’s 5 GW compute-capacity plan , the adjacent Crusoe Microsoft Abilene 900 MW site with a planned 900 MW on-site power plant , SINES’ 1.2 GW secured IT capacity , ODATA QR03’s 300 MW full-build IT capacity with a 200 MW energized first energy phase , DataVolt HUMAIN’s 100 MW tranche of a 360 MW first phase under construction , and SoftBank Tomakomai’s more-than-300 MW expected power capacity . These figures mix source-defined boundaries and states and must not be summed. The clearest discrete-GPU signals are Sakura’s approximately 1,100 Blackwell GPUs operating , DataSection’s 4,696 B200 GPUs acquired , Noetra NVIDIA Japan’s 27,500 Rubin GPUs planned , and Firebird’s 41,000 additional GB300 GPUs authorized for sale and delivery .

Data-center project comparison

The table keeps possible or announced capacity, IT load, electricity grid evidence, roles, backing, and price basis in distinct fields. Source dates are included in the status field; evidence IDs map to the accepted evidence register.

Record Status source date Possible announced capacity IT load Electricity grid generation boundary Developer operator roles Investor funder Price cost basis GPU at site Evidence --- --- --- --- --- --- --- --- --- --- NA-01 · Crusoe Microsoft Abilene, US Announced; site preparation underway; first building expected mid-2027 2026-03-27 900 MW new campus site; not labelled IT ND Planned 900 MW behind-the-meter on-site plant; new-campus interconnect ND; adjacent existing campus has a separate 1.2 GW grid signal Crusoe developer operator; Microsoft customer support context; Lancium energy role Named lender equity sponsor ND Project price ND; power-plant figure is not capex ND for this new campus; related 50,000-system design is not assigned here E-NA-01, E-NA-02 · Crusoe 900 MW · Crusoe 1.2 GW NA-02 · Meta Hyperion, Louisiana, US Under construction expanding; completion and energized IT phase ND 2026-07-13 5 GW source-defined compute capacity ND Entergy reports 2,260 MW planned generation, substations, and 500 230 kV transmission; import interconnect rating ND Meta developer operator; Entergy Louisiana utility Meta reports more than US$50B Richland Parish investment; external lender equity sponsor ND Regional site investment scope, not GPU or electricity price ND E-NA-03, E-NA-04 · Meta · Entergy NA-03 · Meta Sturgeon, Alberta, Canada Groundbreaking; under construction; completion ND 2026-07-08 1 GW data center; source does not label it IT or utility import ND Grid generation MW MVA ND; Meta says it funds new generation and grid infrastructure and will match use with clean energy Meta developer operator; Frost Collective general-contractor JV; Greenlight, AltaLink, Capital Power, AESO partners Meta more than CA$13B site investment; approximately CA$60M local infrastructure; financing split ND Investment scope, not power or GPU price ND E-NA-05 · Meta Canada NA-04 · Applied Digital Polaris Forge 2, North Dakota, US Under construction; initial delivery in 2026, full 200 MW targeted early 2027 2026-01-07 800 MW additional right of first refusal; source-described 1 GW expansion potential 200 MW critical IT Power secured through Cass County Electric Cooperative; rating interconnect ND Applied Digital developer operator; unnamed hyperscaler tenant; Macquarie financing partner Macquarie facility context; senior secured notes US$2.15B notes priced 2026-03-04; financing for development construction, not quoted construction or GPU price ND E-NA-06, E-NA-07, E-NA-08 · Applied results · Applied notes LAT-01 · ODATA QR03, Querétaro, Mexico Under construction; first energy phase energized; later commissioning not independently revalidated 2025-02-27 2025-04-30 300 MW five-building full-build IT capacity Up to 300 MW full build; first building 72 MW IT 200 MW first energy phase; 400 kV switching substation, 6.2 miles of lines, and path to 400 MW energy infrastructure ODATA, an Aligned company; Ammper renewable-energy partner; anchor client names ND ODATA reports US$3.3B total campus investment; external split ND Campus investment, not tariff or GPU price ND E-LAT-01, E-LAT-02 · ODATA capacity · ODATA energization LAT-02 · Scala AI City, Brazil Planned; LOI and authorized connection; construction energization ND 2024-09-11 2025-05-14 4.75 GW source-defined full-build campus 54 MW initial IT 5 GW National Interconnected System connection authorization; PPAs described without MW or strike price Scala developer operator; Rio Grande do Sul government Sedec partner Project equity lender split ND US$500M initial first-phase investment; not GPU or electricity price ND E-LAT-03, E-LAT-04 · Scala investment · Scala connection EU-01 · SINES Data Campus, Portugal SIN01 operating; SIN02 ready-for-service from 2026; six-building completion targeted by 2030 2024-08-08 2025-04-04 2026-05-12 1.2 GW secured IT ; SIN02–06 up to 240 MW each; phases are nested 1,200 MW secured campus IT; SIN01 26 MW current facility Grid power secured; exact connection rating ND; 100% renewable design target, PUE 1.1 Start Campus developer operator; REN transmission-system role Davidson Kempner and Pioneer Point named shareholders funders €8.5B combined privately funded construction investment; not tariff or GPU price More than 66,000 Vera Rubin GPUs planned from late 2027; separate GPU row E-EU-01, E-EU-02, E-EU-03 · SINES · SIN01 · GPU plan EU-02 · Pure DC SJK01, Finland Announced early development; first data-hall substation live; completion ND 2026-07-14 550 MW+ full-campus pathway; repeatable 40 MW modules 110 MW phase 1; 550 MW+ full-campus IT pathway More than 700 MVA renewable power access; MVA retained and not converted to MW Pure Data Centres developer operator; SDC Ventures local partner; Seinäjoki municipality host Lender equity allocation ND More than €1.5B phase 1 and more than €7.5B full campus; investment scopes ND E-EU-04 · Pure DC MEA-01 · DataVolt HUMAIN Oxagon, Saudi Arabia Construction underway; first 100 MW anticipated in 2028 2026-08-31 100 MW tranche of 360 MW first phase; wider planned AI factory 1.5 GW ND Renewable integration planned; utility import, generation, storage, and interconnect ratings ND DataVolt developer operator; HUMAIN joint-development partner; NEOM Oxagon host infrastructure role DataVolt initial first-phase funding; equity lender split ND US$5B initial first-phase funding; not GPU price ND E-MEA-01, E-MEA-02 · HUMAIN DataVolt · NEOM agreement MEA-02 · iXAfrica NBOX1, Kenya NBOX1.1 live; NBOX1.2 planned; page accessed 2026-09-10 22.5 MW full-site IT; NBOX1.2 is 18 MW planned 4.5 MW live; 22.5 MW full site Initial 2×4 MVA via 2×11 kV; planned 2×40 MVA 66 kV substation; MVA not converted iXAfrica operator developer; Future-tech design; KenInvest support Helios Investment Partners committed US$50M growth capital in 2022 Growth-capital scope, not current site capex, tariff, or GPU price ND E-MEA-03, E-MEA-04 · Campus specs · 4.5 MW facility MEA-03 · Firebird AI factory, Armenia Operational launch announced; expansion roadmap remains future 2026-08-08 300 MW AI-infrastructure target by end-2027; not labelled IT ND Grid interconnect generation MW ND Firebird developer operator; NVIDIA technology intended-investment role; CoreWeave earlier-investment role; Armenian US government participation NVIDIA CoreWeave investment amounts ND US$4B Phase 2 and prior US$500M Phase 1; project scope, not GPU price More than 70,000 Rubin Blackwell roadmap GPUs; 41,000-GB300 procurement is separate E-MEA-05, E-MEA-06 · Phase 2 · Launch APAC-01 · STT Jakarta, Indonesia Jakarta 2 launched; Jakarta 3 topped out; Jakarta 5 6 groundbreaking 2026-06-10 More than 360 MW AI-ready sustainable IT pipeline; phase split ND More than 360 MW pipeline; not current live load Secured power and high-voltage substation reported; rating import MW ND; PLN renewable arrangement referenced STT GDC developer operator; BKPM and Singapore government partnership context Green-financing amount, equity, and lender allocation ND Price capex ND GPU model and units ND E-APAC-01 · STT GDC APAC-02 · SoftBank Tomakomai, Japan Under construction; 60–70% complete; fiscal-2026 opening target ending March 2027 2025-05-01 2026-02-25 Expected power capacity more than 300 MW ; not labelled IT ND Local renewable procurement planned; utility import and generation MW ND SoftBank Corp. and IDC Frontier developer operator; Hokkaido government and METI roles METI subsidy acknowledged; amount ND Project cost and GPU price ND ND E-APAC-02, E-APAC-03 · Progress · Construction APAC-03 · Noetra NVIDIA Japan national AI factory Announced; site, construction, delivery, and financial close ND 2026-07-16 140 MW data-center capacity; not relabelled IT ND Grid interconnect generation MW ND Noetra project; NVIDIA technology partner; METI FRONTia and Japanese industrial leaders Investor lender allocation ND Project price ND 27,500 Rubin GPUs + 13,750 Vera CPUs planned E-APAC-04, E-SPEC-04 · NVIDIA Japan APAC-04 · DataSection Bangkok-area AI data center, Thailand Under development; equipment introduction planned; DCD reported phased operations but live status not independently confirmed 2026-06-04 2026-06-05 ND ND DCD’s secondary 10 MW power reference is not treated as IT load; utility interconnect ND DataSection developer operator; ASRock Rack supplier; unnamed US technology customer Named institutional investor ND; financing roles reported by DCD DCD secondary estimate US$257.4M for server systems; official HTML count source does not state price 4,696 B200 GPUs acquired for deployment E-APAC-05, E-APAC-06 · DataSection · DCD

GPU inventory and procurement comparison

These rows retain the unit boundary and state. Rack system counts are not multiplied into discrete-GPU totals, and planned or authorized units are not presented as operating inventory.

Record site Units and state Model specification boundary Company operator customer Investor funder Price cost basis Evidence --- --- --- --- --- --- --- GPU-01 · Crusoe Abilene, US Up to 50,000 GB200 NVL72 systems per building ; design target, not installed discrete GPUs GB200 NVL72 reference: 72 Blackwell GPUs + 36 Grace CPUs in one liquid-cooled rack Crusoe Lancium; Oracle workload context Project investor and GPU procurement price ND ND; system count is not a card price E-GPU-01, E-SPEC-01 · Crusoe · NVIDIA GB200 GPU-02 · SINES, Portugal More than 66,000 GPUs planned from late 2027; not delivered stock Vera Rubin NVL72 reference: 72 Rubin GPUs + 36 Vera CPUs Start Campus and Nscale; customer operator allocation beyond announcement ND GPU-specific funder and price ND; project funding is reported in EU-01 ND E-EU-03, E-SPEC-04 · Start Campus · NVIDIA Rubin GPU-03 · Japan national AI infrastructure 27,500 Rubin GPUs + 13,750 Vera CPUs planned; not delivered or installed Rubin Vera program on Vera Rubin NVL72 DSX; model is Rubin, not Blackwell Noetra and NVIDIA; METI FRONTia national-program context Investor lender and GPU price ND ND; 140 MW is source-defined data-center capacity, not per-GPU draw E-APAC-04, E-SPEC-04 · NVIDIA Japan GPU-04 · DataSection Bangkok, Thailand 4,696 B200 GPUs in 587 servers ; acquired for planned deployment; live status ND B200 DGX B200 reference: 8 GPUs , 1,440 GB total GPU memory, 64 TB s HBM3e; approximately 14.3 kW is a DGX system reference, not cluster draw DataSection operator buyer; ASRock Rack supplier; unnamed US customer Customer advances borrowings reported by DCD; named institutional investor ND DCD secondary estimate US$257.4M for server systems; not card-only MSRP E-APAC-05, E-APAC-06, E-SPEC-02 · DataSection · NVIDIA DGX B200 GPU-05 · Firebird Armenia Additional 41,000 GB300 GPUs authorized for sale delivery in Phase 2; procurement authorization, not installed inventory GB300 NVL72 reference: 72 Blackwell Ultra GPUs + 36 Grace CPUs , liquid-cooled rack Firebird; NVIDIA and US export-authorization context; Armenia host NVIDIA intended investment and CoreWeave earlier investment; amounts ND US$4B Phase 2 scope, not GPU purchase price E-MEA-05, E-SPEC-03 · Firebird Phase 2 · NVIDIA GB300 GPU-06 · Sakura Internet Ishikari, Japan Approximately 1,100 Blackwell GPUs operating as of 2026-02-25; company-reported live infrastructure Exact SKU, memory, TDP, and rack topology ND; do not substitute DGX B200 or GB300 specs Sakura Internet operator site owner; NVIDIA platform supplier Project finance and GPU purchase price ND ND E-GPU-06 · Sakura

Product specification anchors — interpretation only

Manufacturer specifications describe products, not site inventory, interconnection ratings, or cluster draw unless a project source explicitly says so.

Spec ID Product Manufacturer-reported specification Power boundary Evidence --- --- --- --- --- E-SPEC-01 NVIDIA GB200 NVL72 36 Grace CPUs and 72 Blackwell GPUs in one liquid-cooled rack-scale design; 130 TB s low-latency GPU communications No site load inferred NVIDIA GB200 E-SPEC-02 NVIDIA DGX B200 8 Blackwell GPUs; 1,440 GB total GPU memory; 64 TB s HBM3e bandwidth Approximately 14.3 kW maximum system power; not DataSection cluster draw NVIDIA DGX B200 E-SPEC-03 NVIDIA GB300 NVL72 72 Blackwell Ultra GPUs and 36 Grace CPUs in a fully liquid-cooled rack-scale architecture No site load inferred NVIDIA GB300 E-SPEC-04 NVIDIA Vera Rubin NVL72 72 Rubin GPUs and 36 Vera CPUs in a rack-scale platform No site load inferred NVIDIA Vera Rubin

Capital, price, and electricity findings

Economic signal Reported value and source date Correct boundary Evidence --- --- --- --- Meta Hyperion More than US$50B Richland Parish investment 2026-07-13 Site program investment; not GPU, electricity, or tariff price E-NA-03 · Meta Meta Sturgeon More than CA$13B site investment and approximately CA$60M local infrastructure 2026-07-08 Site and local-infrastructure scopes; not GPU or power price E-NA-05 · Meta Canada ODATA QR03 US$3.3B total campus investment 2025-04-30 Campus investment; not tariff, grid fee, or GPU price E-LAT-01 · ODATA Scala AI City US$500M initial first-phase investment 2024-09-11 Initial project investment; not 5 GW connection cost or GPU price E-LAT-03 · Scala SINES Data Campus €8.5B combined privately funded construction investment value 2024-08-08 2025-04-04 Campus investment; not tariff or GPU price E-EU-01, E-EU-02 · SINES Pure DC SJK01 More than €1.5B phase 1 and more than €7.5B full-campus pathway 2026-07-14 Project investment pathways; not electricity or GPU price E-EU-04 · Pure DC DataVolt NEOM US$5B initial first-phase funding 2025-02-10 Phase program funding; equity split, tariff, and GPU price ND E-MEA-02 · NEOM Applied Digital Polaris Forge 2 US$2.15B senior secured notes 2026-03-04 Financing proceeds for 200 MW critical IT development construction; not quoted cost per MW E-NA-07 · Applied DataSection Thailand Secondary estimate US$257.4M approximately ¥41.13B for 587 server systems 2026-06-05 ; official HTML confirms units but not price Server-system acquisition estimate; not card-only price or live-deployment proof E-APAC-05, E-APAC-06 · DataSection · DCD Electricity and OEM price screen No selected primary source disclosed a campus tariff, demand charge, interconnection fee, PPA strike price, or OEM GPU quote in this pass cutoff 2026-09-10 No $ MW, $ MWh, $ GPU, or annual operating-cost estimate is inferred Evidence register E-NA-01–E-SPEC-04

Current-run analytical view and unresolved risks

The pinned baseline’s quantitative chart slots remain unchanged: they are source-defined capacity and status-qualified GPU or hosted-rate bars. The current packet does not support adding a second chart shape without mixing non-comparable boundaries: project observations mix IT, facility, compute, grid, generation, phase, and full-build measures, while GPU observations mix operating, acquired, planned, and authorized states. A line, range, or GPU-versus-power scatter is therefore not asserted; missing observations are not plotted as zero.

Risk Treatment in this update Evidence and implication --- --- --- Global roll-up No global MW, electricity, installed-GPU, investment, or comparable-price total is asserted Parent programs, child phases, utility resources, GPU systems, and procurement states overlap or use different units Electricity boundary IT load is distinct from source-defined facility compute capacity, grid interconnection, generation PPA, and MVA ODATA’s 300 MW IT and 200 MW energized energy phase; SJK01’s more-than-700 MVA access; Meta’s 2,260 MW planned generation are not interchangeable Inventory state Operating, acquired, planned, and authorized units remain separate Sakura, DataSection, Noetra, and Firebird are not summed; GB200 GB300 rack-system counts are not converted into discrete GPUs Source strength Self-reported announcements are retained with dates and evidence IDs; secondary values are marked secondary DataSection’s official HTML supports the unit count, while the price remains a DCD estimate; some current operating states remain unverified Price boundary Project investment, financing, server-system acquisition, tariffs, and OEM price remain separate No accepted OEM quote, site tariff, PPA strike price, or normalized $ MW or $ GPU value was found

Current-run update — 11 September 2026

Cutoff: 11 September 2026 UTC. This incremental update uses the accepted current-pass evidence and leaves prior dated records, chart slots, and the established structured report block in place. ND means not disclosed or not verified in this pass; it is not zero. Parent programs, child phases, options, utility assets, GPU systems, and financial signals are not added together.

Current conclusion

The current source set confirms a large, non-additive pipeline rather than a defensible global total. The most decision-relevant new signals are an Australia-wide NVIDIA DSX ecosystem target of up to 2 GW by 2027, live B300 capacity at QumulusAI’s Philadelphia colocation site with no disclosed unit count, and a separate QumulusAI purchase of 1,632 B300 GPUs across 204 HGX B300 systems. The current pass also strengthens evidence for Abilene site preparation, Meta Hyperion’s compute plan and associated generation, SINES and Pure DC IT envelopes, DataVolt construction, SoftBank Tomakomai progress, and the named regional GPU programs. These observations remain source-qualified and are not a global capacity, electricity, or installed-GPU census.

Program and completion updates kept outside site totals

Program or site Current state New evidence signal Explicit boundary Evidence --- --- --- --- --- NVIDIA Australia DSX ecosystem Announced 9 Sep 2026; target through 2027 Up to 2 GW of DSX AI-factory buildout across Firmus, Sharon AI, IREN, Megaport, ResetData, CDC, NEXTDC, and AirTrunk; DCD reports Sharon AI up to 68,000 NVIDIA GPUs Ecosystem compute target, not site IT load, utility import, installed GPU stock, or project financing; IREN’s 800 MW Bundey reference is not added NVIDIA · DCD QumulusAI Philadelphia colocation Company-reported live capacity after full customer handoff on 1 Sep 2026; release dated 10 Sep NVIDIA Blackwell B300 capacity is live and in production; a two-year US$18M take-or-pay agreement is activated Completion release gives no GPU quantity or site IT MW; US$18M is a customer contract signal, not GPU price or site capex QumulusAI handoff QumulusAI B300 procurement Purchased 20 Jul 2026; deployment location not assigned from the purchase release 1,632 B300 GPUs across 204 HGX B300 systems, plus 192 RTX PRO 6000 Blackwell GPUs; primary financing arrangements named Company-level order across a national network; do not assign all units to Philadelphia or treat financing as hardware price QumulusAI purchase

Regional project and construction register

Region Project location Status and timing IT load Facility compute Grid interconnection Generation PPA Company role Investor funder Evidence --- --- --- --- --- --- --- --- --- --- North America Crusoe Microsoft adjacent AI factory campus, Abilene, Texas, US Announced; land clearing and site preparation underway 27 Mar 2026; first building expected energized mid-2027 ND 900 MW new campus site; not labelled IT New-campus grid interconnect rating ND; existing adjacent Crusoe campus separately has a 1.2 GW grid interconnection Planned 900 MW behind-the-meter on-site power plant; operating generation ND Crusoe developer operator; Microsoft AI-infrastructure customer context; Lancium energy-infrastructure role Project price and named lender or equity sponsor ND E-NA-01 Crusoe · E-NA-02 Crusoe North America Meta Hyperion, Richland Parish, Louisiana, US Under construction expansion; Meta update 13 Jul 2026; completion and energized IT phase ND ND 5 GW compute capacity; source-defined compute headline, not relabelled IT Interconnection import rating ND; Entergy reports substations, nearly 100 miles of 500 kV lines, and eight new 230 kV lines 2,260 MW planned generation reported by Entergy; separate from Meta’s compute figure Meta developer operator customer; Entergy Louisiana utility More than US$50B Richland Parish investment; external lender or equity sponsor ND Meta · Entergy North America Meta Sturgeon Data Centre, Sturgeon County, Alberta, Canada Groundbreaking 8 Jul 2026; under construction; completion ND ND 1 GW data center; source does not label it IT or utility import Grid rating ND; Meta says it is funding new generation and grid infrastructure Meta says it will match 100% of electricity use with clean energy; generation rating ND Meta developer operator; Frost Collective general-contractor JV; Greenlight, AltaLink, Capital Power, and AESO partner roles More than CA$13B site investment and approximately CA$60M local infrastructure improvements Meta Canada North America Applied Digital Polaris Forge 2, Harwood, North Dakota, US Under construction; initial delivery in 2026; full 200 MW planned in early 2027 200 MW leased critical IT 800 MW additional right of first refusal; source-described 1 GW expansion potential Power secured through Cass County Electric Cooperative; rating and interconnect MW ND Generation and PPA details ND Applied Digital developer operator; unnamed investment-grade hyperscaler tenant; Macquarie financing partner US$2.15B senior secured notes for development and construction; not a quoted construction cost or GPU price Applied results · Applied financing Latin America ODATA DC QR03, Querétaro, Mexico Under construction first energy phase energized; later commissioning not independently revalidated Up to 300 MW full build; first building 72 MW IT Five-building campus; 300 MW total IT capacity 200 MW energized first energy phase; 400 kV switching substation, 6.2 miles of transmission lines, two new substations; path described to 400 MW energy infrastructure Ammper renewable-energy supply role; tariff and PPA price ND ODATA, an Aligned Data Centers company; three anchor hyperscale clients named by ODATA, names ND US$3.3B total campus investment; external lender equity allocation ND ODATA capacity · ODATA energization Latin America Scala AI City, Rio Grande do Sul, Brazil Planned LOI and authorized connection; construction and energization ND 54 MW initial IT 4.75 GW source-defined full-build campus capacity Brazil MME authorized a 5 GW ramp-up connection to the National Interconnected System Long-term wind and hydro PPAs described; MW and strike price ND Scala Data Centers developer operator; Rio Grande do Sul government strategic partner US$500M initial first-phase investment; named project equity lender split ND Scala investment · Scala connection Europe SINES Data Campus, Sines, Portugal SIN01 operating; SIN02 ready-for-service from 2026; six-building campus targeted for completion by 2030 1,200 MW secured campus IT; SIN01 26 MW current facility Campus phases are nested; SIN02–06 up to 240 MW each and are not added to 1,200 MW Grid power secured; exact connection rating ND 100% renewable-energy design target; seawater cooling and PUE 1.1; no hourly delivery claim Start Campus developer operator; REN transmission-system role Davidson Kempner Capital Management and Pioneer Point Partners named shareholders funders; €8.5B combined privately funded construction investment value SINES campus · SIN01 · GPU plan Europe Pure DC SJK01, Seinäjoki, Finland Announced early development; first data-hall substation live; phase-1 permissions and power requirements secured 110 MW phase 1 IT; 550 MW+ full-campus IT pathway Repeatable 40 MW AI-ready modules More than 700 MVA renewable power access; MVA not converted to MW Renewable access reported; tariff and PPA strike price ND Pure Data Centres Group developer operator; SDC Ventures local partner; municipality host More than €1.5B phase 1 and more than €7.5B full-campus investment; lender equity allocation ND Pure DC Middle East Africa DataVolt–HUMAIN AI data center, Oxagon, NEOM, Saudi Arabia Construction underway as of 31 Aug 2026; first 100 MW anticipated in 2028 ND 100 MW tranche of a 360 MW first phase; wider planned campus 1.5 GW Utility import, storage, and interconnect ratings ND Renewable-energy integration planned; generation rating ND DataVolt developer operator; HUMAIN joint-development partner; NEOM Oxagon host and infrastructure role DataVolt initial US$5B first-phase funding; equity and lender split ND NEOM HUMAIN · NEOM agreement Middle East Africa iXAfrica NBOX1, Nairobi, Kenya NBOX1.1 live; NBOX1.2 planned; campus page accessed 10 Sep 2026 4.5 MW live; 22.5 MW full site; NBOX1.2 18 MW planned Two-building campus; values are nested Initial 2×4 MVA via 2×11 kV feeders; planned 2×40 MVA 66 kV substation; MVA not converted Utility feeds reported; tariff and PPA price ND iXAfrica operator developer; Future-tech design partner; KenInvest support Helios Investment Partners committed US$50M growth capital in 2022; growth capital is not site capex Campus specs · 4.5 MW facility Middle East Africa Firebird AI factory, Hrazdan, Armenia Operational launch announced 8 Aug 2026; expansion roadmap future ND 300 MW AI-infrastructure target by end-2027; not labelled IT Grid and interconnect MW ND Generation and PPA details ND Firebird developer operator; NVIDIA technology and intended-investment role; CoreWeave earlier-investment role; Armenian and US government participation US$4B Phase 2 and prior US$500M Phase 1; project scope, not GPU price Firebird Phase 2 · Firebird launch Asia-Pacific STT Jakarta campus, Greater Jakarta, Indonesia Jakarta 2 launched; Jakarta 3 topped out; Jakarta 5 and 6 groundbreaking More than 360 MW AI-ready sustainable IT pipeline; phase split ND Aggregate development pipeline, not current live load Secured power and high-voltage substation reported; rating and import MW ND PLN renewable arrangement referenced; tariff and PPA price ND STT GDC developer operator; Indonesian BKPM and Singapore government partnership context Green-financing amount, equity, and lender allocation ND STT GDC Asia-Pacific SoftBank Tomakomai AI data center, Hokkaido, Japan Under construction; about 60–70% complete on 25 Feb 2026; fiscal-2026 opening target ND Expected power capacity more than 300 MW; not labelled IT Utility import rating ND Local Hokkaido renewable procurement planned; generation MW ND SoftBank Corp. and IDC Frontier developer operator; Hokkaido government and METI roles METI subsidy acknowledged; amount and project cost ND Progress · Construction Asia-Pacific Noetra NVIDIA national AI factory, Japan; site undisclosed Announced 16 Jul 2026; siting, construction, delivery, and financial close ND ND 140 MW data-center capacity based on DSX; not relabelled IT Grid, interconnect, and generation MW ND ND Noetra project; NVIDIA technology partner; METI FRONTia and Japanese industrial leaders support national program Investor lender allocation and price ND NVIDIA Japan Asia-Pacific DataSection AI data center, Bangkok vicinity, Thailand Under development; equipment introduction planned; live status not independently confirmed ND ND; secondary 10 MW power reference not treated as IT load Utility and interconnect rating ND Generation and PPA details ND DataSection developer operator; ASRock Rack supplier; unnamed US technology customer DCD secondary estimate of US$257.4M for server systems; official HTML count source does not state price DataSection · DCD

Boundary ledger: IT, facility or compute, grid or interconnection, and generation or PPA

Project IT load Facility compute Grid interconnection Generation PPA Evidence --- --- --- --- --- --- Abilene new campus ND 900 MW new campus site New-campus rating ND; adjacent existing campus has 1.2 GW grid interconnection Planned 900 MW on-site plant Crusoe Meta Hyperion ND 5 GW compute capacity 500 kV and 230 kV transmission and substations reported; import rating ND 2,260 MW planned generation Meta · Entergy ODATA QR03 Up to 300 MW full-build IT; first building 72 MW IT Five-building campus 200 MW energized energy phase; 400 kV switching substation and path to 400 MW energy infrastructure Renewable supply role reported; tariff and PPA price ND ODATA Scala AI City 54 MW initial IT 4.75 GW full-build campus 5 GW ramp-up connection authorization Wind and hydro PPAs described; MW and strike price ND Scala SINES 1.2 GW secured campus IT; SIN01 26 MW Nested six-building campus; SIN02–06 up to 240 MW each Grid power secured; exact connection rating ND 100% renewable design target; no hourly delivery claim Start Campus Pure DC SJK01 110 MW phase 1 IT; 550 MW+ full-campus pathway 40 MW repeatable AI-ready modules More than 700 MVA renewable access; MVA not converted Renewable access reported; tariff and PPA strike price ND Pure DC DataVolt–HUMAIN ND 100 MW tranche of 360 MW first phase; 1.5 GW wider campus Utility and interconnect ratings ND Renewable integration planned; generation rating ND NEOM SoftBank Tomakomai ND More than 300 MW expected power capacity Utility import rating ND Local renewable procurement planned SoftBank

GPU inventory, state, and specification ledger

Record site Units and state at cutoff Model and specification boundary Company operator customer Investor funder Price or price basis Evidence --- --- --- --- --- --- --- Crusoe Abilene, US Up to 50,000 GB200 NVL72 systems per building; design target, not installed discrete GPUs GB200 NVL72 reference: 72 Blackwell GPUs and 36 Grace CPUs in one liquid-cooled rack Crusoe Lancium; Oracle workload context Project investor and GPU procurement price ND ND; system count is not a card price Crusoe · NVIDIA GB200 SINES, Portugal More than 66,000 GPUs planned from late 2027; not delivered stock Vera Rubin NVL72 reference: 72 Rubin GPUs and 36 Vera CPUs Start Campus and Nscale; allocation beyond announcement ND GPU-specific funder and price ND; project funding is in the SINES row ND Start Campus AI future · NVIDIA Vera Rubin Japan national AI infrastructure 27,500 Rubin GPUs and 13,750 Vera CPUs planned; not delivered or installed Rubin Vera program on Vera Rubin NVL72 or DSX; model is Rubin, not Blackwell Noetra and NVIDIA; METI FRONTia national-program context Investor lender and GPU price ND ND; 140 MW is source-defined data-center capacity, not per-GPU draw NVIDIA Japan DataSection Bangkok, Thailand 4,696 B200 GPUs in 587 servers; acquired for planned deployment; live status ND B200 DGX B200 reference: 8 GPUs, 1,440 GB total GPU memory, 64 TB s HBM3e; approximately 14.3 kW is a DGX system reference, not cluster draw DataSection operator buyer; ASRock Rack supplier; unnamed US customer Customer advances and borrowings reported by DCD; named institutional investor ND DCD secondary estimate US$257.4M for server systems; not card-only MSRP DataSection · DCD · NVIDIA DGX B200 Firebird Armenia Additional 41,000 GB300 GPUs authorized for sale and delivery in Phase 2; procurement authorization, not installed inventory GB300 NVL72 reference: 72 Blackwell Ultra GPUs and 36 Grace CPUs, liquid-cooled rack Firebird; NVIDIA and US export-authorization context; Armenia host NVIDIA intended investment and CoreWeave earlier investment; amounts ND US$4B Phase 2 scope, not GPU purchase price Firebird · NVIDIA GB300 Sakura Internet Ishikari, Japan Approximately 1,100 Blackwell GPUs operating as of 25 Feb 2026; company-reported live infrastructure Exact SKU, memory, TDP, and rack topology ND Sakura Internet operator site owner; NVIDIA platform supplier Project finance and GPU purchase price ND ND Sakura QumulusAI national AI-cloud network 1,632 B300 GPUs purchased 20 Jul 2026 across 204 HGX B300 systems, plus 192 RTX PRO 6000 Blackwell GPUs; not assigned to a site B300 system memory, TDP, and topology ND in the purchase release QumulusAI operator; Supermicro and Lenovo named as OEM examples Technology Finance Corporation and USD.ai primary financing arrangements Hardware purchase price ND; more than US$124M inference agreements are customer-contract context QumulusAI purchase QumulusAI Philadelphia colocation B300 capacity live after 1 Sep 2026 handoff; unit count ND B300 model identified; exact site topology and specifications ND QumulusAI operator; customer and colocation landlord ND US$18M two-year take-or-pay agreement activated; not hardware financing Customer contract signal, not GPU price or site capex QumulusAI handoff Sharon AI Australia DSX ecosystem Up to 68,000 NVIDIA GPUs targeted for deployment; program target, not installed inventory GPU model s not disclosed; DSX is an infrastructure platform reference, not a GPU SKU Sharon AI partner; NVIDIA platform and software role; other Australian partners named in source Investor, lender, hardware price, and site allocation ND ND; up-to-2 GW is ecosystem compute program scope NVIDIA Australia · DCD

Product specification anchors — interpretation only

Manufacturer specifications describe products, not site inventory, interconnection ratings, or cluster draw unless a project source explicitly says so.

Spec ID Product Manufacturer-reported specification Power boundary Evidence --- --- --- --- --- E-SPEC-01 NVIDIA GB200 NVL72 36 Grace CPUs and 72 Blackwell GPUs in one liquid-cooled rack-scale design; 130 TB s low-latency GPU communications No site load inferred NVIDIA GB200 E-SPEC-02 NVIDIA DGX B200 8 Blackwell GPUs; 1,440 GB total GPU memory; 64 TB s HBM3e bandwidth Approximately 14.3 kW maximum system power; not DataSection cluster draw NVIDIA DGX B200 E-SPEC-03 NVIDIA GB300 NVL72 72 Blackwell Ultra GPUs and 36 Grace CPUs in a fully liquid-cooled rack-scale architecture No site load inferred NVIDIA GB300 E-SPEC-04 NVIDIA Vera Rubin NVL72 72 Rubin GPUs and 36 Vera CPUs in a rack-scale platform No site load inferred NVIDIA Vera Rubin

Company, investor, and funder role ledger

Program site Developer, operator, customer Utility, supplier, host, or government role Investor funder Role boundary Evidence --- --- --- --- --- --- Abilene new campus Crusoe developer operator; Microsoft customer context; Lancium energy role On-site power plant planned; new-campus utility rating ND Named project lender and equity sponsor ND Existing adjacent Crusoe campus and its 1.2 GW interconnection are not assigned to the new 900 MW site Crusoe Meta Hyperion Meta developer operator customer Entergy Louisiana utility; construction partners named by Meta Meta more-than-US$50B investment; external lender equity sponsor ND Investment and generation are not IT load or GPU price Meta · Entergy Meta Sturgeon Meta developer operator; Frost Collective general-contractor JV Greenlight, AltaLink, Capital Power, and AESO partner roles Meta fully funds new generation and grid infrastructure; more than CA$13B site investment Local infrastructure and site investment are separate scopes Meta Canada SINES Start Campus developer operator; Nscale GPU deployment context REN transmission-system role; seawater cooling design Davidson Kempner Capital Management and Pioneer Point Partners named shareholders funders €8.5B campus value is not a GPU price or tariff Start Campus DataVolt–HUMAIN DataVolt developer operator; HUMAIN joint-development partner NEOM Oxagon land and infrastructure host role DataVolt initial US$5B first-phase funding; equity lender split ND Funding is phase program scope, not electricity price or GPU cost NEOM iXAfrica NBOX1 iXAfrica developer operator; Future-tech design partner KenInvest government-investment support; utility feeds described by operator Helios Investment Partners US$50M growth capital commitment Growth capital is not current site capex or GPU price iXAfrica Applied Digital Polaris Forge 2 Applied Digital developer operator; unnamed hyperscaler tenant Cass County Electric Cooperative power role Macquarie Asset Management financing partner; US$2.15B senior secured notes Debt proceeds are financing scope, not quoted cost per MW or GPU price Applied QumulusAI B300 network QumulusAI operator; Supermicro and Lenovo OEM examples Customer and site allocation vary across network Technology Finance Corporation and USD.ai primary financing arrangements Financing is attached to company-level procurement; Philadelphia unit count remains ND QumulusAI

Price and economic signals

Economic signal Reported value and source date Economic type and boundary Evidence --- --- --- --- Meta Hyperion More than US$50B Richland Parish investment, 13 Jul 2026 Site program investment; not GPU, electricity, or tariff price Meta Meta Sturgeon More than CA$13B site investment and approximately CA$60M local infrastructure, 8 Jul 2026 Site and local-infrastructure scopes; not GPU or power price Meta Canada ODATA QR03 US$3.3B total campus investment, 30 Apr 2025 Campus investment; not tariff, grid fee, or GPU price ODATA Scala AI City US$500M initial first-phase investment, 11 Sep 2024 Initial project investment; not 5 GW connection cost or GPU price Scala SINES Data Campus €8.5B combined privately funded construction investment value, 8 Aug 2024 and 4 Apr 2025 Campus investment; not tariff or GPU price SINES · SIN01 Pure DC SJK01 More than €1.5B phase 1 and more than €7.5B full-campus pathway, 14 Jul 2026 Project investment pathways; not electricity or GPU price Pure DC DataVolt NEOM US$5B initial first-phase funding, 10 Feb 2025 Phase program funding; equity split, tariff, and GPU price ND NEOM Applied Digital Polaris Forge 2 US$2.15B senior secured notes, 4 Mar 2026 Financing proceeds for 200 MW critical IT development and construction; not quoted cost per MW Applied DataSection Thailand Secondary estimate US$257.4M for 587 server systems, 5 Jun 2026; official HTML confirms units but not price Server-system acquisition estimate; not card-only price or live-deployment proof DataSection · DCD Electricity and OEM price screen No selected primary source disclosed a campus tariff, demand charge, interconnection fee, PPA strike price, or OEM GPU quote in this pass No $ MW, $ MWh, $ GPU, or annual operating-cost estimate is inferred Evidence links retained beside each row above

Current-run analytical view and unresolved risks

The established report’s quantitative chart slots remain unchanged, including the non-additive capacity comparison and the GPU or hosted-rate comparison. No additional chart is added from this pass because the observations mix IT, facility, compute, grid, generation, phase, and full-build measures, while GPU observations mix operating, acquired, planned, targeted, and authorized states. Missing observations are not plotted as zero.

Risk Retained treatment Evidence and implication --- --- --- Global roll-up No global MW, electricity, installed-GPU, investment, or comparable-price total is asserted Parent programs, child phases, utility resources, GPU systems, and procurement states overlap or use different units Electricity boundary IT load is distinct from facility or compute capacity, grid interconnection, generation PPA, and MVA ODATA’s 300 MW IT and 200 MW energized energy phase, SJK01’s more-than-700 MVA access, and Meta’s 2,260 MW planned generation are not interchangeable Inventory state Operating, acquired, planned, targeted, and authorized units remain separate Sakura, DataSection, Noetra, Firebird, QumulusAI, and Sharon AI are not summed; rack-system counts are not converted into discrete GPUs Source strength Self-reported announcements retain source dates and direct citations; secondary values are marked secondary DataSection’s official HTML supports the unit count while the price remains a DCD estimate; several operating states remain unverified Price boundary Project investment, financing, server-system acquisition, tariffs, and OEM price remain separate No accepted OEM quote, site tariff, PPA strike price, or normalized $ MW or $ GPU value was found Commissioning and allocation Future phases, site preparation, and company-level GPU orders are not upgraded to live site inventory Abilene, Hyperion, ODATA, DataVolt, Noetra, DataSection, Firebird, Australia, and QumulusAI retain their source-defined states

Current-run update — 14 September 2026

Cutoff: 14 September 2026 UTC. This additive update uses only the accepted current-run evidence and leaves the prior dated records, structured report graph, and quantitative chart fences in place. ND means not disclosed or not verified in the reviewed source; it is not zero. Parent programs, site phases, utility assets, GPU systems, financing, and cloud rates are not added together.

New project, construction, and power records

Region location Stage source date IT capacity Facility program capacity Grid interconnection Generation PPA Company roles Investor financing Price cost basis Uncertainty evidence --- --- --- --- --- --- --- --- --- --- North America · Crusoe Lancium Childress AI campus, Texas, US Announced 15 Jul 2026; construction expected to begin Q3 2026 ND 1 GW grid-connected campus; source says it can support hundreds of thousands of AI accelerators at full scale 1 GW grid-connected campus ND Lancium owns land, secures power interconnect, and manages energy orchestration; Crusoe designs, builds, and operates Investor, lender, and project-finance allocation ND ND Campus grid figure, not IT load or installed-GPU stock; Crusoe and Lancium North America · Crusoe contracted US portfolio and Crusoe Cloud Contracted; site work and construction stages vary; source 9 Jun 2026 ND 4.9 GW contracted AI infrastructure across five US campuses and cloud capacity; development pipeline exceeds 40 GW ND ND Crusoe developer operator; hyperscaler, enterprise, and AI-native customer allocation varies ND ND Portfolio cloud scope is not a site total and is not additive to the Childress or Abilene rows; Crusoe portfolio North America · Crusoe Bergen Engines multi-site generation program Contract plus letter of intent; deliveries phased through 2027; source 3 Jun 2026 ND ND ND Approximately 750 MW: 438 MW contract plus 310 MW letter of intent Bergen Engines supplies natural-gas gensets; Piller Power Systems supplies SHIELD-X stabilization; Crusoe is recipient operator ND ND Generation capacity is not grid import, facility power, IT load, or GPU inventory; Bergen Engines agreement Europe · LUMI-AI, Kajaani, Finland Procurement contracted; deployment scheduled for H2 2027; source 31 Aug 2026 ND ND ND 100% renewable-energy basis; MW, tariff, and PPA strike price ND EuroHPC JU owns the system; CSC hosts; Bull delivers; six-country LUMI AI Factory consortium supports access Funding shared equally by EuroHPC JU and the LUMI AI Factory consortium €387.8M system contract including acquisition, delivery, installation, and maintenance; not GPU-only price or site capex Supercomputer procurement has no disclosed GPU count or site-MW value; AMD LUMI-AI

Current-run GPU inventory and specification addition

Site region Inventory or deployment evidence GPU system specification Company investor or financing Possible price basis Uncertainty evidence --- --- --- --- --- --- LUMI-AI · Kajaani, Finland GPU count ND; AMD-based system planned for H2 2027 deployment AMD Instinct MI430X GPUs with 6th Gen AMD EPYC 256-core processors; no GPU count or system topology disclosed EuroHPC JU owner; CSC host; Bull delivery; funding shared equally by EuroHPC JU and the LUMI AI Factory consortium €387.8M system contract for acquisition, delivery, installation, and maintenance; not a GPU quote Planned procurement is not installed inventory; AMD LUMI-AI

The other current-run GPU records remain state-qualified in the preceding inventory tables: operating Sakura infrastructure, acquired DataSection B200s, purchased QumulusAI B300s, planned Noetra Rubin GPUs, authorized Firebird GB300s, planned SINES Nscale Vera Rubin GPUs, targeted Sharon AI GPUs, and Crusoe GB200 NVL72 system capacity. System counts are not converted into discrete GPUs, and no GPU record is promoted from planned, targeted, authorized, or acquired to operating without direct evidence.

Regional and power comparison

Region Current-run signal Interpretation --- --- --- North America 1 GW Childress grid-connected campus; 4.9 GW contracted Crusoe portfolio; approximately 750 MW Crusoe generation program Three different boundaries—campus grid, portfolio contract, and generation—remain separate and non-additive Europe €387.8M LUMI-AI procurement; GPU count and site-MW basis ND A system procurement and funding signal, not a regional IT-capacity total Latin America, Middle East Africa, and Asia-Pacific Existing regional records retained from earlier dated updates; no new comparable current-run total is asserted here Reported IT, facility, grid, generation, MVA, GPU, and financing fields remain source-defined

No regional capacity or electricity total is calculated. The evidence supports comparisons of boundary and stage, not a roll-up of global or regional MW.

Price, methodology, and unresolved gaps

Subject Current signal Economic type and control Evidence --- --- --- --- Crusoe Cloud H100 US$3.90 GPU-hour; H200 US$4.29 GPU-hour; A100 US$2.00–2.30 GPU-hour; B200 and GB200 NVL72 contact sales Public hosted rental rates captured 14 Sep 2026; not OEM acquisition price, project capex, electricity tariff, or annual operating cost Crusoe Cloud pricing LUMI-AI €387.8M System contract including acquisition, delivery, installation, and maintenance; not GPU-only price or data-center construction price AMD LUMI-AI

The current delta keeps publisher date separate from fetch date, preserves official versus secondary source labels, and does not convert MVA or kV to MW. Missing observations remain ND ; they are not fabricated zeroes. Unresolved gaps are utility delivery and energization for Childress, site allocation within the 4.9 GW contracted portfolio, delivery and site mapping for the 750 MW generation program, the LUMI-AI GPU count and installed state, and any OEM GPU quote or electricity tariff. These gaps prevent a defensible global total or normalized $ MW or $ GPU comparison.

Current-run update — 15 September 2026

As of: 15 September 2026 UTC. This additive evidence screen uses only the accepted evidence review and its cited source passages. It contains 12 data-center records plus one distinct Nscale Start Campus Sines re-check, and 13 accelerator deployment configuration records plus the Sines addition. Earlier selected rows remain intact. ND means not disclosed or not verified in the reviewed evidence; it does not mean zero.

Method: preserve each source’s metric class and status; show only explicit GW-to-MW readability conversions; do not convert MVA or kV to MW; do not turn facility, campus, compute, generation, grid, allocation, system, rack, or cloud-offer figures into IT load, installed GPU stock, or a global total. Capital, debt, lease value, contract value, rental rates, and purchase proxies remain separate economic objects.

The report’s existing evidence-backed analytics remain in place, including source-labelled IT-capacity and hosted-rate comparisons. The new records below are not plotted when their units, scope, or status are not comparable; no missing observation is represented as a zero.

Data-center plans, construction, power, and ownership

Project site Status as-of Possible IT capacity Electricity capacity boundary Company operator Investor owner financing Sourced price cost basis date Unknowns evidence --- --- --- --- --- --- --- --- DC-01 · SB Energy OpenAI Stargate — Milam County, Texas, US Under construction in the latest SB Energy release; current energization ND ND; 1,200 MW data-center site lease, not labelled IT Generation MW, binding interconnection, and usable IT load ND; initial facilities expected from 2026 SB Energy, a SoftBank subsidiary, builds and operates for OpenAI OpenAI and SoftBank each invested US$500M in SB Energy; Ares preferred equity US$800M ; platform financing, not site ownership or capex No site capex or project price disclosed; the funding signals are not a site price GPU model count ND; DC-S1 · DC-S2 DC-02 · Meta Hyperion — Richland Parish, Louisiana, US Under construction and expanding as of 13 July 2026 ND; 5 GW compute capacity , not utility or IT MW Entergy agreement disclosed; contracted grid MW and usable load ND Meta provides construction and property management Blue Owl-managed funds own 80% and Meta 20% of the JV About US$27B JV development cost; about US$7B Blue Owl cash contribution; later more than US$50B regional investment has a different scope GPU count model and site load ND; DC-S3 · DC-S4 DC-03 · Microsoft Fairwater — Mount Pleasant, Wisconsin, US First facility fully operational 23 June 2026; adjacent second facility under construction for 2028 IT capacity ND Campus site MW, grid allocation, and usable IT MW ND Microsoft develops and operates Microsoft owner operator; no separate site investor disclosed US$3.3B initial pledge, additional US$4B for a similar second facility, and US$4.7B local hyperscale construction spend for 2024–2028; scopes differ Microsoft says “hundreds of thousands” of NVIDIA GPUs, but exact count model is ND; DC-S5 · DC-S6 DC-04 · xAI SpaceXAI Colossus — Memphis, Tennessee, US Operational claim on the current operator page accessed 15 September 2026 IT capacity ND Current site, grid, and IT MW ND; MLGW and TVA are named utility collaborators xAI SpaceXAI operator Project investors and project capex ND; corporate financing is not site ownership US$20B Series E is corporate financing, not project capex or GPU price; project price ND 200,000 H100s is a single-cluster claim; the 1M-GPU roadmap is excluded from current stock; DC-S7 · DC-S8 · GPU-E2 DC-05 · DataVolt NEOM HUMAIN Oxagon — Saudi Arabia Construction underway as of 31 August 2026 IT capacity ND; wider planned campus 1,500 MW , first phase 360 MW , and 100 MW under development are AI-ready facility figures Binding grid and generation MW ND; the reported figures are not confirmed site load DataVolt developer investor operator; HUMAIN, a PIF company, jointly develops the 100 MW tranche; NEOM Oxagon provides land and infrastructure support DataVolt investment role; ownership and project-finance allocation beyond the named roles ND US$5B initial phase investment; not full-campus capex or GPU price First 100 MW anticipated in 2028; GPU count model ND; DC-S9 · DC-S10 DC-06 · Stargate UAE Khazna-G42 — Abu Dhabi, UAE Construction well underway in the 16 October 2025 update toward planned 2026 delivery; newer completion confirmation not found IT capacity ND; 5 GW UAE–US campus, nested 1,000 MW cluster, and 200 MW first build are program campus signals Utility, interconnection, and usable load MW ND G42 leads; Khazna, a G42 company, develops the cluster; OpenAI, Oracle, NVIDIA, Cisco, and SoftBank are named partners Project ownership, capex, and financing allocation ND beyond named roles Project price capex ND The 5 GW, 1 GW, and 200 MW figures are nested, not additive; DC-S11 DC-07 · YTL Johor Green Data Center Park JDC1 — Kulai, Malaysia Current YTL Data Centers page says JDC1 is operational; page date ND, accessed 15 September 2026 IT load ND; current park claim up to 600 MW and JDC1 72 MW facility capacity National-grid access stated; binding feed MW, generation boundary, and usable IT load ND; older 500 MW solar claim is generation, not data-center load YTL Data Center Holdings YTL Power develops; YTL Communications announced ownership management of AI infrastructure YTL platform ownership context; separate investor allocation ND Project price ND H100 quantity and current deployment ND; DC-R1 · DC-R2 · DC-S12 · DC-S13 DC-08 · Yotta D2 — Greater Noida, India 20,736-card deployment announced in February April; later 30,000-card statement in June; 30 September 2026 completion target not independently verified by 15 September IT load ND; 60 MW site scalable to 250 MW , basis not separated Site figure is not normalized into IT, facility, or grid MW; interconnection and energization ND Yotta operator Yotta deployment investment more than US$2B ; NVIDIA engagement more than US$1B ; later supporting contracts about US$2.8B ; investor split ND These are program contract signals, not GPU or site purchase prices 20,736 and 30,000 B300 claims conflict; installed count ND; DC-S15 · DC-S16 · GPU-E3 DC-09 · Nebius AI Factory — Pajarila Lappeenranta, Finland Construction announced 31 March 2026; first customer capacity expected in 2027 IT load ND; up to 310 MW AI-factory capacity Grid IT split and live load ND Nebius operator; Polarnode implements; City of Lappeenranta leases the site City lease context; project investor allocation ND City describes a multi-billion-euro project; exact capex ND Blackwell and Vera Rubin named, but project GPU count and price ND; DC-S17 · DC-S18 DC-10 · AirTrunk MEL2 — Melbourne North-West, Australia Planned site acquired in the 24 December 2025 announcement; current construction or completion not confirmed IT load ND; 354+ MW hyperscale campus A diverse 66 kV feed is disclosed; feed MW, interconnection, and energization ND AirTrunk developer operator Site ownership investor split ND; Blackstone platform context is not assigned to this site in the accepted record A$5B+ direct MEL2 investment and A$7B+ total Melbourne investment; scopes differ and neither is GPU price GPU count model and site purchase price ND; DC-S19 · DC-S20 DC-11 · NEXTDC S4 — Horsley Park, Western Sydney, Australia In development on the current facility page accessed 15 September 2026 350 MW IT capacity Grid allocation, energization, and usable live capacity ND; secondary interconnection lead not accepted as binding evidence NEXTDC developer operator S4-specific investor owner allocation ND A$2.2B capital plan is portfolio-level; isolated S4 capex ND GPU count model and price ND; DC-S21 · DC-S22 · DC-C1 DC-12 · YTL Power JLand Sedenak Tech Park West — Johor, Malaysia Proposed joint development announced 19 August 2026; construction and delivery date ND IT load ND; up to 1,200 MW proposed data-center campus Utility, generation, and interconnection MW ND SIPP Power YTL Power and JLG Technopark JLand Group JLand is a Johor Corporation real-estate infrastructure platform; financing and ownership split ND Project price, financing, lease, and GPU price ND Proposed capacity is not construction, energization, or installed stock; DC-S14 DC-13 · Nscale Start Campus Sines Data Campus — Portugal Planned committed deployment starting late 2027; current installed and energized status ND IT load ND; fully permitted 1,200 MW campus and separate 200 MW second building Permitted campus capacity is not usable load or grid load Nscale deploys for Microsoft; Start Campus builds operates Davidson Kempner identified as Start Campus shareholder; Microsoft customer role; other project financing allocation ND €230M shared-infrastructure investment and €465M second-building investment; separate scopes, not GPU price More than 66,000 Rubin GPUs planned and more than 12,600 Blackwell Ultra GPUs in the first building; no delivery proof; DC-R3

GPU inventories, specifications, company roles, and price basis

Model deployment Disclosed inventory state Specifications Company vendor Investor owner financing Possible price basis date Unknowns evidence --- --- --- --- --- --- --- GPU-01 · xAI Colossus — Memphis, US 200,000 NVIDIA H100 GPUs in one interconnected cluster; separate more than 1M H100-equivalents across Colossus I II is not additive H100 product anchor: fourth-generation Tensor Cores, FP8 Transformer Engine, and 900 GB s GPU-to-GPU interconnect; exact project variant ND NVIDIA xAI SpaceXAI operator; NVIDIA vendor Project investors and GPU procurement financing ND GPU purchase price ND; US$20B Series E is corporate financing, not hardware price Operator claim, not independent stocktake; GPU-E1 · GPU-E2 GPU-02 · Yotta NM1 — Navi Mumbai, India Operator reports 8,192 H100s and 1,024 L40s live ; derived subtotal 9,216 , not independently audited H100 and L40 models disclosed; detailed L40 specification not retained in the accepted evidence. NM1 separately lists 52 MW IT power , 7,000 racks, and PUE 1.5 as facility context NM1 Yotta Data Services Hiranandani Group context; NVIDIA partner Yotta operator site context; investor split and procurement financing ND Purchase price and rental basis ND; no unit-price date verified Live claim is operator-reported and facility power is not GPU draw; GPU-E3 GPU-03 · CoreWeave MERLIN Edged — Barcelona, Spain Site announcement says it will host 10,224 NVIDIA H200 GPUs ; current 2026 availability ND H200: 141 GB HBM3e at 4.8 TB s ; NVIDIA labels specifications preliminary H200 CoreWeave operator; MERLIN Properties and Edged Energy facility partners; NVIDIA vendor Site lease procurement and investor roles ND GPU purchase price and rental basis ND; no current unit-price date verified Planned hosted quantity is not a current inventory stocktake; GPU-E5 GPU-04 · DataSection ASRock Rack — Bangkok, Thailand 4,696 B200 GPUs across 587 servers purchased; delivery, commissioning, and availability ND DGX B200 reference: 8 Blackwell GPUs, 1,440 GB total GPU memory, 64 TB s HBM3e bandwidth, about 14.3 kW maximum system power; system reference is not proof of deployment configuration DGX B200 DataSection operator buyer; ASRock Rack supplier; NVIDIA vendor; unnamed US customer Investor, lender, and customer-finance allocation ND Reported US$257.4M package estimate 5 June 2026 , potentially including servers equipment; not per-card price Purchased is not delivered or installed; official filing detail is a search-captured record; GPU-E6 · GPU-E6a GPU-05 · Mistral AI Eclairion — Bruyères-le-Châtel, France DCD reports access to 13,800 GB300 GPUs ; earlier 18,000 GB200 disclosure is not reconciled or additive GB300 NVL72 reference: liquid-cooled rack with 72 Blackwell Ultra GPUs and 36 Grace CPUs , plus 800 Gb s per-GPU connectivity GB300 Mistral customer developer; Eclairion owner operator; NVIDIA vendor Seven-bank lender group; US$830M debt is infrastructure GPU financing, not ownership or card price GPU purchase price ND; US$830M debt dated 30 March 2026 is financing, not a unit quote No current commissioning proof after the end-June target; quantity is secondary-reported; GPU-E7 GPU-06 · Yotta D2 — Greater Noida, India February plan: 20,736 liquid-cooled Blackwell Ultra B300 GPUs; April describes an additional 20,736-card tranche; June says 30,000 B300s ; installed count ND Liquid cooling, 800 Gbps Quantum-X800 InfiniBand , and more than 40 PB parallel storage stated; GPU VRAM ND Yotta operator; NVIDIA technology customer partner Yotta deployment investment more than US$2B ; NVIDIA engagement more than US$1B ; supporting contract framework about US$2.8B ; not GPU price No per-GPU purchase price or rental basis; dated program values are 18 February, 30 April, and 23 June 2026 Conflicting dated commitments are preserved without addition; GPU-E8 · GPU-E3 GPU-07 · Nscale Aker OpenAI Narvik — Norway 100,000 NVIDIA GPUs targeted by end-2026; separate more than 30,000 Rubin GPUs for Microsoft in 2027; not additive Vera Rubin NVL72 reference: 72 Rubin GPUs and 36 Vera CPUs , ConnectX-9, BlueField-4, and NVLink 6 Rubin Nscale Aker OpenAI project roles; Microsoft customer deployment role; NVIDIA vendor 2025 plan described Nscale Aker 50 50 JV; 2026 update says solely managed by Nscale; initial commitment about US$1B with up to US$250M equity and later US$790M financing Financing and commitments are project economics, not GPU price; dates 31 July 2025 and 11 May 2026 Installed count, legal ownership, and tranche overlap unresolved; GPU-E9 · GPU-E10 · GPU-E19 GPU-08 · HUMAIN NVIDIA AI factories — Saudi Arabia First phase described as an 18,000 GB300 Grace Blackwell AI supercomputer plus several hundred thousand GPUs over five years; unit type scope ambiguous GB300 NVL72 reference is 72 Blackwell Ultra GPUs and 36 Grace CPUs, liquid cooled GB300 HUMAIN, a PIF subsidiary; NVIDIA, SDAIA, and Aramco Digital named Site financing and ownership allocation ND beyond the named program roles GPU price ND; program capacity up to 500 MW over five years is not price or installed stock Do not count 18,000 as individual GPUs until the unit is clarified; GPU-E11 GPU-09 · Noetra NVIDIA national AI infrastructure — Japan 27,500 Rubin GPUs plus 13,750 Vera CPUs planned; delivery and energization ND Vera Rubin NVL72 reference: 72 Rubin GPUs, 36 Vera CPUs, ConnectX-9, BlueField-4, and NVLink 6 Rubin Noetra establishes; NVIDIA and METI-supported FRONTia context Other financing and investor roles ND GPU price ND; source’s 140 MW data-center capacity is not a hardware price or GPU power total Planned program is not delivered or installed stock; GPU-E12 GPU-10 · AWS Project Rainier Anthropic — US AWS reports an operational cluster with nearly 500,000 Trainium2 chips ; this is accelerator-chip inventory, not GPU inventory AWS Trn2 reference uses 16 Trainium2 chips per instance and 64 per UltraServer, with up to 6 TB HBM and 185 TB s for UltraServer Trn2 AWS Amazon operator; Anthropic customer US$5B current investment and up to US$20B more are corporate investment signals, not site or chip price Chip price and site price ND; separate Anthropic commitment up to 5 GW is not Rainier site load Keep Trainium2 separate from NVIDIA AMD GPU inventory; GPU-E13 · GPU-E14 GPU-11 · Israel national AI supercomputer Nebius — Israel Infrastructure reported live and operational; 1,000 NVIDIA B200 accelerators allocated over coming years, 70% high-tech and 30% research B200 DGX B200 reference: 8 GPUs, 1,440 GB total GPU memory, 64 TB s HBM3e, about 14.3 kW system maximum DGX B200 Israel Innovation Authority selected Nebius under Telem; NVIDIA vendor National program access; investor and project-finance allocation ND Access is discounted versus market; tariff and GPU purchase price ND; source dated 18 January 2026 Allocation ceiling is not installed count; GPU-E15 GPU-12 · IBM Cloud AMD Zyphra — US Large dedicated MI300X cluster announced; count ND; current availability not re-confirmed MI300X: 192 GB HBM3 , about 5.325 TB s bandwidth, and 750 W OAM power in signed indexed official evidence; detailed values provisional AMD MI300X IBM cloud operator; AMD vendor; Zyphra customer Project capex, investor, and financing allocation ND GPU purchase and cloud price ND; announcement dated 1 October 2025 Count, site power, and current deployment state ND; GPU-E16 GPU-13 · Google Cloud TPU7x Ironwood Supported configurations include 256 and 9,216 chips ; configuration ceiling customer option, not global installed fleet Per chip: 2,307 BF16 TFLOPs , 4,614 FP8 TFLOPs , 192 GiB HBM , 7,380 GB s HBM bandwidth, 1,200 GB s bidirectional ICI; pod power ND TPU7x Google designs and operates Project investor and purchase-finance allocation ND Public purchase price and rental basis ND; current page accessed 15 September 2026 TPU7x is an accelerator, not NVIDIA AMD GPU inventory; GPU-S7 GPU-14 · Nscale Start Campus Sines — Portugal 66,000+ Rubin GPUs planned for Microsoft from late 2027; first building has more than 12,600 Blackwell Ultra GPUs ; no delivery proof Rubin NVL72 reference: 72 Rubin GPUs 36 Vera CPUs; Blackwell Ultra rack reference: 72 GPUs 36 Grace CPUs Rubin · GB300 Nscale and Start Campus; Microsoft customer; NVIDIA vendor Davidson Kempner shareholder context; project-funding allocation ND €230M shared infrastructure and €465M second-building investment 5 May 2026 ; not GPU price Separate site tranche claims; do not add to Narvik or the Europe-wide rollout; GPU-E18

Current price checkpoints

Provider scope Observed value Price basis and date Boundary --- --- --- --- RunPod Pods H100 PCIe US$2.89 , H100 SXM US$3.49 , H200 US$4.59 , B200 US$6.79 , B300 US$7.89 , L40 US$0.82 , L40S US$1.09 per GPU-hour Dynamic hosted rental list rates; page updated 13 September 2026; region, tier, tax, storage, and networking can change effective price Not OEM purchase price, project capex, electricity tariff, or installed inventory; PRICE-1 RunPod H100 PCIe About US$25,000–30,000 per card Provider purchase comparison; page updated 27 August 2026 Secondary provider proxy, not NVIDIA MSRP or a signed project quote; excludes server, network, cooling, software, support, tax, and availability; PRICE-2 Lambda self-serve instances H100 SXM US$3.99 and B200 SXM6 US$6.69 per GPU-hour Current catalog accessed 15 September 2026 Hosted rental list rates; sales tax VAT GST may apply; not hardware price; PRICE-3 CoreWeave HGX instances H100 US$49.24 , H200 US$50.44 , HGX B200 US$68.80 per 8-GPU instance-hour Current provider table accessed 15 September 2026 Instance-hour prices, not per-card prices, procurement quotes, site capex, or electricity tariffs; PRICE-4

No accepted source supplies a clean OEM purchase price or executed per-card quote for H100, H200, B200 B300 GB300, Rubin, MI300X MI355X, Trainium2, TPU7x, or Gaudi 3. Rental rates and the H100 provider comparison are useful market checkpoints only.

Evidence controls and unresolved risks

Risk unresolved field Treatment in this report Residual uncertainty --- --- --- xAI’s 200,000-H100 single-cluster claim versus its more-than-1M H100-equivalent roadmap Keep current single-cluster and broader roadmap as separate scopes; exclude the roadmap from current stock Self-reported inventory; no independent stocktake Yotta D2’s 20,736, additional 20,736, and 30,000 B300 statements Preserve dated conflict; do not add; installed count remains ND Current delivery and commissioning evidence is required after the stated target date Nscale Narvik, Sines, and Europe-wide Rubin announcements Keep exact sites and envelopes separate; do not add counts; retain the ownership-language change Legal ownership, delivery, and overlap across tranches remain unresolved Mistral Eclairion’s 13,800 GB300 access report versus prior 18,000 GB200 plan Use later reported access as non-live evidence and retain the prior conflict without addition Secondary-only quantity and no current commissioning proof HUMAIN NVIDIA’s 18,000 GB300 “supercomputer” wording Do not treat 18,000 as individual-GPU inventory until the unit and tranche are clarified System card scope remains ambiguous YTL’s current up-to-600 MW park and 72 MW JDC1 versus older 500 MW park solar claims Keep park, facility, and solar-generation metrics separate 72 MW is not proven IT or grid MW NEXTDC S4’s 350 MW IT claim versus an unaccepted secondary interconnection lead Use 350 MW IT only; leave grid interconnection MW ND Primary approval or energization evidence is absent DataSection’s official filing search capture and DCD package estimate Retain purchased count with quality qualification; keep US$257.4M as a package estimate, not per-card price Direct filing extraction and package scope remain unresolved Mixed facility, IT, compute, AI-ready, campus, grid, generation, and MVA measures Preserve source units and boundaries; no MVA-to-MW or MW-to-GPU conversion No field-level scope bridge supports a global or regional roll-up Missing OEM prices and electricity tariffs Keep ND and show only dated rental, provider, financing, and cost proxies with their original basis Procurement requires dated supplier quotes, contracts, and site-specific tariffs Global coverage Present a bounded source-linked landscape, not a census No defensible global capacity, electricity, installed-GPU, capital, or comparable-price total is available in this evidence set

Current-run update — 16 September 2026

As of: 16 September 2026 UTC. This additive evidence screen uses only the accepted current-run register. It contains 24 data-center project site records, 24 accelerator or specification records, one separate electricity PPA record, and three rental benchmarks. Earlier selected rows and the existing analytical charts remain intact. ND means not disclosed or not verified in the reviewed evidence; it does not mean zero.

The current screen reinforces a non-additive picture: operating claims and planned targets coexist with construction, procurement, allocation, and product-topology records. IT load, facility or compute capacity, grid interconnection, generation PPA, MVA, GPU cards, accelerator chips, servers, racks, pods, and economic objects remain separate. No global or regional total is asserted.

Data-center plans, construction, operations, and electricity boundary

ID Geography project Lifecycle and status Possible announced capacity IT load Facility compute electricity Grid interconnection Generation PPA Company operator developer Investor backer Price cost basis Source date Direct evidence Verification qualification --- --- --- --- --- --- --- --- --- --- --- --- --- --- DC-NA-01 North America · Richland Parish, Louisiana, US · Meta Hyperion Under construction expanding; 2026-07-13 update; current completion ND 5 GW compute capacity ND 5 GW is source-defined compute capacity, not facility or electricity MW ND ND in reviewed project source Meta; Entergy Louisiana, DPR, Turner, Mortenson named in project context Meta reports more than US$50B regional investment; site capex split ND More than US$50B regional investment; US$2.65B customer energy savings over 20 years; neither is site capex, tariff, or GPU price 2026-07-13 Meta Official source; live load, commissioning date, GPU count, utility allocation, and financing split ND; compute capacity is not relabelled. DC-NA-02 North America · Mount Pleasant, Wisconsin, US · Microsoft Fairwater First facility fully operational; adjacent second facility under construction and scheduled for 2028 ND ND Campus site MW and electricity boundary ND Grid allocation ND Generation PPA ND Microsoft; Village of Mount Pleasant and local contractors partners Microsoft owner operator; separate investor allocation ND US$4.7B local hyperscale construction spend for 2024–2028; not facility-only capex or GPU price 2026-06-23 Microsoft Official source; operating status is reported, but site MW, grid, generation, GPU inventory, and facility financing are ND. DC-NA-03 North America · Memphis, Tennessee, US · xAI SpaceXAI Colossus Operating claim; current page read 2026-09-16; page date ND ND ND Site electricity and facility boundary ND MLGW TVA are referenced in continuity material; binding allocation ND ND xAI SpaceXAI Project investors, capex, and site financing ND ND ND page read 2026-09-16 xAI Colossus Official operator page; 200,000 H100 claim is not an independent stocktake; site MW and commissioning proof beyond the page are ND. DC-NA-04 North America · Abilene, Texas, US · Crusoe Microsoft AI-factory campus Site preparation underway; first building expected energized mid-2027 900 MW dedicated site; approximately 2.1 GW projected full Abilene site ND 900 MW onsite power plant; site power figure, not IT load ND Onsite generation is reported; nameplate grid import boundary ND Crusoe; Microsoft customer Project finance and investor allocation ND ND 2026-03-27 Crusoe Official source; 900 MW is not relabelled as IT load or grid import; 900 MW and 2.1 GW are nested. DC-NA-05 North America · Childress, Texas, US · Crusoe Lancium AI campus Announced; construction expected Q3 2026 1 GW AI data-center grid-connected campus at full scale ND Solar and storage referenced; generation nameplate MW ND 1 GW grid-connected capacity Solar power and energy storage referenced; exact generation MW ND Crusoe designs, builds, and operates; Lancium owns land, secures interconnect, and manages energy orchestration Crusoe Lancium direct local infrastructure investment; project financing allocation ND; unnamed hyperscaler ND 2026-07-15 Crusoe Lancium Official source; 1 GW is a grid-connected campus figure, not IT load or installed accelerators; construction is expected, not confirmed. DC-NA-06 North America · Harwood, North Dakota, US · Applied Digital Polaris Forge 2 Financed development construction scope; offering close conditional in source ND 200 MW critical IT load Facility compute MW ND Grid capacity and interconnection ND Generation PPA ND Applied Digital APLD ComputeCo 2 Senior secured notes; Applied Digital completion guarantee; qualified institutional buyers US$2.15B, 6.750% senior secured notes due 2031, issue price 98%; financing for 200 MW critical IT load, not GPU price 2026-03-04 Applied Digital Official investor-relations source; financing does not establish normalized construction cost, utility tariff, current completion, or GPU price. DC-LAT-01 Latin America · Querétaro, Mexico · ODATA DC QR03 First building available; first energy phase energized; full five-building build staged 300 MW full build; 72 MW first building 300 MW full build; 72 MW first building are source-labelled IT figures Five-building campus 400 kV switching substation; 6.2 miles of transmission; two new substations Ammper renewable-energy supply partner; exact PPA MW price ND ODATA, an Aligned Data Centers company ODATA Aligned; site-specific investor allocation ND; three anchor hyperscale clients More than US$3B launch investment; later US$3.3B campus investment; not GPU price 2025-04-30; energization source 2025-02-27 ODATA capacity · ODATA energization Official company sources; 200 MW energized is an energy-phase figure, not IT load; installed GPU count, tariff, and site finance split ND. DC-EU-01 Europe · Sines, Portugal · Start Campus SINES Data Campus SIN01 live; SIN02 next planned; later phases not treated as energized 1.2 GW total campus; 200 MW next planned facility 37.5 MW live first facility; 200 MW next facility is planned, not live 1.2 GW total campus capacity 1.2 GW secured IT grid capacity; current draw ND 100% renewable-energy claim; PPA MW and price ND Start Campus develops, builds, and operates; REN transmission role; Siemens Energy supplier context Investor allocation ND in current source ND in current source ND page read 2026-09-16 Start Campus Current official page; full-load energization, exact grid draw, and site GPU inventory ND; campus grid figure is not live IT load. DC-EU-02 Europe · Sines, Portugal · Nscale Start Campus Microsoft deployment Committed deployment starting late 2027; first-building current installed state ND 200 MW second building; campus fully permitted for 1.2 GW ND Second 200 MW building; facility IT split ND Permitted campus figure only; usable grid draw ND 100% renewable-campus claim; PPA MW and price ND Nscale deployment customer infrastructure; Start Campus builds and operates Microsoft customer; Davidson Kempner named Start Campus shareholder; financing draw ND €230M shared infrastructure plus €465M second 200 MW building; separate investment objects, not GPU prices 2026-05-05 Nscale Start Campus Official source; does not establish current energization, delivered GPU count, IT split, or financing draw; nested with SINES campus. DC-EU-03 Europe · Lappeenranta, Finland · Nebius AI factory Construction announced; first customer capacity expected in 2027 Up to 310 MW AI-factory capacity ND 310 MW is source-defined AI-factory capacity; boundary not otherwise stated ND Clean-energy design stated; exact generation PPA MW ND Nebius; Polarnode implements; city local industrial-site partners City lease context; exact project investor allocation ND Multi-billion-euro project context in city materials; exact site capex ND 2026-03-31 Nebius Official company source; 310 MW is not relabelled as IT load or grid draw; construction progress and energization ND. DC-EU-04 Europe · Seinäjoki, Finland · Pure DC SJK01 Phase 1 launched; first data-hall substation live; expansion conditional on permissions and contracts 110 MW phase 1; 550+ MW full-campus IT pathway 110 MW phase 1; 550+ MW full-campus IT pathway 550+ MW AI campus at full pathway More than 700 MVA renewable power access; not converted to MW Renewable power access; PPA strike price ND Pure Data Centres Group SDC Ventures; Seinäjoki municipality; lender equity allocation ND More than €1.5B phase 1; more than €7.5B full campus; not GPU prices 2026-07-14 Pure DC Official source; MVA is not converted to MW; live IT load, grid import, installed GPUs, and financing draw ND. DC-EU-05 Europe · Bruyères-le-Châtel, France · Mistral AI Eclairion Planned; target date passed without current commissioning evidence ND ND 44 MW powered capacity; not labelled IT ND ND Mistral AI customer developer; Eclairion owner operator Seven-bank lender consortium including Bpifrance, BNP Paribas, Crédit Agricole CIB, HSBC, La Banque Postale, MUFG, Natixis CIB US$830M debt financing; infrastructure GPU financing, not per-card price 2026-03-30 DCD report Secondary source; no current commissioning or delivery proof; powered capacity is not IT load. DC-EU-06 Europe · Kvandal near Narvik, Norway · Nscale Aker OpenAI Stargate Norway Planned; preparatory mobilization announced; current energization ND 230 MW planned; ambition for an additional 290 MW ND 230 MW source-defined planned capacity ND Powered entirely by renewable energy in source claim; exact generation PPA MW ND Nscale designs builds and after JV roll-up solely manages; Aker planned 50 50 JV partner; OpenAI Microsoft customer roles Approximately US$1B initial 20 MW phase including up to US$250M equity; later financing context separate Approximately US$1B initial phase; up to US$250M equity; project commitment, not GPU price 2025-07-31; expansion source 2026-04-14 Nscale Norway · Nscale Microsoft Official sources; 100,000 target, 230 MW plan, 290 MW ambition, and 30,000+ Rubin tranche have no verified current delivery map; values are nested. DC-MEA-01 Middle East · Oxagon, NEOM, Saudi Arabia · DataVolt HUMAIN AI data center Construction underway; first 100 MW anticipated in 2028 100 MW under development within 360 MW first phase; wider 1.5 GW campus ND 100 360 1,500 MW are source-defined project factory scopes ND Renewable integration planned; exact generation import MW ND DataVolt developer, financier, operator; HUMAIN PIF company and joint-development partner; NEOM Oxagon host DataVolt initial US$5B phase investment; ownership financing allocation beyond named roles ND Initial US$5B phase investment; not tariff, unit construction cost, or GPU price 2026-08-31 NEOM DataVolt HUMAIN Official source; project scopes are not verified IT load or grid-import totals; installed GPUs and current energization ND. DC-MEA-02 Middle East · Saudi Arabia · HUMAIN NVIDIA sovereign AI factories Announced multi-year program; site delivery current operation ND Up to 500 MW over five years ND 500 MW program figure; site facility boundary ND ND ND HUMAIN PIF subsidiary; NVIDIA technology platform partner; SDAIA and Aramco Digital named HUMAIN PIF investment stated; amount and site allocation ND ND 2025-05-13 NVIDIA Saudi AI factories Official vendor source; 500 MW is not assigned to one site or IT load; 18,000 figure is not treated as unconditional individual-GPU stock. DC-MEA-03 Middle East Europe fringe · Hrazdan, Armenia · Firebird AI factory Operational launch announced; expansion roadmap future 300 MW AI infrastructure targeted by end-2027 ND 300 MW source-defined AI-infrastructure capacity ND ND Firebird; NVIDIA intends to invest; CoreWeave earlier investment; government strategic partners NVIDIA intends to invest; CoreWeave invested earlier; amounts ND ND 2026-08-08 Firebird Official source; launch is source-reported; current installed GPUs, IT load, grid allocation, and financing draw ND; broader 2 GW pipeline excluded. DC-AF-01 Africa · Cape Town, South Africa · Cassava AI Factory Near-live first deployment in place; full rollout not independently re-confirmed ND ND Facility and electricity boundary ND ND ND Cassava Technologies; NVIDIA Cloud Partner; CSIR and Rafay ecosystem Company-level investor details and site capex ND ND 2026-03-18; secondary update 2026-05-04 Cassava · ITWeb update Official announcement plus secondary status update; first deployment is reported, but commissioning record, live GPU count, MW, grid, and price are ND. DC-AF-02 Africa · Johannesburg, South Africa · Cassava planned AI Factory Planned aspiration; construction and commissioning not evidenced ND 20 MW planned installation; boundary otherwise unspecified ND ND ND Cassava Technologies; NVIDIA platform context ND ND 2026-05-04 ITWeb Secondary source; 20 MW is a planned installation statement, not energized IT load or grid capacity; GPU allocation and capex ND. DC-AP-01 Asia-Pacific · Kulai, Johor, Malaysia · YTL Johor Green Data Center Park JDC1 JDC1 operational; park expansion remains a development envelope Park up to 600 MW data-center capacity; JDC1 72 MW capacity ND Park facility boundary; live IT utilization ND National-grid access stated; binding feed MW ND Renewable solar-powered facility claim; generation nameplate MW ND YTL Data Center Holdings YTL Power; NVIDIA technology context Backer and project-finance allocation ND in current page ND ND page read 2026-09-16 YTL park · YTL JDC1 Current official pages; park, JDC1 facility, solar generation, grid access, and live IT GPU utilization remain separate. DC-AP-02 Asia-Pacific · Sedenak Tech Park West, Johor, Malaysia · YTL Power JLand campus Proposed joint development; construction and delivery date ND Up to 1.2 GW planned campus capacity ND Planned data-center campus; facility electricity boundary ND ND ND SIPP Power YTL Power; JLG Technopark JLand Group master developer Johor Corporation ecosystem; financing and ownership split ND ND 2026-08-19 YTL Power JLand Official source; 1.2 GW is a planned campus envelope, not energized load or confirmed grid allocation; GPU plan and capex ND. DC-AP-03 Asia-Pacific · Bangkok, Thailand · DataSection AI data center Under development; phased operations expected July–August 2026 in source; current operation not verified ND ND ND ND 10 MW planned power supply in prior company plan; not relabelled as IT or grid capacity DataSection; ASRock Rack supplier; unnamed US technology-company customer Customer advances, US$180M borrowings, and a share acquisition reported; exact site allocation ND US$257.4M estimated GPU purchase agreement package; may include servers equipment, not per-card MSRP 2026-06-05 DCD report Secondary source; no current delivery, energization, IT-load, grid, or site-level financing evidence in the accepted pass. DC-AP-04 Asia-Pacific · Horsley Park, Western Sydney, Australia · NEXTDC S4 In development ND 350 MW IT capacity 124,000 m² technical space; facility power ND Grid allocation and interconnection ND ND NEXTDC developer operator S4-specific financing ND; portfolio capital not assigned ND for S4 ND page read 2026-09-16 NEXTDC S4 Current official facility page; construction completion, energization, usable load, GPU inventory, and site capex ND. DC-AP-05 Asia-Pacific · Melbourne North-West, Australia · AirTrunk MEL2 Current location page; construction energization status ND 354+ MW hyperscale campus ND 66 kV diverse high-voltage feed; feed MW ND Feed voltage is disclosed; interconnection and usable draw ND ND AirTrunk developer operator Backers and site ownership split ND in current page ND ND page read 2026-09-16 AirTrunk MEL2 Current official location page; capacity and voltage do not establish current construction, energization, IT-load split, GPU inventory, or investment. DC-AP-06 Asia-Pacific · Sejong, South Korea · NAVER GAK Sejong AI-factory alliance Existing campus launchpad; expansion planned 55 MW operations in H1 2027; 200 MW by 2028; broader gigawatt alliance context ND 55 200 MW are roadmap capacity statements; facility electricity boundary ND ND ND NAVER; NVIDIA partner Alliance capital collaboration; site-specific equity lender allocation ND ND 2026-06-08 NAVER Official source; roadmap figures are not current live IT grid draw; broader 1 GW alliance capacity is not assigned to GAK Sejong.

GPU and accelerator inventory, specification, company, backing, and price basis

ID Geography deployment Model and disclosed count State Specification Company vendor Investor backer Price cost basis Source date Direct evidence Verification qualification --- --- --- --- --- --- --- --- --- --- --- ACC-01 North America · Memphis, US · xAI SpaceXAI Colossus NVIDIA H100 · 200,000 Operating claim; single interconnected cluster H100 reference: fourth-generation Tensor Cores, FP8 Transformer Engine, 900 GB s GPU-to-GPU interconnect; project SKU form factor ND xAI SpaceXAI; NVIDIA ND; separate 1M-GPU roadmap is future context ND; no OEM or project purchase price ND page read 2026-09-16 xAI Colossus Official operator claim, not an independent stocktake; roadmap excluded from current inventory. ACC-02 Asia-Pacific · Navi Mumbai, India · Yotta NM1 NVIDIA H100 · 8,192 Live claim; operator-reported H100 vendor reference separate; project form factor ND Yotta Data Services; NVIDIA Hiranandani Group company context; site investor allocation ND ND 2026-06-23 Yotta NM1 claim Official company release; live count not independently audited and not combined with Yotta D2 or future roadmap. ACC-03 Asia-Pacific · Navi Mumbai, India · Yotta NM1 NVIDIA L40 · 1,024 Live claim; operator-reported L40 specification not assigned in this pass Yotta Data Services; NVIDIA Hiranandani Group company context; site investor allocation ND ND 2026-06-23 Yotta NM1 claim Official company release; distinct model count, no monetary value inferred, and no independent audit. ACC-04 Europe · Barcelona, Spain · CoreWeave MERLIN Edged NVIDIA H200 · 10,224 Host commitment planned in 2025 source H200 reference: 141 GB HBM3e at 4.8 TB s; project form factor ND; same source says 15 MW site CoreWeave; MERLIN Edged; NVIDIA CoreWeave’s US$2.2B initial Europe investment is not Barcelona-only ND 2025-05-13 CoreWeave Barcelona Official investor-relations source; site will host claim, not a current delivery or availability count; 15 MW is not GPU draw. ACC-05 Asia-Pacific · Bangkok, Thailand · DataSection NVIDIA B200 · 4,696 across 587 servers Purchased deployment planned B200 SXM6 catalog reference 180 GB VRAM; project SKU form factor not independently confirmed DataSection; ASRock Rack; NVIDIA Unnamed US customer; customer advances and US$180M borrowings reported; financing allocation ND US$257.4M estimated purchase agreement package; not per-card MSRP 2026-06-05 DCD DataSection Secondary source; purchased is not delivered or commissioned, and 587 is a server count rather than GPU count. ACC-06 Europe · Bruyères-le-Châtel, France · Mistral AI Eclairion NVIDIA GB300 · 13,800 Planned access financed GB300 NVL72 reference: 72 Blackwell Ultra GPUs and 36 Grace CPUs, liquid cooled, 800 Gb s per GPU; not site inventory Mistral AI; Eclairion; NVIDIA Seven-bank consortium; US$830M debt financing US$830M infrastructure GPU financing; not unit price 2026-03-30 DCD Mistral Eclairion Secondary source; installation current state unverified; earlier 18,000 GB200 disclosure is conflicting and not added. ACC-07 Asia-Pacific · Greater Noida, India · Yotta D2 NVIDIA Blackwell Ultra HGX B300 · 20,736 announced 2026-02-18; 30,000 stated 2026-06-23 Planned committed; installed count ND Liquid-cooled Blackwell Ultra; project memory and TDP ND; 800 Gbps Quantum-X800 InfiniBand and more than 40 PB storage stated Yotta Data Services; NVIDIA Yotta investment above US$2B; separate NVIDIA engagement above US$1B Program deployment investment signals; not GPU price 2026-02-18; 2026-06-23 Yotta 20,736 · Yotta 30,000 Official company claims conflict by date; counts are preserved without addition and no installed stock is asserted. ACC-08 Europe · Narvik, Norway · Nscale Aker OpenAI Stargate NVIDIA GPUs; exact SKU mix ND · 100,000 Planned target by end-2026 Project-specific specification ND Nscale; planned Nscale Aker JV; OpenAI offtaker; NVIDIA vendor Approximately US$1B initial 20 MW phase including up to US$250M equity Project commitment, not GPU price 2025-07-31 Nscale Norway Official source; target is not delivery proof and is not added to Rubin or Europe-wide rollout claims. ACC-09 Europe · Narvik, Norway · Nscale campus Microsoft agreement NVIDIA Vera Rubin · more than 30,000 Announced deployment in 2027 Vera Rubin NVL72 reference: 72 Rubin GPUs and 36 Vera CPUs, ConnectX-9, BlueField-4, NVLink 6; site mix ND Nscale; Microsoft customer; NVIDIA vendor Narvik project-finance context in ACC-08; tranche price ND ND; not inferred from project finance 2026-04-14 Nscale Microsoft Norway Official source; tranche within the Narvik campus target, not an incremental global total. ACC-10 Europe · Sines, Portugal · Nscale Start Campus Microsoft NVIDIA Rubin · more than 66,000 Planned expansion starting late 2027 Vera Rubin NVL72 topology named; site-specific memory and rack count ND Nscale; Start Campus; Microsoft customer; NVIDIA vendor Nscale €230M shared infrastructure plus €465M second-building investment Investment signals, not GPU price 2026-05-05 Nscale Sines Official source; not added to ACC-11, Narvik, or a wider rollout; delivery proof absent. ACC-11 Europe · Sines, Portugal · first Start Campus building NVIDIA Blackwell Ultra · more than 12,600 Deployed claim; current operation ND Blackwell Ultra; project-specific topology ND Nscale; Start Campus; Microsoft customer; NVIDIA vendor See Sines project financing; no GPU unit price ND 2026-05-05 Nscale Sines Official source; foundation for the expansion claim, not automatically additive to ACC-10 or campus capacity. ACC-12 Middle East · Saudi Arabia · HUMAIN NVIDIA AI factories NVIDIA GB300 Grace Blackwell · 18,000 First-phase deployment announcement Source calls it an AI supercomputer; GB300 NVL72 is a separate 72-GPU rack reference; cards versus systems unresolved HUMAIN NVIDIA; SDAIA and Aramco Digital partners HUMAIN is a PIF subsidiary; site financing ND ND 2025-05-13 NVIDIA Saudi AI factories Official vendor source; 18,000 is not treated as unconditional individual-GPU stock until unit scope is clarified. ACC-13 Middle East Europe fringe · Hrazdan, Armenia · Firebird AI factory NVIDIA Vera Rubin and Blackwell · more than 70,000 Planned expansion target by end-2027 Model families named; per-model memory and topology ND Firebird; NVIDIA vendor; CoreWeave context NVIDIA intends to invest; CoreWeave invested earlier; amounts ND ND 2026-08-08 Firebird Official source; target is paired with 300 MW infrastructure and is not launch-day installed stock. ACC-14 Asia-Pacific · Japan · Noetra NVIDIA national AI infrastructure NVIDIA Rubin · 27,500 Planned; delivery and energization ND Vera Rubin NVL72 racks; paired 13,750 Vera CPUs are not included in GPU count Noetra; NVIDIA METI FRONTia support context; project finance allocation ND ND 2026-07-16 NVIDIA Japan Official vendor source; planned count is not delivered or installed stock; CPU count remains separate. ACC-15 North America · AWS Project Rainier Anthropic AWS Trainium2 · nearly half a million chips Fully operational in AWS source Trn2 reference: 16 chips per instance and 64 per UltraServer; up to 6 TB HBM and 185 TB s per UltraServer AWS; Anthropic workloads AWS Anthropic strategic relationship; project capex allocation ND ND 2025-10-29 Amazon Project Rainier Official AWS source; custom accelerator-chip inventory, not NVIDIA AMD GPU inventory; separate 1M-chip target excluded. ACC-16 Middle East · Israel national supercomputer Nebius NVIDIA B200 · 1,000 Program allocation over coming years; infrastructure live and operational B200 reference separate; program hardware detail ND Nebius selected provider; Israel Innovation Authority program; NVIDIA vendor National Program for AI R&D Infrastructure Telem Program Discounted access versus market; numeric tariff ND 2026-01-18 Israel Innovation Authority Official government source; allocation is not current installed stock and the 70% 30% split is not two inventories. ACC-17 Global product reference · NVIDIA H100 vendor page NVIDIA H100 · ND Available product reference; not inventory Fourth-generation Tensor Cores; FP8 Transformer Engine; 900 GB s GPU-to-GPU interconnect; 60 TFLOPS FP64 NVIDIA OEM Purchaser financier ND OEM price ND ND page read 2026-09-16 NVIDIA H100 Official product specification; does not validate project counts or form factors. ACC-18 Global product reference · NVIDIA H200 vendor page NVIDIA H200 · ND Available product reference; not inventory 141 GB HBM3e at 4.8 TB s; preliminary specifications NVIDIA OEM Purchaser financier ND OEM price ND ND page read 2026-09-16 NVIDIA H200 Official product specification; does not re-count the Barcelona host commitment. ACC-19 Global catalog reference · Lambda GPU instances NVIDIA B200 SXM6 · ND On-demand catalog orderability; not fleet inventory 180 GB VRAM GPU; listed 8x context includes 208 vCPUs and 2,900 GiB RAM Lambda cloud; NVIDIA OEM ND US$6.69 GPU-hour plus applicable sales tax VAT GST; hosted rental only ND catalog read 2026-09-16 Lambda instances Official provider catalog; orderability and 1–8 GPU or 16–2,000+ cluster listings do not prove physical inventory. ACC-20 Global product reference · NVIDIA GB300 NVL72 NVIDIA GB300 NVL72 · 72 GPUs per rack-scale platform Available product reference; topology, not site inventory Fully liquid cooled; 72 Blackwell Ultra GPUs, 36 Grace CPUs, 800 Gb s connectivity per GPU NVIDIA OEM Customer project allocation ND OEM price ND ND page read 2026-09-16 NVIDIA GB300 NVL72 Official product specification; topology is not mapped onto Yotta, Mistral, or HUMAIN counts. ACC-21 Global product reference · NVIDIA Vera Rubin NVL72 NVIDIA Vera Rubin NVL72 · 72 GPUs per rack-scale platform Full-production product reference; topology, not site inventory 72 Rubin GPUs, 36 Vera CPUs, ConnectX-9 SuperNICs, BlueField-4 DPUs, NVLink 6 NVIDIA OEM Customer project allocation ND OEM price ND ND page read 2026-09-16 NVIDIA Vera Rubin NVL72 Official product specification; rack topology is not a delivery record for Nscale or Noetra. ACC-22 Global cloud product reference · Google Cloud TPU7x Ironwood Google TPU7x · 9,216 chips per pod Available cloud product; configuration footprint, not installed inventory 2,307 TFLOPS BF16 chip; 4,614 TFLOPS FP8 chip; 192 GiB HBM; 7,380 GB s HBM; 1,200 GB s bidirectional ICI; 100 Gbps DCN Google Cloud Google; customer allocation ND ND ND documentation read 2026-09-16 Google TPU7x Official cloud documentation; TPU is an accelerator, not NVIDIA AMD GPU inventory. ACC-23 Global product reference · AMD Instinct MI355X AMD Instinct MI355X · ND Product reference; not inventory 4th Gen AMD CDNA; 288 GB HBM3E; 8 TB s bandwidth; MXFP6 and MXFP4 support AMD OEM Customer project allocation ND OEM price ND ND page read 2026-09-16 AMD MI355X Official product specification; no MI355X project count assigned. ACC-24 Global cloud product reference · Amazon EC2 Trn2 Trainium2 AWS Trainium2 · 16 chips per Trn2 instance; 64 per UltraServer Available cloud product; topology, not Project Rainier global stock Up to 6 TB HBM and 185 TB s bandwidth per UltraServer; NeuronLink AWS AWS; customer allocation ND ND ND documentation read 2026-09-16 AWS Trn2 Official cloud documentation; custom accelerator topology is not added to NVIDIA AMD GPU rows.

Separate electricity and price checkpoints

ID Geography scope Status Observed value Company role backing Price or cost basis Source date Direct evidence Verification qualification --- --- --- --- --- --- --- --- --- ENERGY-01 North America · Susquehanna, Pennsylvania · Talen Energy Amazon AWS nuclear PPA PPA with delivery ramp; not a data-center IT-load record 1,920 MW carbon-free nuclear power at full contract quantity; full volume expected no later than 2032 Talen generator and retail supplier; Amazon AWS offtaker; PPL Electric Utilities transmission delivery PPA tariff ND 2025-06-11 Talen Amazon PPA Official investor-relations source; generation PPA only; do not add to any data-center IT, facility, or grid row. PRICE-01 Global provider catalog · RunPod Hosted GPU rental list rate B300 US$7.89, H200 US$4.59, B200 US$6.79, H100 PCIe US$2.89, H100 SXM US$3.49, L40 US$0.82 per GPU-hour RunPod provider; underlying hardware vendor allocation ND Dynamic rental list rates; not hardware purchase price, site capex, electricity tariff, or installed inventory 2026-09-13 page update; read 2026-09-16 RunPod pricing Official provider catalog; region, tier, tax, storage, networking, and availability can change effective rates. PRICE-02 Global provider catalog · Lambda Hosted GPU instance rental list rate B200 SXM6 US$6.69, H100 SXM US$3.99, A100 SXM 80 GB US$2.79 per GPU-hour Lambda provider; NVIDIA OEM; customer allocation ND Rental only; applicable sales tax VAT GST; not hardware price ND catalog read 2026-09-16 Lambda instances Official provider catalog; no physical fleet count or OEM purchase price inferred. PRICE-03 Global provider catalog · CoreWeave On-demand cloud instance rental HGX H100 US$49.24, HGX H200 US$50.44, HGX B200 US$68.80 per 8-GPU instance-hour; GB300 contact sales CoreWeave provider; customer project allocation ND Instance-hour rental; do not divide or call it per-card purchase price ND pricing page read 2026-09-16 CoreWeave pricing Official provider catalog; instance scope is not comparable with per-GPU-hour rates without an explicit normalization.

Method and limitations

This is a bounded source-qualified register across North America, Latin America, Europe, the Middle East, Africa, and Asia-Pacific—not a global census or underwriting model. Official operator, developer, vendor, government, investor-relations, utility, and clearly labelled secondary sources were retained with their source dates and direct links. ND means not disclosed or not verified in the reviewed evidence. Product specifications are kept separate from deployment inventory, and rental rates, financing, lease or contract value, generation PPA, and construction-cost signals retain their original economic basis.

The principal limitations are current commissioning and delivery uncertainty for several projects, operator vendor government claims that are not independent stocktakes, unresolved parent child and tranche overlap, mixed IT facility grid generation MVA boundaries, and the absence of clean OEM prices or site-specific electricity tariffs. No MVA-to-MW or MW-to-GPU conversion was made, and no missing observation was represented as zero. Existing quantitative views remain bounded comparisons; the new values are not forced into a roll-up when units or states are unlike.

Current-run update — 17 September 2026

As of: 17 September 2026 UTC. This additive evidence screen uses only the accepted current-run evidence. It contains 12 data-center project or construction records, 8 GPU, accelerator, or model records, and 7 price-basis checkpoints. Earlier selected rows and the existing quantitative chart slots remain intact. ND means not disclosed or not verified in the reviewed evidence; it does not mean zero.

The current screen keeps announced, planned, under-construction, energized, operational, product-reference, and cloud-orderability states separate. IT load, compute capacity, facility or campus power, grid interconnection, generation PPA, product topology, deployment claims, and economic objects remain in their source-defined boundaries. No global or regional total is asserted.

Data-center project and construction records

ID Geography project Lifecycle status Possible announced capacity IT data-center capacity Electricity grid capacity Company behind operator Investor backer Capex possible price Price basis Source date Direct source Evidence status --- --- --- --- --- --- --- --- --- --- --- --- --- P1 North America · Abilene, Texas, US · Crusoe Oracle Stargate flagship First phase operational; six-building expansion under construction; final completion ND 1.2 GW campus; approximately 4M sq ft; mid-2026 completion target not confirmed Completion design supports hundreds of thousands of GPUs; each building designed for up to 50,000 GB200 NVL72 systems; design figures, not inventory 1.2 GW campus power grid-interconnection signal; initial two buildings 200+ MW and energized Crusoe; Oracle Cloud Infrastructure customer operator context Crusoe, Blue Owl Real Assets funds, Primary Digital Infrastructure; Oracle customer Approximately US$15B Joint-venture funding for the 1.2 GW second phase; not normalized campus capex or GPU price 2025-03-18; 2025-05-21; 2025-09-30; 2026-06-09 Abilene expansion · JV Official sources; current update still describes part of the campus under construction, so the 1.2 GW figure is not treated as energized or operational total P2 North America · Richland Parish, Louisiana, US · Meta Hyperion Under construction and expanding; operational completion date ND 5 GW compute capacity 5 GW compute capacity; source does not label it IT load Grid and facility MW ND Meta; Entergy Louisiana, DPR, Turner, and Mortenson in project context Blue Owl funds 80%; Meta 20%; pro-rata JV commitment Approximately US$27B development costs; approximately US$7B Blue Owl cash contribution Transaction and development figures; not grid cost, GPU bill of materials, or rental rate 2025-10-21; 2026-07-13 Meta Blue Owl · Meta expansion Official sources; compute capacity is retained as compute and is not relabelled as IT, facility, or electricity MW P3 North America · Mount Pleasant, Wisconsin, US · Microsoft Fairwater First facility fully operational; adjacent second facility under construction and scheduled for 2028 315-acre, three-building campus; 1.2M sq ft; campus GPU count exact stock ND Hundreds of thousands of latest NVIDIA GPUs described at campus level; exact inventory ND MW and grid interconnection ND; energy electrical infrastructure reported prepaid Microsoft Microsoft; separate project-investor allocation ND Approximately US$4.7B local hyperscale construction spend for 2024–2028 Local portfolio-period construction spend; not facility-only capex or GPU price 2025-09-18; 2026-06-23 Microsoft facility · Microsoft campus Official sources; first-facility operation is reported, but site MW, grid, generation, exact GPU inventory, and financing split remain ND P4 North America · Memphis, Tennessee, US · xAI SpaceXAI Colossus First cluster operating by official claim; Colossus 2 expansion reported but not independently confirmed 200,000 H100 official cluster claim; 1M roadmap is future; secondary 230,000 mixed-GPU claim conflicts Model-level deployed count is an inventory claim, not a disclosed IT-load figure 150 MW TVA MLGW grid connection, split 8 MW existing substation and 142 MW new substation; search-result evidence xAI SpaceXAI Valor Equity Partners reported in secondary financing context; terms under negotiation US$12B debt sought Reported debt seek for Colossus 2 and GPU compute; secondary and not confirmed capex or purchase price Undated current; 2025-05-05; 2025-07-22 xAI Colossus · MLGW update · DCD report Official operator claim plus official search-result excerpt and secondary financing report; utility read failed, and conflicting inventory claims are not combined P5 Europe · Kvandal near Narvik, Norway · Nscale Aker OpenAI Stargate Announced planned; preparatory construction steps disclosed; energization ND 230 MW planned; ambition for an additional 290 MW; 100,000 NVIDIA GPU target IT-load split ND 230 MW planned; possible 520 MW including ambition; renewable-power claim, exact generation PPA MW ND Nscale designs builds; planned 50 50 JV with Aker; OpenAI offtaker Nscale and Aker; OpenAI offtake option; Microsoft later expansion customer context Approximately US$1B initial 20 MW phase; more than US$250M equity Project funding on a 100% basis; not full 230 520 MW build cost or GPU price 2025-07-31 Nscale Norway Official target and project-commitment evidence; GPU, power, and capacity figures are planned or ambitious, not delivered or energized stock P6 Europe · Glomfjord, Norway · Nscale AI data center Operational claim; expansion timing ND 30 MW operational capacity, expandable to 60 MW IT-load split ND Renewable hydropower stated; a 120 MW hydropower plant is in the town, not identified as dedicated site capacity Nscale Backer and capex ND ND No price disclosed Undated current Nscale Glomfjord Official location page; operating capacity is source-reported and is not converted into IT load or dedicated generation capacity P7 Europe · Lappeenranta, Finland · Nebius AI factory Announced planned construction; first customer capacity expected in 2027 Up to 310 MW AI-factory capacity IT-load split ND Clean-energy design stated; exact interconnection and generation PPA MW ND Nebius; Polarnode implements; city and local industrial-site partners City lease context; exact project investor allocation ND ND Multi-billion-euro project context is not assigned as site capex 2026-03-31 Nebius Official source; 310 MW remains source-defined AI-factory capacity and construction progress energization are ND P8 Middle East · Abu Dhabi, UAE · Stargate UAE UAE–US AI campus Announced planned; first live phase target in 2026 1 GW Stargate cluster within a 5 GW AI-campus data-center envelope IT-load split ND Nuclear, solar, and natural-gas power mix stated; exact grid interconnection ND G42 builds; OpenAI and Oracle operate; NVIDIA, Cisco, and SoftBank partners G42, OpenAI, Oracle, NVIDIA, Cisco, SoftBank, and UAE–US government framework ND No project price disclosed 2025-05-22 OpenAI Stargate UAE · Abu Dhabi Media Office Official sources; 1 GW cluster and 5 GW campus are separate scopes, not installed GPU, IT-load, or grid totals P9 Middle East · Oxagon, NEOM, Saudi Arabia · DataVolt net-zero AI factory campus Announced planned; first phase targeted for operation in 2028 1.5 GW factory campus IT grid split ND Entirely renewable energy is the stated ambition; exact grid capacity and generation split ND DataVolt developer, investor, and operator; NEOM Oxagon host DataVolt and NEOM partnership US$5B Initial first-phase investment; not full-campus capex or per-GPU price 2025-02-10; 2025-02-11 NEOM DataVolt · DataVolt Official sources; 1.5 GW is factory campus capacity and renewable ambition has no disclosed MW split P10 Asia-Pacific · Kulai, Johor, Malaysia · YTL Green Data Centre Park JDC1 JDC1 operational claim; park expansion remains a development envelope; Q3-2025 target not re-verified Park up to 600 MW; JDC1 72 MW facility claim; AI supercomputer plan using NVIDIA GB200 via DGX Cloud Live IT utilization ND National-grid access and solar-powered facility claims; binding feed and generation MW ND YTL Data Center Holdings YTL Power; YTL Cloud Backer and project-finance allocation ND ND No current project price disclosed 2025-05-23; current location pages YTL park · YTL JDC1 · YTL AI supercomputer Official sources; park, JDC1 facility, solar generation, grid access, and live IT GPU utilization remain separate; current status needs refresh P11 Asia-Pacific · Iskandar Puteri, Johor, Malaysia · TM Nxera AI-ready green campus Under construction; first building topped out; first-phase operations targeted H2 2026 First building scalable up to 200 MW; campus scalable above 200 MW IT facility boundary ND 280 MW secured under a multiyear electricity agreement with TNB TM Nxera JV between Telekom Malaysia and Nxera, Singtel’s regional data-center arm TM and Nxera Singtel Digital InfraCo; TNB electricity provider ND Secured electricity agreement; no capex or tariff disclosed 2026-01-12; 2026-04-16 TM Nxera campus · Topping out Official sources; 200 MW scalability and 280 MW electricity are separate source-defined objects, not an energized IT total P12 Asia-Pacific · Inzai East Tokyo, Japan · AirTrunk TOK1 Under construction and continuing phased development; later completion ND 40+ MW phase; 300+ MW total campus 100+ MW IT load being added in financing source; 40+ MW phase and campus total have different scopes Grid interconnection MW ND AirTrunk developer operator Financing led by SMBC, MUFG, Crédit Agricole CIB, and Société Générale among the green-loan syndicate JP¥191.6B US$1.24B Green loan for refinancing and continued development; not total TOK1 capex or GPU price 2025-06-19; 2026-03-10 TOK1 phase · TOK1 green loan Official sources; 40+ MW, 300+ MW, and 100+ MW IT load are not additive, and current commissioning remains ND

GPU inventory and specification records

ID Geography record Model specification Public quantity state Company behind vendor Investor backer Possible price Price basis Source date Direct source Evidence status --- --- --- --- --- --- --- --- --- --- --- G1 Global product reference · NVIDIA H100 Tensor Core Hopper; 80 GB HBM; fourth-generation Tensor Cores; FP8 Transformer Engine; 900 GB s GPU-to-GPU interconnect ND; product reference, fleet count ND NVIDIA; cloud availability via Azure Purchaser, investor, and financing roles ND US$3.26 GPU-hour DigitalOcean HGX H100 12-month reserved cloud rental, effective 2026-08-01; not purchase price Undated current; 2026-08-01 NVIDIA H100 · DigitalOcean GPU pricing Official product and provider-rate references; rate is not a physical inventory or OEM price G2 Global product reference · NVIDIA H200 Tensor Core Hopper; 141 GB HBM3e; 4.8 TB s memory bandwidth; Azure 8-GPU VM and 900 GB s NVLink context ND; product reference, fleet count ND NVIDIA; cloud availability via Azure Purchaser, investor, and financing roles ND US$3.40 GPU-hour DigitalOcean HGX H200 12-month reserved cloud rental, effective 2026-08-01; not purchase price Undated current; 2026-08-01 NVIDIA H200 · DigitalOcean GPU pricing Official product and provider-rate references; hosted rate is not hardware price or installed count G3 Global system reference · NVIDIA DGX B200 Eight Blackwell GPUs; 1,440 GB total HBM3e; 64 TB s HBM3e bandwidth; approximately 14.3 kW maximum system power 8 GPUs per DGX B200 system topology; fleet count ND NVIDIA OEM and system vendor Purchaser, investor, and financing roles ND ND OEM system price not disclosed; no per-GPU estimate inferred Undated current DGX B200 Official product specification; system topology is not deployment inventory G4 Global rack reference · NVIDIA GB200 NVL72 Rack-scale Blackwell plus Grace; 36 Grace CPUs and 72 Blackwell GPUs; 130 TB s low-latency communications; 1.8 TB s GPU-to-GPU interconnect 72 GPUs per rack topology; fleet count ND NVIDIA OEM and platform provider Customer, investor, and financing roles ND ND OEM rack price not disclosed Undated current GB200 NVL72 Official product specification; rack topology is not a site inventory count G5 Global product reference · NVIDIA A100 80 GB Ampere; 80 GB; more than 2 TB s memory bandwidth; up to seven MIG GPU instances ND; product reference, fleet count ND NVIDIA OEM Purchaser, investor, and financing roles ND US$3.09 GPU-hour DigitalOcean dedicated GPU listing; term not stated; hosted rental, not purchase price Undated current NVIDIA A100 · DigitalOcean additional GPUs Official product and provider-rate references; listed rate does not establish hardware value or fleet count G6 Global product reference · AMD Instinct MI300X CDNA 3; 192 GB HBM3; 5.325 TB s peak theoretical bandwidth; 750 W OAM accelerator ND; product reference, fleet count ND AMD OEM Purchaser, investor, and financing roles ND US$1.91 GPU-hour DigitalOcean MI300X 12-month reserved cloud rental, effective 2026-08-01; not purchase price Undated current; 2026-08-01 AMD MI300X · DigitalOcean GPU pricing Official product and provider-rate references; rental basis is not comparable OEM pricing G7 North America · Memphis, US · xAI Colossus NVIDIA H100 deployment; H100 reference specifications are listed in G1 200,000 H100; official operating claim for one interconnected cluster; 1M roadmap excluded xAI SpaceXAI; NVIDIA ND on cited source ND No project or OEM price disclosed Undated current xAI Colossus Official operator claim, not an independent stocktake; not added to product-reference rows G8 Europe · Mäntsälä, Finland · Nebius Mixed deployment: H100 installed, H200 deployment from November, Blackwell planned in 2025; exact model mix ND Up to 60,000 GPUs enabled by expansion; model-level counts ND; site expansion to 75 MW Nebius; NVIDIA models Investor, lender, and financing roles ND ND No price disclosed; source is historical and does not establish current installed state 2024-10-08 Nebius Finland expansion Official company source; current installed state and model-level inventory need refresh

Price-basis checkpoints

Record Published figure Economic type price basis Source date Direct source Evidence status --- --- --- --- --- --- Abilene Stargate JV Approximately US$15B Joint-venture funding for the 1.2 GW second phase; not comparable full-campus capex or GPU price 2025-05-21 Crusoe Blue Owl Primary Digital JV Official funding signal; phase and campus scope retained Meta Hyperion Approximately US$27B total development costs; approximately US$7B Blue Owl cash contribution Buildings and long-lived power, cooling, and connectivity infrastructure plus ownership transaction terms; not GPU price 2025-10-21 Meta Blue Owl Official transaction development figures; site-level allocation and normalized capex unavailable Stargate Norway Approximately US$1B initial 20 MW phase; more than US$250M equity Initial project funding on a 100% basis; not full 230 520 MW build cost or GPU price 2025-07-31 Nscale Norway Official project commitment; not a unit price NEOM DataVolt US$5B Initial first-phase factory investment; not full 1.5 GW campus capex or GPU price 2025-02-10; 2025-02-11 NEOM · DataVolt Official investment signal; full-campus and per-unit economics ND AirTrunk TOK1 JP¥191.6B US$1.24B Green loan for refinancing and continued development; not total TOK1 capex or GPU price 2026-03-10 AirTrunk green loan Official financing signal; 100+ MW IT addition remains separate from campus and phase values DigitalOcean H100 H200 MI300X US$3.26 US$3.40 US$1.91 per GPU-hour 12-month reserved hosted-cloud rental, effective 2026-08-01; not OEM purchase price or site operating cost 2026-08-01 DigitalOcean GPU pricing Official provider rate card; terms and availability can change DigitalOcean A100 US$3.09 per GPU-hour Listed dedicated GPU rental rate; term not stated; not OEM purchase price Current listing DigitalOcean additional GPUs Official provider rate card; no physical fleet count inferred

Methodology and limitations

This is a bounded, source-qualified snapshot rather than a global census or underwriting model. Capacity remains in the source-defined class: compute, IT, facility, campus, grid, generation PPA, and MVA are not substituted for one another. Lifecycle labels retain the evidence state; a target, planned phase, financing signal, product topology, or catalog listing is not promoted to energized, operating, delivered, or installed inventory. Prices retain their original object, currency, unit, term, and date.

The packet combines official operator, developer, vendor, government, and investor-relations sources with one clearly labelled secondary record. The xAI utility value is based on an official search-result excerpt whose direct document read failed, and the xAI financing report is secondary; these qualifications remain visible in the table. Product specifications and cloud rental rates do not establish deployment counts or OEM purchase prices. Parent and child phases, project targets, and overlapping tranches are not added. No MVA-to-MW or MW-to-GPU conversion was made, and no missing observation was represented as zero.

Current-run update — 18 September 2026

As of: 18 September 2026 UTC, with the quote snapshot observed at 12:22 UTC. This additive premarket overlay adds no data-center project, electricity, GPU, or price-ledger row. It retains the 17 September capacity GPU register and adds only the current market screen and attributed industry context. ND remains not disclosed or not verified, not zero.

Market context and key observed levels

The liquid sample was mixed-to-positive on derived bid ask midpoints, with $QQQ at 718.105 +0.165% , $SMH at 561.445 +0.149% , and $SPY at 760.465 -0.032% . Within the AI and semiconductor basket, $GOOGL at 354.415 +2.040% , $META at 691.855 +1.399% , $INTC at 109.925 +1.034% , $MU at 983.295 +0.593% , $AVGO at 348.895 +0.459% , $ANET at 200.150 +0.311% , and $AMZN at 251.845 +0.261% were higher. $NVDA at 219.315 -0.011% was effectively flat; $TSM at 428.525 -0.403% , $ORCL at 149.720 -0.578% , $MSFT at 496.320 -0.287% , and $SMCI at 40.055 -0.731% were lower. $DLR and $EQIX are retained in the quote table but excluded from the directional read because their quote fields were wide and older.

Symbol Role Prev. close Bid–ask Indicative mid Gap vs. prev. close Spread Read --- --- ---: --- ---: ---: ---: --- $SPY S&P 500 proxy 760.711 760.45–760.48 760.465 -0.032% 0.03 Tight; broad context $QQQ Nasdaq-100 proxy 716.920 718.08–718.13 718.105 +0.165% 0.05 Tight; broad tech context $SMH Semiconductor ETF 560.610 561.30–561.59 561.445 +0.149% 0.29 Moderate; sector context $NVDA AI accelerator OEM 219.340 219.29–219.34 219.315 -0.011% 0.05 Tight; effectively flat $AMD AI accelerator OEM 545.090 545.49–545.98 545.735 +0.118% 0.49 Moderate; marginally higher $AVGO AI networking custom silicon 347.300 348.79–349.00 348.895 +0.459% 0.21 Tight; higher $INTC CPU foundry AI infrastructure 108.800 109.90–109.95 109.925 +1.034% 0.05 Tight; higher $MU HBM memory 977.500 983.00–983.59 983.295 +0.593% 0.59 Moderate; higher $MRVL Custom silicon networking 240.760 240.65–241.10 240.875 +0.048% 0.45 Moderate; near flat $ANET AI networking 199.530 199.80–200.50 200.150 +0.311% 0.70 Moderate; higher $SMCI AI servers 40.350 39.99–40.12 40.055 -0.731% 0.13 Tight; lower $TSM Semiconductor foundry 430.260 428.35–428.70 428.525 -0.403% 0.35 Moderate; lower $ORCL Hyperscaler AI infrastructure customer 150.590 149.64–149.80 149.720 -0.578% 0.16 Tight; lower $MSFT Hyperscaler AI infrastructure 497.750 496.10–496.54 496.320 -0.287% 0.44 Tight; lower $GOOGL Hyperscaler Google Cloud 347.330 354.35–354.48 354.415 +2.040% 0.13 Tight; higher $AMZN Hyperscaler AWS 251.190 251.79–251.90 251.845 +0.261% 0.11 Tight; higher $META Hyperscaler AI infrastructure 682.310 691.50–692.21 691.855 +1.399% 0.71 Moderate; higher $DLR Data-center operator 184.260 180.57–186.00 183.285 -0.529% 5.43 Wide older; exclude $EQIX Data-center operator 1,025.940 986.10–1,050.00 1,018.050 -0.769% 63.90 Very wide older; exclude

These are observed quote levels and derived gaps, not support resistance targets or trade signals. The returned regular-session last and prior-close fields were flat, so they were not used as premarket performance.

Catalysts and power context

• CNBC and Bloomberg attributed a forecast from NVIDIA CEO Jensen Huang that NVIDIA expects to sell twice as many chips in the following year. The evidence is attributed search-result reporting, not a verified shipment or installed-GPU count: CNBC · Bloomberg.
• TipRanks’ 18 September market report associated higher AI-semiconductor premarket trading with stronger demand signals. It is secondary market context, not a primary company disclosure or proof of causation: TipRanks.
• TrendForce estimates 161 GW of global data-center power-demand capacity in 2026, about 31% year over year, with AI servers at 33.4% of total demand capacity. This is industry context and is not additive to named project IT load, facility power, grid capacity, generation PPA, or GPU inventory: TrendForce.

Risks, confidence, and evidence controls

Confidence in the directional premarket read is low. Bid ask indications are not executed trades; no dedicated extended-hours print or numeric futures level was separately verified; $DLR and $EQIX are not directionally usable in this pull; and the relationship between the cited headlines and the quote snapshot is not independently established. Huang’s forecast does not prove delivery, while the 161 GW industry estimate does not establish a global site total. The overlay is analysis only and is not an instruction to place a trade.

The capacity and GPU conclusion is unchanged: the evidence remains a bounded, source-qualified ledger with no defensible global MW, electricity, installed-GPU, capital, or comparable hardware-price total. Parent programs, child phases, generation, grid access, IT load, compute capacity, system topology, cloud rental, and financing remain separate economic or physical objects.

Source list

• Dated bid ask snapshot observed 18 September 2026 at 12:22 UTC; midpoint, gap, spread, and quote-quality labels are derived fields.
• CNBC and Bloomberg for attributed chip-demand commentary.
• TipRanks for secondary premarket context.
• TrendForce for the source-defined power-demand estimate.

Latest evidence update — 21 September 2026

As of 21 September 2026 UTC, the accepted packet is a bounded representative sample, not a global census. It covers nine facility or enabling-power records across North America, Europe, Asia, the Middle East, and Africa; eight GPU inventory or deployment records; three vendor specification anchors; and seven price checkpoints. ND means not disclosed or not verified, not zero. The rows below keep construction status, facility or IT capacity, electricity or interconnection, GPU inventory or deployment state, company or operator, investor or funder, and price or cost evidence separate from one another and from the historical dated sections above.

Project and construction status

ID project Region Status supported by source Facility IT capacity Electricity interconnection Company operator backing Capital price Source --- --- --- --- --- --- --- --- DC-01 · Stargate additional Oracle capacity — United States Abilene North America Under development; cited source reported parts of Abilene up and running 4,500 MW additional U.S. capacity; 5,000 MW total platform under development ND OpenAI and Oracle; SoftBank and CoreWeave named as platform partners; site allocation ND US$500B four-year U.S. AI-infrastructure program commitment; not site capex OpenAI DC-02 · Crusoe Lancium Clean Campus — Abilene, Texas North America Under construction; present-day status beyond the release was not refreshed 1,200 MW total facility power; 200+ MW initial phase 1,200 MW grid interconnection Crusoe and Lancium; funder allocation ND No project capex; about US$1B economic-impact estimate is not capex Crusoe DC-03 · Nebius AI factory — Lappeenranta, Finland Europe Construction announced; first customer capacity expected in 2027 Up to 310 MW AI-factory capacity Site-specific MW ND; company-wide target not assigned to site Nebius; named project funder ND Site capex and project funder ND Nebius DC-04 · Nebius AI factory — near Lille, France Europe Building status reported; present-day status not independently refreshed 240 MW when fully deployed ND Nebius; named project funder ND Site capex and project funder ND Nebius DC-05 · YTL Green Data Center Park — Johor, Malaysia Asia Campus planned; historical construction and service targets not independently refreshed 500 MW campus; 168 MW GDS subset; 72 MW separate construction 500 MW equivalent on-site solar generation planned; grid MW ND YTL Data Center Holdings YTL Power; GDS co-developer; Sea anchor tenant RM1.5B first-phase investment; not full-campus capex YTL · MIDA DC-06 · Google data center and Cloud region — Selangor, Malaysia Asia Ground broken; construction underway in cited source; operation not refreshed ND ND; renewable-energy programs lack site MW Google Alphabet; program allocation ND US$2B Malaysia investment spanning data center, Cloud region, and related programs Google DC-07 · DataVolt AI factory campus — Oxagon, Saudi Arabia Middle East Announced planned; phase one expected operational in 2028 1,500 MW total factory capacity Renewable-powered ambition; exact MW ND DataVolt; NEOM Oxagon host; partnership allocation disclosed US$5B initial phase-one investment; not full-campus capex DataVolt · SPA DC-08 · Stargate UAE cluster — Abu Dhabi, UAE Middle East Announced planned; 200 MW expected to go live in 2026 in cited announcement 1,000 MW cluster; 200 MW first-phase target ND OpenAI, G42, Oracle, NVIDIA, Cisco, SoftBank; project split ND Project-level price and funding split ND Official URL · Readable mirror DC-09 · Teraco 120 MW solar PV plant — Free State, South Africa Africa Under construction; expected online late 2026 Supported data-center IT facility capacity ND 120 MW solar PV plant; 120 MW Eskom grid-capacity allocation Teraco Digital Realty; JUWI and Subsolar delivery partners Green loan raised; lender and amount ND Teraco construction · Grid allocation

Electricity and interconnection records

Record Value Metric Boundary Source --- ---: --- --- --- DC-02 · Crusoe Abilene 1,200 MW Grid interconnection Electricity interconnection; not automatically IT load Crusoe DC-05 · YTL Green Data Center Park 500 MW Equivalent on-site solar generation planned Generation; not grid interconnection or delivered output YTL DC-09 · Teraco Free State 120 MW Solar PV plant Enabling generation; supported IT capacity ND Teraco DC-09 · Teraco Free State 120 MW Eskom grid-capacity allocation Grid allocation; not site IT load Teraco grid allocation

GPU inventory, deployment state, and possible price

ID record Status Model specification Public quantity or state Company vendor Investor funder Possible price cost Source --- --- --- --- --- --- --- --- GPU-01 · Crusoe Abilene Construction design disclosure NVIDIA GB200 NVL72; 72 Blackwell GPUs and 36 Grace CPUs per rack reference Up to 50,000 per building; total not published Crusoe; NVIDIA ND No hardware price disclosed Crusoe GPU-02 · OpenAI Oracle Stargate U.S. Early workloads; first racks delivered in cited source NVIDIA GB200 racks; GB200 NVL72 reference is 72 Blackwell GPUs per rack Over 2,000,000 chips; not normalized to GPUs OpenAI; Oracle; NVIDIA OpenAI, Oracle, SoftBank, and other partners; split ND US$500B program commitment; not GPU purchase price OpenAI GPU-03 · Oracle OCI Supercluster Generally available service offer NVIDIA GB200 NVL72 Blackwell Up to 131,072 Blackwell GPUs available; owned installed count ND Oracle; NVIDIA Commercial funding split ND Customer price ND Oracle GPU-04 · xAI Colossus Online cluster and planned expansion NVIDIA H100 H200; H100 reference includes 900 GB s GPU-to-GPU interconnect 100,000 H100 online; 200,000 planned total; 50,000 H200 in cited expansion xAI SpaceXAI; NVIDIA ND No hardware price disclosed Tom's Hardware GPU-05 · Yotta Shakti Cloud, India Planned procurement; delivery not independently verified NVIDIA H100 and GH200; exact model split ND 16,000 in latest purchase plan; prior order kept separate Yotta; NVIDIA ND US$500M latest AI purchase plan; broader US$1B GPU-spend plan DCD Reuters GPU-06 · CoreWeave global GPU cloud Active fleet reported; model mix undisclosed Mostly NVIDIA Hopper including H100 250,000 NVIDIA GPUs reported; not a named-site stocktake CoreWeave Financing not isolated to GPU fleet US$0.98 hour H100 versus US$2.20 hour GB200 rack-system market comparison CNBC GPU-07 · Meta global AI infrastructure 2025 target; current delivery not independently verified Model mix undisclosed Over 1,300,000 GPUs targeted by year-end 2025; about 1 GW compute online separately Meta Corporate self-funded CapEx; external project funder ND US$60B–80B 2025 CapEx range; not GPU purchase price TechCrunch GPU-08 · Nebius Finland AI infrastructure Operational planned platform status; exact current count ND NVIDIA GB300 NVL72; Vera Rubin NVL72 planned later; exact specs counts ND Count undisclosed; Mäntsälä operational GB300 deployment; Lappeenranta planned Blackwell Vera Rubin service Nebius; NVIDIA ND No GPU price disclosed Nebius

Vendor specification anchors

Platform Vendor Exact sourced specification Interpretation Source --- --- --- --- --- H100 Tensor Core GPU NVIDIA Hopper architecture; fourth-generation NVLink with 900 GB s GPU-to-GPU interconnect Specification anchor only; no fleet count inferred NVIDIA H100 H200 Tensor Core GPU NVIDIA Hopper architecture; 141 GB HBM3e at 4.8 TB s; same power profile as H100 stated by NVIDIA Specification anchor only; no fleet count inferred NVIDIA H200 GB200 NVL72 NVIDIA Rack-scale Grace Blackwell design with 36 Grace CPUs and 72 Blackwell GPUs; 130 TB s NVLink Switch System bandwidth Rack topology is not a deployed site inventory NVIDIA GB200 NVL72

Price and investment checkpoints

Entity project Published figure Economic type Scope and boundary Source --- ---: --- --- --- Stargate U.S. US$500B Program commitment Four-year U.S. AI-infrastructure commitment; not site capex or GPU price OpenAI YTL Green Data Center Park first phase RM1.5B First-phase investment Johor first phase; not full-campus capex or GPU price MIDA Google Malaysia US$2B Country infrastructure investment Includes data center, Cloud region, and related programs; not isolated facility capex Google DataVolt Oxagon phase one US$5B Phase-one investment Initial investment; not full-campus capex or per-GPU price DataVolt Yotta latest GPU purchase plan US$500M Hardware order plan 16,000 H100 GH200 plan; not paid invoice or delivered inventory DCD Reuters Meta 2025 CapEx US$60B–80B Corporate CapEx range Primarily data centers and AI teams; not GPU unit price TechCrunch H100 GB200 market comparison US$0.98 hour vs US$2.20 hour Market comparison CNBC comparison; not CoreWeave tariff, OEM price, or project economics CNBC

Methodology and limitations

The packet was fetched on 21 September 2026 UTC and retains source-defined units and dates. GW-to-MW conversions use an exact factor of 1,000 and are labelled as conversions; lower bounds such as “over,” ceilings such as “up to,” per-building designs, targets, planned procurement, operating claims, service availability, and chips are not silently converted into delivered GPU inventory. No MVA-to-MW or MW-to-GPU conversion was made, and no unlike measures were summed.

Announced or planned figures are not construction or operating proof. Under-construction records are not operating capacity. Operating claims such as xAI’s 100,000 H100 figure and CoreWeave’s reported fleet remain source-qualified rather than independently audited stocktakes. Estimates and economic signals retain their original object: program commitment, first-phase investment, country investment, CapEx range, hardware-order plan, or market comparison. Unknowns include most site-specific electricity or interconnection values, current status for historical announcements, investor allocation, exact model mix, and OEM purchase price. The Stargate UAE official page remains unresolved; its mirror is retained only with that qualification.

Latest evidence update — 22 September 2026

As of 22 September 2026 UTC, the accepted packet is a bounded ten-project evidence set, not a global census. It covers planned, pre-construction, under-construction, operational, and expanding examples across North America, Europe, the Middle East, and Asia-Pacific. The source-defined measurement boundary is retained on every row: a facility, campus, factory, compute, IT-load, grid, generation, GPU, financing, or rental figure is not silently relabelled as another object. ND means not disclosed or not verified, not zero.

Current evidence read

The strongest directly comparable current observations are Yotta D1's 30 MW IT power and AirTrunk TOK1's more than 100 MW of additional IT load under construction . For GPU inventory, xAI SpaceXAI reports 200,000 NVIDIA H100 GPUs in the Colossus cluster, while Yotta reports a first NM1 cluster of more than 4,000 NVIDIA H100 GPUs . Nscale's 100,000 NVIDIA GPU figure is a target for end-2026, not deployed inventory. The remaining capacity and GPU records are retained as source-defined, non-additive signals.

Project comparison — normalized table

Project Region Status at evidence date Capacity boundary Explicit IT load GPU inventory state Company operator Investor backer Public price cost signal and type Dated evidence --- --- --- --- --- --- --- --- --- --- STG-TX-MILAM — Stargate SB Energy Milam County North America · Texas, US Under construction service preparation; initial facilities expected to enter service from 2026 1.2 GW data-center site; source does not define IT versus utility connection capacity ND Model and count unknown; do not assign Stargate-wide chip figures SB Energy builds and operates; OpenAI is customer partner OpenAI and SoftBank each invested US$500M in SB Energy; Ares supplied US$800M preferred equity 1.2 GW lease with value undisclosed; the US$1B and US$800M figures are platform company financing, not site capex OpenAI — 2026-01-09 · OpenAI platform context — 2025-07-22 HYPERION-LA — Meta Hyperion North America · Louisiana, US Under construction and expanding; completion date ND 5 GW compute capacity; not labelled IT or grid capacity ND Model and count unknown Meta and JV; Meta provides construction and property management Blue Owl-managed funds 80%; Meta 20% Approximately US$27B JV development cost; about US$7B Blue Owl cash contribution and about US$3B Meta distribution; development funding, not GPU price Meta Blue Owl — 2025-10-21 · Meta Louisiana update — 2026-07-13 COLOSSUS-MEMPHIS — xAI SpaceXAI Colossus North America · Memphis, Tennessee, US Operational full-scale cluster by first-party claim; expansion roadmap active No public MW capacity in the cited project pages ND 200,000 NVIDIA H100 GPUs deployed in one interconnected cluster; 1M is a roadmap xAI SpaceXAI; NVIDIA supplier xAI Series C named A16Z, BlackRock, Fidelity, Kingdom Holdings, Lightspeed, MGX, Morgan Stanley, OIA, QIA, Sequoia, Valor Equity Partners, Vy Capital, and strategic investors NVIDIA and AMD US$6B Series C corporate financing; not Colossus-specific capex or GPU price xAI Colossus — current page · xAI Memphis — current page · xAI Series C — 2024-12-23 DATAVOLT-OXAGON — DataVolt NEOM AI factory Middle East · Oxagon, Saudi Arabia Planned; phase one expected operational by 2028 1.5 GW factory; basis not identified as IT, site, or grid capacity ND Model and count unknown DataVolt developer, investor, and operator; NEOM Oxagon host and infrastructure partner DataVolt and NEOM Oxagon partnership US$5B initial phase-one development investment; not full-campus capex or GPU price NEOM — 2025-02-10 UAE-US-AI-CAMPUS — UAE–US AI Campus Stargate UAE Middle East · Abu Dhabi, UAE Announced planned; Phase 1 unveiled 2025-05-15; construction start ND 5 GW AI-data-center campus; secondary source reports a 1 GW first phase, not in the official page ND Model and count unknown; capacity reserved for hyperscalers and approved providers G42 builds; US companies and hyperscalers operate in partnership Project-level backing and financing ND; separate corporate investment context is not assigned to the campus No campus capex or lease price disclosed U.S. Embassy UAE — 2025-05-17 · G42 — 2025-05-22 · DCD — 2025-05-16 YOTTA-SHAKTI — Yotta Shakti Cloud Asia-Pacific · Navi Mumbai and Greater Noida, India Operational and expanding; H100 delivery reported in 2024 and platform expansion in 2025 D1: 30 MW IT power, 5,000 racks, 300,000 sq ft; NM1 and D1 are not merged into a single capacity total 30 MW IT power at D1 More than 4,000 NVIDIA H100 GPUs in first NM1 cluster; 16,384 32,768 roadmap figures are targets, not current D1 stock; B200 future deployment noted Yotta Data Services; NVIDIA technology cloud partner IndiaAI Mission Government of India partnership; financial investor and project-equity amount ND On-demand or reserved GPU procurement model; no numeric rate verified Yotta H100 delivery — 2024-03-19 · Yotta Shakti — 2025-05-19 · Yotta D1 — retrieved 2026-09-22 STARGATE-NORWAY — Nscale Aker OpenAI Europe · Kvandal near Narvik, Norway Planned pre-construction mobilization as of 2025-07-31 230 MW planned, with ambition for an additional 290 MW; basis not identified as IT versus utility connection ND 100,000 NVIDIA GPUs targeted by end-2026; model and installed state ND Nscale designs builds; planned 50 50 Nscale-Aker JV; OpenAI initial offtaker Nscale and Aker; OpenAI offtake option; later Microsoft customer context About US$1B initial 20 MW phase, including more than US$250M equity on a 100% basis; initial-phase funding, not full-campus cost or GPU price Nscale — 2025-07-31 · Nscale project context — 2025-08-14 AIRTRUNK-TOK1 — TOK1 hyperscale campus Asia-Pacific · East Tokyo, Japan Under construction phased expansion; 40+ MW phase broke ground 2025-06-19; later completion ND 300+ MW total campus and 40+ MW phase; separate from the later IT-load statement More than 100 MW of new IT load under construction in the 2026 financing source Model and count unknown AirTrunk developer operator 12 arranging lending banks named; equity investor not identified in cited project sources JP¥191.6B US$1.24B green loan for refinancing and continued development capex; not TOK1 total capex or GPU price; Japan-wide investment exceeds US$8B AirTrunk — 2025-06-19 · AirTrunk — 2026-03-10 YTL-AI-CLOUD-JOHOR — YTL Green Data Center Park YTL AI Cloud Asia-Pacific · Kulai, Johor, Malaysia Operational by secondary report dated 2025-11-03; official plan dated 2023-12-08; scope remains unresolved Official 500 MW facility; secondary 600 MW park observation; neither is IT load and neither is merged ND Official H100 plan; secondary report cites liquid-cooled GB200 NVL72; exact count unknown YTL Power International; YTL Communications owns manages AI infrastructure; NVIDIA partner Project-equity investor not disclosed; policy support is not treated as project backing RM10B secondary program investment, roughly split between data-center infrastructure and AI solutions; not facility-specific capex or GPU price YTL Power — 2023-12-08 · W.Media — 2025-11-03 STTGDC-SG — ST Telemedia Global Data Centres AI-ready facility Asia-Pacific · Singapore Proposed selected for development under DC-CFA2; implementation remains approval-dependent as of 2026-08-21 50 MW new data-center capacity; basis not identified as IT versus utility connection ND Model and count unknown; liquid-cooled AI-ready design ST Telemedia Global Data Centres; selected by Singapore EDB and IMDA Project investor backer and financial amount ND No public capex, lease, rental, or hardware price disclosed STT GDC — 2026-08-21

Status treatment

Planned pre-construction: DataVolt Oxagon, Stargate UAE, Stargate Norway, and STT GDC remain announcements, mobilization, or approval-dependent development signals. Their stated capacity is not operating capacity, and Nscale's GPU figure is a target rather than an installed count.

Under construction: Milam County, Hyperion, and TOK1 have source-supported build activity. Their project or campus figures do not establish full energization; TOK1 is the only record here with a directly stated additional IT-load figure, and it is a lower-bound expansion observation.

Operational expanding: Colossus, Yotta Shakti, and YTL have source-supported operating or service claims, but the evidence is not a common stocktake. Colossus and Yotta provide deployed H100 observations; YTL's operational status and GB200 NVL72 detail are secondary and its count remains unknown.

Capacity and electricity boundaries

The table below keeps site facility compute capacity, IT load, and electricity or grid evidence in separate columns. A voltage, power-source description, generation plan, or financing amount is not converted into an IT-load or interconnection number.

Project Source-defined site facility compute signal Explicit IT-load signal Electricity, grid, or generation evidence Boundary and limitation --- --- --- --- --- Milam County 1.2 GW data-center site ND New generation planned; MW not disclosed Site capacity is not promoted to IT or grid MW Hyperion 5 GW compute capacity ND Power, cooling, and connectivity included in development scope; MW not disclosed Compute capacity is not IT or utility capacity Colossus No public MW in cited project pages ND MLGW TVA coordination and temporary turbines referenced; MW not disclosed GPU count is not converted to power DataVolt Oxagon 1.5 GW factory ND Renewable-powered and net-zero design intent; MW not disclosed Factory figure is not an IT or connection measure Stargate UAE 5 GW AI-data-center campus; 1 GW first phase is secondary ND Nuclear, solar, and gas power sources named; MW not disclosed Power mix is not grid capacity or delivered generation Yotta D1 5,000 racks and 300,000 sq ft 30 MW IT power Dual feeders and 220 kV on-site substation; MW connection value ND Voltage is retained as voltage, not converted to MW Stargate Norway 230 MW planned plus 290 MW ambition ND Renewable hydropower; local transmission constraint noted; connection MW ND Planned capacity and hydropower are separate objects AirTrunk TOK1 300+ MW campus and 40+ MW phase More than 100 MW new IT load Grid interconnection MW ND Campus, phase, and incremental IT load are not additive YTL Johor 500 MW official facility; 600 MW secondary park signal ND Official on-site solar intent; secondary 500 MW solar and 600 MW national-grid backup not reconciled Dated observations remain separate and not comparable STT GDC Singapore 50 MW new data-center capacity ND More than 50% eligible green-energy pathways; grid generation MW ND Capacity basis is not defined as IT or grid

GPU inventory and specification table

Project inventory Model or specification as stated Quantity and state Company operator Investor backer Public price cost type Dated evidence --- --- --- --- --- --- --- xAI SpaceXAI Colossus NVIDIA H100 Hopper; H100 vendor reference is 80 GB HBM3 and 3.35 TB s 200,000 H100 deployed in one interconnected cluster; 1M roadmap is future xAI SpaceXAI; NVIDIA supplier xAI Series C investor list; project allocation of financing ND US$6B corporate Series C; not GPU unit price xAI Colossus — current page · NVIDIA H100 · xAI Series C — 2024-12-23 Yotta Shakti NM1 and D1 NVIDIA H100; future B200 deployment mentioned More than 4,000 H100 delivered to first NM1 cluster; 16,384 32,768 are roadmap targets; current D1 count ND Yotta Data Services; NVIDIA partner IndiaAI Mission Government of India context; financial backer ND On-demand reserved cloud procurement model; no numeric rate Yotta H100 — 2024-03-19 · Yotta Shakti — 2025-05-19 Nscale Aker Stargate Norway NVIDIA GPU; model not specified 100,000 targeted by end-2026 ; target, not deployed inventory Nscale Aker planned JV; OpenAI offtaker Nscale and Aker; OpenAI offtake option About US$1B initial-phase project funding; not GPU price Nscale — 2025-07-31 YTL AI Cloud Johor Official H100 plan; secondary liquid-cooled GB200 NVL72 report Model disclosed, count unknown; operating status secondary YTL Power YTL Communications; NVIDIA partner Project-equity investor ND RM10B AI-program investment; not GPU price YTL Power — 2023-12-08 · W.Media — 2025-11-03 Milam County Model and topology unknown Site-specific GPU count unknown; platform-wide chip figures excluded SB Energy OpenAI OpenAI, SoftBank, and Ares financing context; site allocation ND US$1B and US$800M platform financing; not GPU price OpenAI — 2026-01-09 Hyperion Louisiana Model and topology unknown GPU count unknown Meta JV Blue Owl funds 80%; Meta 20% Approximately US$27B JV development cost; not GPU price Meta Blue Owl — 2025-10-21 DataVolt Oxagon Model and topology unknown GPU count unknown DataVolt NEOM-Oxagon DataVolt and NEOM Oxagon partnership US$5B phase-one investment; not GPU price NEOM — 2025-02-10 Stargate UAE Model and topology unknown GPU count unknown G42 and US-partner consortium Project financing backers ND No public price U.S. Embassy UAE — 2025-05-17 AirTrunk TOK1 Model and topology unknown GPU count unknown; cloud and AI workloads supported AirTrunk Lending syndicate named; equity allocation ND JP¥191.6B US$1.24B green loan; not GPU price AirTrunk — 2026-03-10 STT GDC Singapore Model and topology unknown; liquid cooling and AI-ready design GPU count unknown STT GDC Investor backer ND No public price STT GDC — 2026-08-21

Public price and cost signals

Project entity Published signal Price type What it does and does not measure Dated evidence --- ---: --- --- --- Stargate SB Energy Milam US$1B OpenAI SoftBank investment plus US$800M Ares preferred equity Company platform financing Supports SB Energy Stargate financing context; not Milam site capex or lease price OpenAI — 2026-01-09 Meta Hyperion Approximately US$27B development cost; about US$7B Blue Owl cash contribution; about US$3B Meta distribution JV development and funding Buildings and long-lived power cooling connectivity scope; not GPU bill of materials or tariff Meta Blue Owl — 2025-10-21 xAI Series C US$6B Corporate financing xAI company financing; not Colossus-specific capex or GPU price xAI — 2024-12-23 DataVolt Oxagon US$5B Phase-one development investment Initial phase only; not full 1.5 GW campus capex or GPU price NEOM — 2025-02-10 Nscale Aker Stargate Norway About US$1B initial 20 MW phase, including more than US$250M equity Initial-phase project funding and equity Does not price the full 230 MW plan, 290 MW ambition, or GPU target Nscale — 2025-07-31 AirTrunk TOK1 JP¥191.6B US$1.24B Green loan for refinancing and continued development capex Financing scope is not total TOK1 construction cost or GPU price; Japan-wide investment exceeds US$8B AirTrunk — 2026-03-10 YTL AI program RM10B Program investment Secondary program figure, roughly data-center infrastructure plus AI solutions; not facility-specific capex W.Media — 2025-11-03

Analytical chart and dashboard handoff

The report retains the baseline's ten sourced quantitative chart blocks: same-unit IT-load comparisons, status-qualified GPU quantities, and hosted GPU rental-rate comparisons. The blocks remain non-additive and preserve their existing count, order, kinds, display options, and source notes. The current packet's safe chart-ready observations are the two deployed H100 counts above and the two explicit IT-load observations above; facility, compute, generation, grid, and target values are excluded from those like-for-like comparisons rather than estimated.

Handoff requirement Candidate state Evidence treatment --- --- --- Dashboard data must be tabular Met in this report candidate Project, electricity, GPU, and price records use explicit columns rather than raw unlabelled data Dashboard values must display correctly Met for this report candidate Units, status, lower-bound markers, and ND are retained beside values This run must update the dashboard Not evidenced in this writer session Dashboard changes are outside the allowed report result write scope; no dashboard update is claimed Dashboard and report must each contain multiple sourced analytical charts Report retained; dashboard not verified The report retains ten sourced chart blocks; the separate dashboard artifact is not in the accepted evidence or output scope

Caveats and unknowns

• Most gigawatt and 50–500 MW announcements do not identify whether the figure is IT load, facility power, compute capacity, utility capacity, or interconnection. They must not be added into a global MW total.
• Grid-connection and generation MW are not disclosed for Milam, Hyperion, Colossus, DataVolt, Abu Dhabi, Nscale Norway, TOK1, Yotta D1, or STT GDC. Source descriptions of hydropower, solar, nuclear, gas, or green-energy pathways do not establish a numeric connection or delivered-output figure.
• GPU counts and specifications are missing for most sites. The 200,000 xAI H100 and more-than-4,000 Yotta H100 observations are deployed claims; Nscale's 100,000 is a target; Yotta's 16,384 32,768 figures are roadmap targets; YTL's GB200 NVL72 detail is secondary and count-unknown.
• No GPU count is inferred from rack counts, capacity, topology, customer allocation, or cloud orderability. No MW-to-GPU or GPU-to-MW conversion is made.
• Public financial signals answer different questions: corporate financing, JV development cost, phase-one investment, green-loan refinancing, program investment, and cloud procurement model are retained by type and are not summed.
• YTL's official 500 MW facility description and secondary 600 MW park observation are dated and not reconciled. Yotta's roadmap and current D1 deployment are also unresolved.
• Operational claims, targets, and construction states are bounded by their cited evidence dates. Current completion, actual delivery, exact model mix, site equity, project capex, and OEM purchase prices remain unknown where the sources do not disclose them.

Latest evidence update — 23 September 2026

This current-run update adds the accepted 23 September 2026 evidence packet as an incremental, table-first register. It covers 16 named projects, expansions, and program pipelines across North America, Europe, the Middle East, and Asia. Each row retains its source date, measurement class, electricity boundary, GPU state, company and backer roles, and price or cost type. ND means not disclosed or not verified in the reviewed evidence, not zero. Parent programs, child sites, platform targets, generation, grid access, IT load, compute capacity, and financial objects are not added together.

Project, status, capacity class, and electricity register

ID project Geography and status at evidence date Capacity class and value as stated Electricity grid evidence Direct evidence --- --- --- --- --- STG-TX-MILAM — Stargate SB Energy Milam County site Milam County, Texas, US · Under construction service preparation; OpenAI said initial facilities were under construction and expected to enter service from 2026; source dated 9 Jan 2026 1.2 GW data-center site capacity ; IT load versus utility connection basis ND New generation planned to support site energy needs; generation and grid-connection MW ND OpenAI, 9 Jan 2026 HYPERION-LA — Meta Hyperion campus Richland Parish, Louisiana, US · Construction well underway; 5 GW expansion reported 13 Jul 2026 5 GW compute capacity ; not labelled IT load or grid capacity Long-lived power, cooling, and connectivity included in development scope; separate grid utility MW ND Meta, 13 Jul 2026 · JV announcement COLOSSUS-MEMPHIS — xAI SpaceXAI Colossus Memphis, Tennessee, US · 200,000-GPU cluster reported as operating; expansion roadmap remains active; Memphis page updated 30 Jul 2026 No public MW in cited first-party pages; GPU count is not converted to power MLGW TVA coordination and temporary turbines referenced; durable connection and MW ND Colossus · Memphis DATAVOLT-OXAGON — DataVolt NEOM AI factory Oxagon, Saudi Arabia · Planned; phase one expected operational by 2028; announcement dated 10 Feb 2025 1.5 GW factory ; IT, site, and grid basis ND Renewable-powered and net-zero design intent; numeric generation and grid capacity ND NEOM, 10 Feb 2025 UAE-US-AI-CAMPUS — UAE–US AI Campus Stargate UAE Abu Dhabi, UAE · Announced planned; Phase 1 unveiled 15 May 2025; construction start ND 5 GW AI-data-center campus ; secondary report gives a 1 GW first phase, kept separate Nuclear, solar, and gas power named; numeric generation and grid-connection capacity ND U.S. Embassy · DCD, secondary YOTTA-SHAKTI — Yotta Shakti Cloud Navi Mumbai and Greater Noida, India · Operational and expanding; H100 delivery reported 2024 and platform expansion in 2025 D1 reports 30 MW IT power , 5,000 racks, and 300,000 sq ft; NM1 and D1 are not merged Dual feeders and a 220 kV on-site substation ; MW connection value ND Yotta D1 · Yotta H100 delivery META-JAMNAGAR — Meta Reliance AI-enabled data center Jamnagar, Gujarat, India · Agreed planned as of 9 Jun 2026; construction and service date ND 168 MW first phase , with option to scale; IT versus facility utility basis ND Renewable power and desalinated-water cooling described; separate nearly 1 GW renewable backing is not site grid capacity Meta, 9 Jun 2026 AIRTRUNK-MALAYSIA — JHB3 and JHB4 Iskandar Puteri, Johor, Malaysia · Planned expansion announced 30 Apr 2026 More than 280 MW IT load ; four-campus Malaysia platform later exceeds 700 MW IT load and is not additive Recycled-water cooling and energy-efficiency priorities; numeric grid and generation capacity ND AirTrunk, 30 Apr 2026 AIRTRUNK-INDIA-PIPELINE — AirTrunk India program India · Proposed investment pipeline announced 5 Jun 2026 after Lumina CloudInfra entry More than 5 GW program target by 2030; 600 MW existing pipeline across Mumbai, Chennai, and Hyderabad; not one site or IT load Reliable power, renewables, and sustainable water are planning requirements; site grid generation MW ND AirTrunk, 5 Jun 2026 STARGATE-NORWAY — Nscale Aker OpenAI Narvik Kvandal, Norway · Development buildout active in 2026; commissioning not evidenced 230 MW planned , plus 290 MW ambition and separate 115 MW financing expansion; basis not identified; non-additive Renewable hydropower; local transmission constraint noted; connection MW ND Nscale Norway · 2026 financing AIRTRUNK-TOK1 — TOK1 hyperscale campus East Tokyo, Japan · Under construction phased expansion; 40+ MW phase groundbroken 19 Jun 2025; current page accessed 23 Sep 2026 300+ MW campus and more than 100 MW new IT load are separate observations; not additive 66 kV high-voltage feeds and design PUE 1.15; MW grid connection ND TOK1 location · AirTrunk financing YTL-AI-CLOUD-JOHOR — YTL Green Data Center Park AI Cloud Kulai, Johor, Malaysia · Operational status reported by secondary source 3 Nov 2025; scope unresolved Official 500 MW facility and secondary 600 MW park observation; neither is IT load and neither is merged Official on-site solar intent; secondary 500 MW solar and 600 MW grid-backup claims not reconciled YTL Power · W.Media, secondary STTGDC-SG — ST Telemedia Global Data Centres Singapore Singapore · Proposed selected under DC-CFA2; approval-dependent as of 21 Aug 2026 50 MW new data-center capacity ; IT versus utility connection basis ND More than 50% eligible green-energy pathways intended; grid and generation MW ND STT GDC, 21 Aug 2026 STTGDC-JAKARTA — STT GDC Jakarta expansion Greater Jakarta, Indonesia · Expansion under way as of 10 Jun 2026; Jakarta 2 launched, Jakarta 3 topped out, Jakarta 5 6 groundbroken More than 360 MW AI-ready IT capacity ; source-defined IT capacity, not promoted to current IT load Secured power described; numeric grid, generation, and connection capacity ND STT GDC, 10 Jun 2026 META-STURGEON-AB — Meta Sturgeon County data center Sturgeon County, Alberta, Canada · Under construction; groundbreaking 8 Jul 2026; completion ND 1 GW data-center capacity ; IT, compute, and grid basis ND Meta funds new generation and grid infrastructure with named energy grid parties; 100% clean and renewable matching; numeric MW ND Meta, 8 Jul 2026 META-EL-PASO — Meta Compute BlackRock El Paso campus El Paso, Texas, US · Under construction as of 28 Jul 2026; online target 2028 1 GW compute capacity ; not labelled IT load or grid capacity Numeric grid and generation capacity ND Meta, 28 Jul 2026 Platform context — not allocated to named sites OpenAI reported a 10 GW Stargate milestone by 2029; AWS NVIDIA plan 2M additional GPUs in 2027–2028; Meta NVIDIA describe millions of Blackwell Rubin GPUs; NVIDIA IREN describe up to 5 GW of AI infrastructure Platform-level targets or pipeline; no site allocation or common capacity class Grid, generation, and site boundaries ND at platform scope OpenAI · AWS NVIDIA · Meta NVIDIA · NVIDIA IREN

GPU model, count, state, roles, backing, and price register

ID project GPU model, count, and state Company operator roles Investors backers Price or cost type Direct evidence --- --- --- --- --- --- STG-TX-MILAM Site-specific GPU model and count ND ; OpenAI platform-wide chip figures excluded SB Energy build operate role; OpenAI customer partner OpenAI and SoftBank each invested US$500M in SB Energy; Ares supplied US$800M preferred equity; site allocation ND 1.2 GW lease value ND; financing is platform company context, not site capex OpenAI HYPERION-LA GPU model and count ND Meta plus JV develops owns; Meta construction property management Blue Owl-managed funds 80% and Meta 20% Approx. US$27B JV development cost; approx. US$7B Blue Owl cash and US$3B Meta distribution; not GPU price Meta JV announcement COLOSSUS-MEMPHIS 200,000 NVIDIA H100 deployed in one cluster; 1M roadmap is future and excluded xAI SpaceXAI operator customer; NVIDIA OEM Series C names A16Z, BlackRock, Fidelity, Kingdom Holdings, Lightspeed, MGX, Morgan Stanley, OIA, QIA, Sequoia, Valor Equity Partners, Vy Capital, NVIDIA, and AMD US$6B Series C corporate financing; not Colossus capex or GPU price xAI Colossus · Series C DATAVOLT-OXAGON GPU model and count ND DataVolt developer, investor, and operator; NEOM Oxagon land and infrastructure support DataVolt initial-phase investment; no other financial backer named US$5B phase-one development investment; not full-campus capex or GPU price NEOM UAE-US-AI-CAMPUS GPU model and count ND ; capacity reserved for hyperscalers and approved providers G42 builds; US companies and hyperscalers operate in partnership Project-level backing and financing ND ; separate corporate investment context excluded Campus capex, lease, and GPU price ND U.S. Embassy YOTTA-SHAKTI More than 4,000 NVIDIA H100 delivered to the first NM1 cluster; current D1 count ND; future B200 and 16,384 32,768 roadmap figures are not current inventory Yotta Data Services operator; NVIDIA technology cloud partner IndiaAI Mission Government of India partnership context; financial investor and project equity ND On-demand or reserved GPU procurement model; numeric rate and hardware price ND Yotta H100 · Yotta Shakti META-JAMNAGAR GPU model and count ND Reliance builds; Meta leases first phase and covers supporting energy and water costs Project equity investor ND; Meta's separate US$5.7B Jio Platforms investment is not assigned Lease value and site capex ND ; no GPU price Meta AIRTRUNK-MALAYSIA GPU model and count ND AirTrunk developer operator Project equity investor ND; 30-institution lending consortium relates to JHB2, not JHB3 JHB4 equity MYR12B US$3B planned investment for JHB3 JHB4; separate JHB2 green financing not added AirTrunk Malaysia AIRTRUNK-INDIA-PIPELINE GPU model and count ND AirTrunk; Lumina CloudInfra acquired entry platform AirTrunk backed by Blackstone and CPPIB; site allocation ND More than INR3,000B US$30B proposed program investment by 2030; not site capex AirTrunk India STARGATE-NORWAY 100,000 NVIDIA GPUs targeted by end-2026 ; separate more than 30,000 NVIDIA Rubin GPUs planned for Microsoft in 2027; neither is deployed inventory or additive Nscale manages the project; Aker was original JV partner; OpenAI initial offtaker; Microsoft customer Initial Nscale Aker commitment and later lenders ABN AMRO, DNB, Eksfin, Nordea, and SEB; corporate financing context remains separate Approx. US$1B initial 20 MW phase; US$790M additional financing and uncommitted accordion; not GPU price Nscale · Microsoft plan · Financing AIRTRUNK-TOK1 GPU model and count ND ; cloud and AI workloads supported AirTrunk developer operator Twelve arranging lending banks named; equity allocation ND JP¥191.6B US$1.24B green loan for refinancing and development capex; not total construction cost or GPU price AirTrunk financing YTL-AI-CLOUD-JOHOR Official H100 plan; secondary report cites liquid-cooled GB200 NVL72 ; exact count unknown YTL Power International; YTL Communications owns manages AI infrastructure; NVIDIA partner Project-equity investor ND; policy support is not project equity RM10B secondary AI-program investment; not facility-specific capex or GPU price YTL Power · W.Media STTGDC-SG GPU model and count ND ; liquid-cooled AI-ready design STT GDC; selected by Singapore EDB and IMDA Project investor backer and financing amount ND Capex, lease, rental, and hardware price ND STT GDC Singapore STTGDC-JAKARTA GPU model and count ND ; AI-ready colocation capacity is not an inventory count STT GDC campus developer operator Project investor backer and financial amount ND; policy support not project equity Site capex, lease, rental, and hardware price ND STT GDC Jakarta META-STURGEON-AB GPU model and count ND Meta designs operates; energy and grid parties support planning Project equity investor ND More than CAD13B data-center investment plus approx. CAD60M local infrastructure; not a GPU price Meta Alberta META-EL-PASO GPU model and count ND ; Meta NVIDIA platform GPU commitments not allocated Meta and BlackRock venture to develop own; Meta construction and property management; Meta initial sole occupant BlackRock with GIP and HPS; venture financing amount ND Meta investment over US$10B ; not combined with other Meta projects or GPU price Meta El Paso

Platform commitments kept outside site and inventory totals

Platform signal Current evidence Boundary retained Direct evidence --- --- --- --- OpenAI Stargate More than 10 GW committed by 2029 and more than 3 GW added in the prior 90 days as of 29 Apr 2026 Platform capacity; site allocation, IT basis, grid MW, and GPU count ND OpenAI AWS NVIDIA 2 million additional NVIDIA GPUs planned across AWS infrastructure in 2027–2028; separate 100,000-GPU secure build Future platform deployment; not current installed inventory or named-site capacity AWS NVIDIA Meta NVIDIA Millions of Blackwell and Rubin GPUs, including GB300-based systems, in a large-scale partnership Platform plan; no site allocation, current deployed count, or project price NVIDIA Meta NVIDIA IREN Up to 5 GW of AI infrastructure and a 2 GW Sweetwater flagship expectation; potential right to buy 30M shares at $70, up to $2.1B Pipeline and conditional equity right; not operating capacity, GPU inventory, or site capex NVIDIA IREN

Dashboard-ready field discipline and safe comparisons

Field Current-run treatment Display rule --- --- --- status phase Planned, construction, operational, expansion, or preparation only when the source supports it Keep source date beside the state; do not promote a target to commissioning capacity class it load , it capacity , facility capacity , compute capacity , grid or interconnection , generation or ppa , or program capacity Show the class in the column label; do not relabel unspecified MW as IT load capacity value Original unit and lower-bound marker retained, including 280 MW , 100 MW , and 360 MW Do not convert MVA, voltage, energy, or generation into site IT MW electricity evidence Power source, substation, voltage, secured power, generation, or grid statement kept separate from capacity Renewable, solar, nuclear, gas, or hydropower language is not a numeric connection value gpu count and gpu state xAI 200,000 H100 deployed; Yotta more than 4,000 H100 delivered; Nscale 100,000 target and 30,000+ Rubin 2027 plan remain separate Never sum deployed, delivered, target, procured, orderable, or platform quantities gpu specification H100, GB200 NVL72, GB300 NVL72, B200, B300, Rubin, or ND exactly as sourced A rack or system specification is not a site inventory count company operator and investors backers Developer, owner, operator, customer, lender, JV, government, and platform roles are kept distinct Do not infer equity ownership from a customer, supplier, utility, policy, or lender role price signal and price type Financing, development cost, program investment, lease, revenue, backlog, share-purchase right, and hosted rate remain original economic objects Do not sum or present any one as a comparable GPU purchase price source date and access date Publication date retained; 23 Sep 2026 is the current evidence-access date for fresh pages Access date does not change publication date or prove operation

The two retained analytical chart slots in the 23 September packet show only safe comparisons from that packet: source-labelled IT-load observations and status-qualified H100 observations. The three IT-load values are 30 MW, more than 100 MW, and more than 280 MW; the two H100 bars are xAI's 200,000 deployed claim and Yotta's more-than-4,000 delivered lower bound. Facility, compute, grid, generation, PPA, platform, target, and hosted-rental values are excluded from those like-for-like views rather than estimated. They remain dated historical context for continuity; the 24 September packet does not repeat those exact item inventories. The report retains ten quantitative chart blocks in the baseline order and with the baseline bar-chart display options.

Evidence limitations and unresolved risks

Risk Retained treatment Direct evidence and implication --- --- --- Measurement-class risk Most gigawatt and 50–1,000 MW figures are facility, compute, program, or unspecified capacity; STT Jakarta is source-defined IT capacity, not explicit IT load Meta Jamnagar · STT GDC Jakarta · AirTrunk India support separate source-defined treatment Lower-bound risk More-than observations remain visibly lower bounds; chart anchors are not presented as exact totals AirTrunk Malaysia · AirTrunk TOK1 · Yotta D1 GPU-state risk xAI's 200,000 is a first-party operating claim; Yotta's more-than-4,000 is a delivered-cluster observation; Nscale, AWS, and platform counts are targets or plans xAI Colossus · Yotta H100 Electricity risk Power-source descriptions and voltage do not establish connection MW, generation nameplate, or delivered hourly energy Meta Alberta · AirTrunk TOK1 Ownership and financing risk Company, customer, operator, lender, government, and strategic-partner roles remain separate; site equity is ND where not disclosed Nscale financing · Meta El Paso Price comparability risk Development cost, financing, expected revenue, program investment, backlog, conditional equity, cloud rental, and hardware price are different objects; OEM purchase prices remain ND NVIDIA IREN · YTL secondary report Completion risk Current completion, installed counts, exact model mix, site equity, project capex, and grid MW remain unresolved for several announced or expanding projects STT GDC Jakarta · Meta El Paso

Latest evidence update — 24 September 2026

The accepted 24 September packet is an incremental, source-qualified register: 10 project rows across four regions and 12 accelerator or platform records. The structured section above is the reader-facing table view. It retains status, source-defined capacity class, electricity or grid evidence, company role, explicitly named backing, GPU state, technical specification, price type, and as-of date in separate fields. ND means not disclosed or not verified, not zero.

The strongest project signals are Crusoe's 1,200 MW facility-power design in Abilene, Meta Hyperion's 5,000 MW compute-capacity statement, Norway's 230 MW initial plan plus a 290 MW expansion path, HUMAIN's projected 500 MW AI-factory capacity, and Kao Data's 71 MW ITE-load campus figure. These are different measurement classes and states; they are not summed. Electricity evidence is similarly bounded: a 1,200 MW interconnection signal, a 150 MW substation, a 40 MW solar PPA, a 30 MW renewable virtual PPA, and up to 500 MW of on-site solar are not interchangeable with delivered IT load.

The clearest current accelerator evidence is xAI's first-party 200,000-H100 operating claim, Nscale's 100,000-GPU target, HUMAIN's 18,000-GB300 first phase, OCI's H100 H200 B200 service maxima, and documented TPU pod limits. The current price register includes hosted benchmarks such as Lambda H100 at US$3.99 GPU-hour, RunPod H200 at US$3.59–4.59 hour, derived Azure MI300X at US$6 GPU-hour, MI325X at US$2.00–3.80 GPU-hour, TPU v5p at US$1.89–4.20 chip-hour, A100 at US$2.79 GPU-hour, and a grouped L40S listing at US$1.75 hour. The Intel Gaudi 3 system estimate is retained separately from hosted rental. None is an OEM purchase quote or a global inventory total.

The canonical report retains two evidence-backed analytical families in its existing ten chart fences: source-labelled capacity electricity or IT-load comparisons, and accelerator quantity or hosted-rate comparisons. The baseline chart contract requires the same count, order, kinds, and display options; the 24 September packet does not repeat the prior current chart item inventories, so no stale observation is relabeled as current and no replacement or extra chart slot is introduced. Current packet values remain available in the structured tables with their measurement classes and source links.

Primary evidence is linked beside each current row, including Crusoe, Meta, xAI, Nscale, Kao Data, NVIDIA HUMAIN, Khazna, YTL, AirTrunk, Google, Oracle, Google Cloud TPU, AWS Project Rainier, Intel Gaudi, Lambda, and RunPod.

Unresolved risks remain completion and delivery verification, mixed IT facility compute grid boundaries, ambiguous units such as systems versus GPUs, historical rather than current deployment claims, undisclosed site equity and financing allocation, dynamic hosted rates, and the absence of procurement-grade OEM prices. No dashboard state was changed in this report-writing session.

Latest evidence update — 25 September 2026

The accepted 25 September packet is a bounded, source-qualified register: 12 data-center or AI-factory projects and 10 GPU, accelerator, cluster, or platform records. It does not support a global total. IT load, facility or compute capacity, grid interconnection, generation PPA, installed or planned GPU state, company roles, backing, financing, and price objects remain separate. ND means not disclosed or not verified, not zero.

The strongest current project signals are Meta Richland’s 5,000 MW compute-capacity statement, Crusoe Abilene’s 900 MW new site and planned 900 MW on-site plant, SINES’ 1.2 GW secured campus IT capacity, Pure DC’s 110 MW phase-1 IT load, and the 100 MW tranche within DataVolt’s 360 MW first phase. These are unlike boundaries and states. The named inventory signals include approximately 1,100 operating Blackwell GPUs at Sakura, 4,696 acquired B200 GPUs at DataSection, 1,632 purchased B300 GPUs plus 192 RTX PRO 6000 at QumulusAI, 27,500 planned Rubin GPUs in Japan, 41,000 additional GB300 GPUs authorized for delivery at Firebird, and more than 66,000 Rubin GPUs planned at SINES Nscale. They are not additive.

Data-center plans, construction, power, and ownership

ID Region project and location Status and source as-of IT load Facility compute capacity Grid, contracted, renewable or on-site power Company backers Price cost signal Evidence --- --- --- ---: ---: --- --- --- --- DC-01 Meta Richland Parish Data Center Hyperion, Louisiana, US Under construction ; Meta says it has broken ground and is expanding the site. Source 2026-07-13; read 2026-09-25 ND 5,000 MW compute capacity ; not labelled IT load Entergy energy agreement; current source gives savings but no MW rating Meta operator developer; DPR Construction, Turner Construction and Mortenson named as general contractors; outside equity split ND More than US$50B Meta investment in the Richland Parish region; investment, not capex-by-MW or GPU price Meta Richland Parish update DC-02 Crusoe Microsoft Abilene AI factory campus, Texas, US Under construction site preparation for the new campus. Source 2026-03-27; read 2026-09-25 ND 900 MW new site; adjacent full Abilene site projected at 2.1 GW ; existing phases are separate scopes New campus includes a planned 900 MW behind-the-meter on-site power plant ; older Abilene phase has a 1.2 GW grid interconnection Crusoe developer operator; Microsoft supported workload tenant; Lancium owns operates the broader energy-campus model; financing ND ND Crusoe 900 MW announcement; Crusoe 1.2 GW phase DC-03 Applied Digital Polaris Forge 2, Harwood, North Dakota, US Under construction status carried from the prior accepted project read; current financing read 2026-03-04 and reread 2026-09-25 200 MW critical IT load 200 MW critical IT funded by the notes; an additional 800 MW right of first refusal was previously reported, not current load Utility interconnection rating ND in the current financing release Applied Digital APLD ComputeCo 2; qualified institutional buyers; Applied completion guarantee; Macquarie context in prior project evidence US$2.15B senior secured notes, 6.750%, issue price 98%; financing, not audited construction cost or GPU price Applied Digital notes release DC-04 ODATA QR03, Querétaro, Mexico Under construction first energy phase energized in the last successful primary read; current reread on 2026-09-25 returned a 404 concurrency page, so status is not revalidated in this pass 72 MW first building; 300 MW full-build IT 300 MW five-building full-build IT capacity 200 MW energized first energy phase ; 400 kV switching substation, 6.2 miles transmission lines, two substations, and a path to 400 MW energy infrastructure were reported ODATA, an Aligned Data Centers company; Ammper energy partner; anchor-client names ND More than US$3B about US$3.3B reported campus investment; scope differs by release and is not a GPU price ODATA 300 MW; ODATA energization DC-05 Scala AI City, Eldorado do Sul Porto Alegre metro, Brazil Announced permitted power connection ; construction and energization ND. Source dates 2024-09-11 and 2025-05-14; read 2026-09-25 54 MW initial IT capacity 4.75 GW source-described fully built campus capacity; not relabelled IT Brazil MME authorized a 5 GW ramp-up connection to the National Interconnected System; long-term wind hydro PPAs cited with no tariff or strike price Scala Data Centers developer operator; Rio Grande do Sul government Sedec strategic partner; lender equity split ND US$500M initial first-phase investment; a source quote mentions up to US$90B when fully realized, treated as aspirational and not committed Scala initial agreement; Scala 5 GW connection DC-06 SINES Data Campus, Sines, Portugal Operational SIN01; campus expansion planned in construction sequence . Source dates 2024-08-08 and 2025-04-04; read 2026-09-25 26 MW SIN01 operating facility ; campus IT capacity is separate 1.2 GW secured campus IT capacity ; SIN02–06 each up to 240 MW in the original plan Fully secured grid power for the campus; 100% renewable-energy design and seawater cooling; exact interconnection rating ND Start Campus developer operator; REN Portuguese transmission-system operator; Davidson Kempner Capital Management and Pioneer Point Partners private funders shareholders €8.5B combined privately funded construction investment value; not tariff or GPU price SINES 1.2 GW; SIN01 inauguration DC-07 Pure DC SJK01, Seinäjoki, Finland Permitted early development ; first-hall substation constructed and live, planning permissions and phase-1 power requirements secured. Source 2026-07-14; read 2026-09-25 110 MW phase-1 IT 550 MW-plus full-campus IT pathway; future pathway subject to permissions and contracting More than 700 MVA renewable-power access; MVA retained as reported and not converted to MW Pure Data Centres Group developer operator; SDC Ventures local partner; Seinäjoki municipality host; financing split ND More than €1.5B phase 1 and more than €7.5B full-campus investment pathways Pure DC SJK01 DC-08 DataVolt HUMAIN AI data center, Oxagon, NEOM, Saudi Arabia Under construction as of 2026-08-31 ND 100 MW tranche of a 360 MW first phase; wider planned Oxagon campus 1.5 GW ; source does not label these IT Renewable-energy integration and advanced cooling planned; utility import, generation, storage and interconnect ratings ND DataVolt developer operator and investor; HUMAIN, a PIF company, joint-development partner; NEOM Oxagon land and infrastructure support US$5B initial DataVolt investment for first phase under the 2025 agreement; phase scope, not GPU price NEOM HUMAIN 100 MW; DataVolt NEOM agreement DC-09 iXAfrica NBOX1, Mombasa Road, Nairobi, Kenya Operational NBOX1.1; expansion planned . Sources 2025-02-26 and undated campus specs; read 2026-09-25 4.5 MW live NBOX1.1; 18 MW NBOX1.2; 22.5 MW full-site IT Full-site and phase figures are nested; not additive Initial 2×4 MVA through 2×11 kV feeders; planned 2×40 MVA 66 kV substation; four 11 kV and two 66 kV utility feeds; MVA not converted iXAfrica operator developer; Future-tech design partner; Helios Investment Partners committed US$50M growth capital ; KenInvest government support US$50M growth-capital commitment; not necessarily total site capex, tariff or GPU price iXAfrica campus specifications; iXAfrica 4.5 MW DC-10 SoftBank IDC Frontier Hokkaido Tomakomai AI Data Center, Japan Under construction ; groundbreaking 2025-04-15. Source 2025-05-01; read 2026-09-25 ND Expected power capacity above 300 MW ; source does not label it IT Renewable energy procured locally within Hokkaido planned; utility import and generation ratings ND SoftBank Corp. and IDC Frontier developer operator; Hokkaido government and METI public-partnership subsidy roles; subsidy amount ND Project cost and GPU price ND SoftBank Tomakomai DC-11 STT Jakarta campus, Greater Jakarta, Indonesia Mixed phases: Jakarta 2 launched, Jakarta 3 topped out, Jakarta 5 6 groundbreaking. Source 2026-06-10; read 2026-09-25 Source reports pipeline as AI-ready IT capacity but no phase split More than 360 MW AI-ready sustainable IT-capacity pipeline ; aggregate pipeline, not live load Backed by secured power; high-voltage substation rating, import MW, PPA tariff and generation ND ST Telemedia Global Data Centres developer operator; Indonesian BKPM and Singapore government partnership context; financing amount ND Green financing mentioned but amount and allocation ND STT GDC Jakarta DC-12 Firebird AI factory, Hrazdan, Armenia Operational launch; expansion roadmap planned . Sources 2026-02-10 and 2026-08-08; read 2026-09-25 ND 300 MW AI-infrastructure capacity targeted by end-2027 ; not labelled IT Grid interconnect generation MW ND Firebird developer operator; NVIDIA intended investment; CoreWeave earlier investment; Armenian and US government export-licensing participation; amounts ND US$4B Phase 2 and prior US$500M Phase 1 project scopes; not GPU purchase price Firebird Phase 2; Firebird Hrazdan launch

GPU inventory, procurement, and cluster register

ID Region operator or project Model cited specification Quantity and state Company backer Price type, amount, and date Evidence and qualification --- --- --- --- --- --- --- GPU-01 Sakura internet, Ishikari, Japan NVIDIA Blackwell; exact SKU not disclosed Approximately 1,100 discrete GPUs; operating from 2026-02-25 Sakura internet; financing backers not stated Hardware price ND; cloud service pricing separate Sakura release GPU-02 DataSection Bangkok-area AI data center, Thailand NVIDIA B200; official release names 587 servers and 4,696 GPUs 4,696 B200 GPUs; acquired for planned deployment ; live status not independently confirmed DataSection operator; ASRock Rack supplier; unnamed US technology customer end user Official hardware price ND. US$257.4M estimated purchase agreement in DCD, 2026-06-05; secondary estimate; customer advances and US$180M borrowings are financing signals, not GPU price DataSection official release; DCD estimate GPU-03 QumulusAI distributed AI cloud, US sites NVIDIA Blackwell B300; plus RTX PRO 6000 Blackwell in the same order 1,632 B300 GPUs across 204 HGX B300 systems purchased , plus 192 RTX PRO 6000 ; company-level procurement, not a site stocktake QumulusAI; Technology Finance Corporation and USD.ai primary financing arrangements Hardware price ND. US$18M two-year take-or-pay customer contract activated for live Philadelphia capacity on 2026-09-01; contract revenue signal, not hardware price QumulusAI purchase; Philadelphia deployment GPU-04 Noetra NVIDIA national AI factory, Japan; site undisclosed NVIDIA Vera Rubin NVL72; Rubin GPUs and Vera CPUs 27,500 Rubin GPUs planned , with 13,750 Vera CPUs; not delivered or installed Noetra establishes project; NVIDIA technology partner; METI FRONTia support GPU price, project capex and financing ND; associated 140 MW data-center capacity is not a GPU power conversion NVIDIA Japan national AI infrastructure GPU-05 Firebird Armenian AI factory, Hrazdan NVIDIA GB300 Blackwell Ultra; later roadmap includes Rubin and Blackwell 41,000 additional GB300 GPUs authorized for sale delivery in Phase 2; separate roadmap more than 70,000 Rubin + Blackwell GPUs by end-2027 ; neither is installed stock Firebird; US export licensing; NVIDIA intended investment; CoreWeave earlier investment US$4B Phase 2 and US$500M Phase 1 project scopes; no disclosed GPU unit price Firebird Phase 2; Firebird launch GPU-06 SINES Nscale, Portugal NVIDIA Vera Rubin NVL72 More than 66,000 planned from late 2027; lower-bound expression and future target, not delivered stock Start Campus site; Nscale announced partner customer; financing price ND; SINES shareholders separately listed in DC-06 GPU price ND; campus €8.5B is separate project investment Start Campus Nscale AI announcement GPU-07 LUMI-AI supercomputer, Kajaani, Finland AMD Instinct MI430X plus 6th Gen AMD EPYC 256-core CPUs GPU count ND; planned deployment H2 2027 EuroHPC JU-owned; CSC host; Bull delivery; six-country LUMI AI Factory consortium €387.8M system contract including acquisition, delivery, installation and maintenance; funding shared equally by EuroHPC JU and the consortium; not GPU-only price AMD LUMI-AI announcement GPU-08 Crusoe Abilene, Texas NVIDIA GB200 NVL72 rack-scale system Up to 50,000 GB200 NVL72 systems per building; design capacity, not installed GPUs Crusoe Lancium; Oracle workload context in the live-campus source; procurement price ND No hardware price disclosed; cloud-rental benchmarks are separate Crusoe 1.2 GW phase; NVIDIA GB200 specification GPU-09 QumulusAI Philadelphia colocation site, US NVIDIA Blackwell B300 capacity Live in production after customer handoff on 2026-09-01; unit count ND . Do not map the company-level 1,632 purchase to this site without direct evidence QumulusAI; unnamed GPU cloud marketplace customer; colocation landlord ND US$18M two-year take-or-pay contract , announced 2026-09-10; customer contract revenue signal, not hardware price or site capex QumulusAI deployment GPU-10 Australia NVIDIA DSX ecosystem NVIDIA DSX platform; specific GPU model and per-partner unit count not disclosed in the primary source Up to 2 GW buildout by 2027 across Firmus, Sharon AI, IREN, Megaport, ResetData, CDC, NEXTDC and AirTrunk; program capacity, not a GPU inventory NVIDIA platform supplier; named Australian cloud infrastructure partners operate factories; investment and financing split ND GPU price and project investment ND; 2 GW is capacity program signal, not IT-load total or GPU count NVIDIA Australia ecosystem release

Product and system specification anchors

These are manufacturer reference specifications for interpretation. They are not site inventories or project power draws.

ID Product Source-backed specification Boundary qualification Source --- --- --- --- --- SPEC-01 NVIDIA DGX B200 8 NVIDIA Blackwell GPUs; 1,440 GB total GPU memory; 64 TB s HBM3e bandwidth; 144 PFLOPS FP4 Tensor Core; approximately 14.3 kW maximum system power; up to 400 Gb s networking Reference system only; do not assign these values to every B200 deployment NVIDIA DGX B200 SPEC-02 NVIDIA GB200 NVL72 72 Blackwell GPUs and 36 Grace CPUs in one liquid-cooled rack; 130 TB s low-latency GPU communications; 1.8 TB s GPU-to-GPU interconnect Rack system reference; Crusoe’s 50,000 figure is systems per building, not discrete GPUs NVIDIA GB200 NVL72 SPEC-03 NVIDIA GB300 NVL72 72 Blackwell Ultra GPUs and 36 Grace CPUs; fully liquid-cooled; 1.5× denser FP4 Tensor Core FLOPS and 2× attention performance versus Blackwell; ConnectX-8 provides 800 Gb s per GPU Vendor performance comparisons are projected; no site load inferred NVIDIA GB300 NVL72 SPEC-04 NVIDIA Vera Rubin NVL72 72 Rubin GPUs and 36 Vera CPUs; ConnectX-9 SuperNICs; BlueField-4 DPUs; NVLink 6 switch NVIDIA’s up-to-10× tokens MW and one-tenth cost-per-million-token claims are projected comparison metrics, not market prices NVIDIA Vera Rubin NVL72 SPEC-05 AMD Instinct MI430X MI400 family sovereign-AI HPC GPU; HBM4 memory; up to 288 TFLOPS hardware-based FP64 in AMD’s portfolio description No LUMI-AI GPU count or system power is disclosed in the source AMD MI400 announcement

Price and economics register

Subject Value Price type date What it does and does not mean Evidence --- ---: --- --- --- DataSection Thailand B200 order US$257.4M Secondary estimate; 2026-06-05 Estimated purchase-agreement value for 4,696 B200 GPUs; official DataSection page confirms units but not price. Do not divide into US$ GPU because taxes, delivery, services and scope are unconfirmed DCD; official DataSection release Applied Digital Polaris Forge 2 US$2.15B ; 6.750% notes; issue price 98% Financing; 2026-03-04 Senior secured notes intended to fund 200 MW critical IT development construction; not a quoted capex total, electricity price or GPU price Applied Digital SINES Data Campus €8.5B Privately funded construction investment scope; source 2025-04-04 Project investment; Davidson Kempner and Pioneer Point named funders shareholders; not tariff or GPU price Start Campus DataVolt NEOM first phase US$5B Initial first-phase investment; 2025-02-10 Funds first phase of 1.5 GW planned AI factory; no equity split, debt share, tariff or GPU quote NEOM DataVolt Firebird Armenia US$500M Phase 1; US$4B Phase 2 Project investment scopes; 2026-02-10 Capital investment tied to AI megaproject; not GPU purchase prices; NVIDIA CoreWeave amounts ND Firebird Pure DC SJK01 More than €1.5B phase 1; more than €7.5B full campus Project investment pathways; 2026-07-14 Subject to permissions and contracting for full campus; not electricity tariff or GPU price Pure DC Scala AI City US$500M initial phase Project investment; 2024-09-11 Initial investment only; separate 5 GW connection and 4.75 GW full-build capacity have no disclosed tariff or GPU quote Scala QumulusAI Philadelphia US$18M over two years Customer take-or-pay contract; announced 2026-09-10 Revenue contract value activated by live customer handoff; not hardware price, site capex or power tariff QumulusAI Crusoe Cloud H100 US$3.90 GPU-hour ; H200 US$4.29 GPU-hour ; A100 US$2.00–2.30 GPU-hour ; L40S US$1.50 GPU-hour ; MI300X US$3.45 GPU-hour ; B200 and GB200 contact sales Public cloud rental rates; page read 2026-09-25 On-demand spot rental benchmarks; not OEM MSRP, purchase price, annualized cost, site capex or electricity price Crusoe Cloud pricing

Method, chart continuity, and unresolved evidence

This packet preserves source units and status labels: announced planned, permitted authorized, under construction, operational, purchased, planned, authorized for delivery, and system-design capacity are not interchangeable. MVA and kV are not converted to MW; systems and racks are not multiplied into GPU counts; financing, backlog, customer contracts, project investment, cloud rental, and hardware estimates are not combined. The established report retains its ten evidence-backed quantitative chart fences and their baseline order, kinds, item inventories, and display options; the accepted packet’s two chart-ready non-additive views remain available for the separate dashboard view without adding a replacement chart slot.

Unresolved evidence includes the failed ODATA current reread; the absence of independent permits, utility offers, meter data, delivery records, audited installed-GPU counts, and financing closes for many entries; the lack of project electricity tariffs, PPA strike prices, interconnection fees, and OEM GPU quotes; Australia’s undisclosed partner-level GPU bill of materials; QumulusAI’s undisclosed Philadelphia unit count; and the distinction between SINES campus capacity and its operating SIN01. No global MW, GPU, investment, $ MW, or $ GPU total is asserted.

Latest evidence update — 28 September 2026

The accepted 28 September packet is a bounded, source-qualified register: 9 data-center projects and 8 GPU operator programs across North America, Europe, the Middle East, and Southeast Asia. It strengthens the current read on status, capacity class, electricity boundary, company role, explicitly named backing, GPU state, model specification, and price proxy, but it does not support a global total. ND means not disclosed or not verified, not zero.

Current conclusion

The largest project signals in this packet are Meta Richland's 5 GW of compute capacity , Crusoe Abilene's 900 MW new site with a planned 900 MW on-site plant , and SINES' 1.2 GW campus IT capacity . The clearest GPU signals are xAI's 200,000 H100 operating claim , HUMAIN's 18,000-GB300 first phase , Oracle's up-to-131,072-Blackwell service ceiling , and Nebius's more-than-22,000-Blackwell deployment program . These figures describe different states and measurement boundaries; they are not additive.

Data-center plans and construction

ID Project region Status and capacity boundary Electricity interconnection Company explicitly sourced backing Capital or price signal Evidence publication date --- --- --- --- --- --- --- DC-01 Meta Richland Parish Hyperion, Louisiana, US Construction and expansion. 5 GW compute capacity ; source does not label it IT load. Entergy LPSC approvals cover 2.2 GW natural-gas generation plus up to 1,500 MW solar hybrid resources ; Meta pays the energy-infrastructure cost. Meta; Entergy Louisiana utility partner. Blue Owl-managed funds own 80% of the JV and Meta 20% . Approximately US$27B later JV development-cost scope, including about US$7B Blue Owl cash contribution and a US$3B Meta distribution; Meta's earlier US$10B site announcement is a different scope and is not added. Meta expansion 2026-07-13 · Meta Blue Owl JV 2025-10-21 · Entergy power 2025-12-18 DC-02 Crusoe Abilene AI Factory, Texas, US Construction site preparation. 900 MW new campus; full Abilene site projected at 2.1 GW ; source does not label either as IT load. First new building expected energized mid-2027. 900 MW dedicated on-site power plant for the new campus; grid-interconnection quantity ND. Crusoe developer operator; Microsoft is the supported AI-infrastructure customer partner. Equity or project-finance backer ND. Project capex and GPU price ND. Crusoe announcement 2026-03-27 DC-03 CyrusOne ECP KKR Bosque County, Texas, US Construction underway; planned operation Q4 2026 . 190 MW campus and 144 MW initial IT capacity . Co-located with Calpine's Thad Hill Energy Center; long-term dedicated-power contract, MW interconnection rating ND. CyrusOne and ECP joint venture. First investment from the ECP–KKR US$50B strategic partnership . Project investment expected to approach US$4B ; not a GPU price. ECP KKR CyrusOne 2025-07-30 DC-04 Microsoft Pecos datacenter campus, Texas, US Announced planned; five-to-seven-year build. Microsoft says it will add approximately 2 GW of global datacenter capacity ; site IT load ND. Dedicated energy supply located on site; Microsoft says it will fund generation and supporting infrastructure; grid details ND. Microsoft; outside equity backer not named. Multibillion-dollar investment; exact capex and GPU price ND. Microsoft Pecos 2026-06-22 DC-05 Nebius Lappeenranta AI factory, Finland Construction announced planned; first customer capacity expected in 2027 . Up to 310 MW across a multi-building site; IT-versus-utility basis ND. Site-specific grid interconnection capacity ND. Nebius's more-than-3-GW contracted-power target is company-wide and not assigned to this site. Nebius; project investor or financing backer not explicitly named. Project capex and GPU procurement price ND. Nebius Lappeenranta 2026-03-31 DC-06 Stargate Norway, Narvik, Norway Planned early-stage. 230 MW planned with ambition for another 290 MW ; site IT basis ND. OpenAI separately targets 100,000 NVIDIA GPUs by end-2026. Renewable hydropower; specific interconnection or contracted-utility quantity ND. Nscale and Aker expected 50 50 JV ; OpenAI may be initial offtaker. No separate project financier named. Capex and GPU price ND. OpenAI Stargate Norway 2025-07-31 DC-07 Stargate UAE, Abu Dhabi, UAE Announced planned. 1 GW cluster; 200 MW expected live in 2026 . Site IT load ND. Utility interconnection source not disclosed. G42, Oracle, NVIDIA, Cisco, SoftBank, and OpenAI are named partners; equity shares and funding amounts ND. Site capex and GPU price ND. OpenAI Stargate UAE 2025-05-22

Operational or expanding data centers

ID Project region Operational or expansion evidence IT capacity Electricity interconnection Company backing and price Evidence publication date --- --- --- --- --- --- --- DC-08 SINES Data Campus, Portugal Operating expanding based on dated milestones; current status of each building is not separately refreshed. 1.2 GW IT capacity campus; SIN02 up to 180 MW IT ; SIN01 was described at 14 MW , expanding to 26 MW . Siemens Energy transformer order supports SIN02-06; design can connect directly to renewables; 100% renewable target; exact interconnection rating ND. Start Campus developer operator; Siemens Energy supplier. Project equity debt backer not named in the cited release. €8.5B combined investment value and €20M transformer order are separate objects. Start Campus Siemens 2024-08-22 DC-09 YTL Green Data Center Park YTL AI Cloud, Johor, Malaysia AI Cloud reported operational by secondary coverage dated 2025-10-31; park remains an expansion program. MIDA's first phase can accommodate up to 72 MW within a 500 MW park; The Edge reports a 600 MW park. Figures are retained as a discrepancy, not reconciled. YTL portfolio page describes up to 500 MW solar ; grid details ND. YTL Power YTL Data Center Holdings; NVIDIA strategic partner; Sea Ltd initial anchor tenant. MIDA reported RM1.5B first-phase investment; a separate YTL–NVIDIA release cites RM10B partnership context without a funding split. MIDA first phase 2022-08-25 · MIDA NVIDIA–YTL 2024-08-23 · YTL portfolio undated, retrieved 2026-09-28 · The Edge 2025-10-31

Electricity and ownership boundaries

Record Source-labelled value Boundary that must remain separate Evidence publication date --- --- --- --- Meta Richland 2.2 GW natural-gas generation; up to 1,500 MW solar hybrid Generation approvals and energy infrastructure; not IT load or delivered hourly energy Entergy 2025-12-18 Crusoe Abilene 900 MW on-site plant Planned generation; grid import and IT load ND Crusoe 2026-03-27 CyrusOne Bosque County Calpine long-term dedicated-power contract Contracted supply; no disclosed MW and not proof of delivered IT load ECP KKR CyrusOne 2025-07-30 SINES 1.2 GW secured campus IT capacity IT capacity; transformer and renewable-design evidence do not establish a separate interconnection MW Start Campus 2024-08-22 YTL Johor 500 MW park solar description versus 600 MW secondary park report Source discrepancy; neither is treated as current IT load YTL portfolio undated · The Edge 2025-10-31

GPU inventory and deployment state

ID Operator program Quantity and state Model and sourced specification Company explicitly sourced backing Possible price or proxy Evidence publication retrieval date --- --- --- --- --- --- --- GPU-01 xAI Colossus, Memphis, US 200,000 H100 GPUs in one interconnected cluster; operating first-party claim. One-million-GPU roadmap is future context, not installed inventory. NVIDIA H100, Hopper ; fourth-generation Tensor Cores, FP8 Transformer Engine, up to 900 GB s fourth-generation NVLink. xAI operator; NVIDIA system partner; investor funder not named in the cited current page. Procurement price ND. CoreWeave access proxy: 8-GPU HGX H100 US$49.24 instance-hour on demand and US$19.71 spot ; single-GPU inference US$6.16 GPU-hour . xAI Colossus live, retrieved 2026-09-28 · NVIDIA H100 live · CoreWeave pricing live, retrieved 2026-09-28 GPU-02 HUMAIN AI factories, Saudi Arabia 18,000 GB300 in first phase; several-hundred-thousand program target over five years; planned, not installed inventory. GB300 Grace Blackwell Blackwell Ultra NVL72 : 72 Blackwell Ultra GPUs + 36 Grace CPUs per liquid-cooled rack-scale system. HUMAIN is a full AI-value-chain subsidiary of the Public Investment Fund PIF ; NVIDIA strategic technology partner. Procurement price ND; public GB300 cloud rates in the cited table are contact sales . NVIDIA HUMAIN 2025-05-13 · NVIDIA GB300 live, retrieved 2026-09-28 · CoreWeave pricing retrieved 2026-09-28 GPU-03 CoreWeave GPU cloud, North America and Europe 51 AI data centers and 1.5 GW active power with 100,000+ GPU megaclusters on the current page. A 250,000-GPU fleet was reported from a 2025 SEC filing; current official total is ND. Primarily NVIDIA Hopper in the filing-based report; current offerings include H100, H200, GB200, GB300, B200, and others. H200 : 141 GB HBM3e and 4.8 TB s. CoreWeave operator cloud provider; separate inventory backer not identified in the cited sources. Current access rates: HGX H100 US$49.24 and HGX H200 US$50.44 per 8-GPU instance-hour on demand; spot US$19.71 and US$20.93 . CoreWeave AI data centers live, retrieved 2026-09-28 · CNBC filing report 2025-03-26 · CoreWeave pricing retrieved 2026-09-28 · NVIDIA H200 live GPU-04 Oracle Cloud Infrastructure supercluster Up to 131,072 Blackwell GPUs available on GB200 NVL72; capacity ceiling orderable service, not asserted installed inventory. GB200 NVL72 : 72 Blackwell GPUs + 36 Grace CPUs in one liquid-cooled rack-scale system. Oracle operates OCI; NVIDIA platform partner; separate equity backer not disclosed. OCI price ND. CoreWeave GB200 access proxy: US$42.00 instance-hour on demand; not Oracle pricing. Oracle NVIDIA 2025-06-12 · NVIDIA GB200 live · CoreWeave pricing retrieved 2026-09-28 GPU-05 Nebius AI Cloud, US and Finland More than 22,000 Blackwell GPUs announced for deployment; pre-order program state; current installed count by site ND. GB200 NVL72 and HGX B200; GB200 is 72 Blackwell GPUs + 36 Grace CPUs, and HGX B200 is an eight-GPU Blackwell platform. Nebius operator; project investor backer not named in the cited release. Nebius access rates: US$7.15 B200-GPU-hour , US$4.50 H200-GPU-hour , US$3.85 H100-GPU-hour on demand; GB200 contact sales. Nebius Blackwell pre-orders 2024-12-04 · Nebius prices live, retrieved 2026-09-28 · NVIDIA GB200 live · NVIDIA DGX B200 live GPU-06 YTL AI Cloud NVIDIA Johor, Malaysia Operational status reported by secondary coverage dated 2025-10-31; GPU quantity undisclosed. Liquid-cooled GB200 NVL72 , reference topology 72 Blackwell GPUs + 36 Grace CPUs. YTL Power YTL Data Center Holdings; NVIDIA strategic partner; Sea Ltd initial anchor tenant; equity funding split ND. Site and procurement price ND. CoreWeave GB200 proxy US$42.00 instance-hour ; not YTL pricing. The Edge 2025-10-31 · MIDA 2024-08-23 · NVIDIA GB200 live · CoreWeave pricing retrieved 2026-09-28 GPU-07 Stargate Norway, Narvik, Norway 100,000 NVIDIA GPUs targeted by end-2026 ; planned target, not installed inventory; model mix not disclosed. NVIDIA platform; architecture and memory specification ND in the cited launch announcement. Nscale Aker planned 50 50 JV; OpenAI may be initial offtaker; GPU-financing backer and price ND. Procurement and cloud-access price ND. OpenAI Stargate Norway 2025-07-31 GPU-08 Core42 G42 next-generation AI cloud, UAE H100 deployment announced; quantity and current installed count ND. Operating expanding status comes from the 2024 announcement. NVIDIA H100, Hopper ; fourth-generation Tensor Cores and FP8 Transformer Engine. Core42, a G42 company; separate investor backer not identified. Site price ND. Public H100 proxies: CoreWeave US$49.24 per 8-GPU instance-hour on demand and Nebius US$3.85 GPU-hour ; neither is Core42 pricing. Core42 G42 2024-10-09 · NVIDIA H100 live · CoreWeave pricing retrieved 2026-09-28 · Nebius prices retrieved 2026-09-28

Capital and price interpretation

Economic object Source value What it means Evidence date --- --- --- --- Meta Blue Owl Hyperion About US$27B development cost; about US$7B Blue Owl cash contribution JV development and funding scope for buildings and long-lived infrastructure; not GPU price or tariff Meta Blue Owl 2025-10-21 CyrusOne Bosque County About US$4B total project investment Project investment; not IT-load unit cost or hardware price ECP KKR CyrusOne 2025-07-30 SINES Data Campus €8.5B combined investment; €20M transformer order Project investment and supplier order are separate; neither is a GPU quote Start Campus Siemens 2024-08-22 YTL Johor RM1.5B first-phase investment; separate RM10B partnership headline First-phase investment and broader partnership context; scope and funding split not disclosed MIDA 2022-08-25 · YTL Community 2025-07-26 Public access proxies CoreWeave H100 H200 US$49.24 US$50.44 per 8-GPU instance-hour on demand; Nebius H100 H200 B200 US$3.85 US$4.50 US$7.15 per GPU-hour Hosted access rates captured 2026-09-28; provider, unit, plan, and availability differ; no OEM purchase price was verified CoreWeave pricing retrieved 2026-09-28 · Nebius prices retrieved 2026-09-28

Chart continuity and dashboard-readable presentation

The report retains its ten baseline quantitative chart blocks in the existing order, with the same bar-chart kinds, item inventories, and display options. They provide two distinct evidence-backed analytical families: source-labelled capacity or IT-load comparisons, and status-qualified GPU quantity or hosted-rate comparisons. The 28 September artifact does not reproduce every baseline item inventory, so historical chart observations remain dated rather than being relabelled current; no replacement or extra chart slot is added, and no missing observation is filled with zero. The new packet is presented in the tables above so units, states, source dates, investor roles, and price types display correctly together.

Evidence limitations and unresolved risks

Risk Current treatment --- --- Comparability IT load, compute capacity, campus facility power, grid interconnection, generation, contracted power, and master-plan capacity are different objects. No global total is calculated. Inventory state xAI is a first-party operating claim; HUMAIN, Norway, and Nebius are planned or program quantities; OCI is a service ceiling; CoreWeave's 250,000 figure is filing-based secondary reporting; YTL and Core42 quantities are ND. None is a verified installed global total. Price Site procurement prices and OEM prices for H100, H200, GB200, GB300, and Blackwell systems are ND in the cited primary sources. Cloud rates are access proxies, not purchase prices. YTL discrepancy MIDA YTL describe a 500 MW park, while The Edge reports a 600 MW park with a 500 MW solar plant. Both are retained and qualified. Source access CoreWeave's Q4 2025 earnings presentation and transcript were not usable in the evidence pass and were excluded; accessible current CoreWeave pages and the filing-based CNBC report were used instead. Coverage The sample is representative rather than a global census. IREN's Bundey South Australia and other candidates were not added without sufficiently usable primary evidence.

Latest evidence update — 29 September 2026

The 29 September source packet is a bounded, source-qualified register rather than a global census. Its visible tables contain nine project rows, five GPU inventory or deployment rows, and four model-reference rows. The packet header says eight GPU operator programs, but that count does not reconcile with the visible GPU register; the discrepancy is retained and no global program count is asserted. All figures below preserve the source-defined unit, state, date, company role, financing relationship, and price type. ND means not disclosed or not verified, not zero.

Current conclusion

The strongest current project signals are Meta Richland's 5 GW compute-capacity statement , Crusoe Abilene's 1.2 GW total power and grid-interconnection signals , Khazna QAJ1's 100 MW total IT load , Scala AI City's 5 GW ramp-up connection authorization , and Nebius Lappeenranta's up to 310 MW facility-capacity signal . The clearest GPU records are xAI's 200,000 H100 operating claim , Crusoe's up to 50,000 GB200 NVL72 systems per building design statement , Microsoft's hundreds of thousands of NVIDIA GPUs with model mix undisclosed, Nebius Patmos's about 35,000-GPU full-potential estimate , and Africa Compute Fund's more than 100,000 H100-class network target . These figures describe different boundaries and states; they are not additive.

Data-center project and construction register

ID Project region Source date Status Developer operator Investor financing sponsor IT compute capacity Utility electrical capacity Target timing Investment possible price Evidence --- --- --- --- --- --- --- --- --- --- --- DC-29-01 Meta Richland Parish Hyperion, Louisiana, US 2026-07-13 Under construction expanding Meta Platforms Meta; no separate outside project investor disclosed 5 GW announced compute capacity ; source does not label it IT load Separate grid or utility MW ND; energy agreement with Entergy Louisiana No completion date in the cited expansion notice More than US$50B regional investment; not site capex or GPU price Meta Data Centers 2026-07-13 DC-29-02 Stargate Crusoe Abilene campus, Texas, US 2025-03-18; 2025-05-21; 2025-09-30 Operational phase 1; phase 2 under construction Crusoe at Lancium Clean Campus; Oracle collaboration; Stargate workload association US$15B phase-2 JV sponsored by Crusoe, Blue Owl Real Assets funds, and Primary Digital Infrastructure; program-level Stargate funders separately include SoftBank, OpenAI, Oracle, and MGX Separate IT load ND; eight-building design supports up to 50,000 NVIDIA GB200 NVL72 systems per building , a design statement rather than installed inventory 1.2 GW total power capacity and 1.2 GW grid interconnection ; phase 1 is 200+ MW; batteries and solar planned Phase-2 source target was energized mid-2026; no later phase-2 commissioning source is in this packet US$15B JV scope; not confirmed construction cost per MW or GPU price Crusoe expansion 2025-03-18 · Crusoe JV 2025-05-21 · Phase 1 live 2025-09-30 DC-29-03 Microsoft Fairwater, Mount Pleasant, Wisconsin, US 2024-05-08; 2025-09-18; 2026-06-23 Phase 1 operational; phase 2 under construction Microsoft Microsoft; no separate project investor named IT load MW ND; first facility described as having hundreds of thousands of NVIDIA GPUs , with exact count and model mix ND MW IT load and separate grid capacity ND; Microsoft says it prepaid energy and electrical infrastructure Phase 1 online in 2026; adjacent second facility scheduled for 2028 Initial US$3.3B pledge plus US$4B for the second facility; not a GPU quote Microsoft investment 2024-05-08 · Fairwater 2025-09-18 · Completion update 2026-06-23 DC-29-04 Digital Edge DC B.Grimm Power EEC campus, Chonburi, Thailand 2025-06-09; 2025-09-04 Under construction; groundbreaking reported 2025-09-04 Digital Edge DC and B.Grimm Power joint venture Digital Edge backed by Stonepeak; B.Grimm Power is the local energy partner 100 MW announced campus capacity ; source does not define it as IT load or utility limit Renewable-power supply described; project-specific grid or interconnection MW ND Ready-for-service target Q4 2026 US$1B investment plan covers Thailand campuses, not expressly this first site Digital Edge B.Grimm 2025-06-09 · Groundbreaking 2025-09-04 DC-29-05 Nebius Lappeenranta AI factory, Finland 2026-03-31 Announced construction planned Nebius Project investor or financing sponsor ND Up to 310 MW facility capacity ; IT-versus-utility basis ND Site-specific grid or interconnection MW ND; Nebius's more-than-3-GW contracted-power target is company-wide First customer capacity expected in 2027 Project capex and GPU procurement price ND Nebius 2026-03-31 DC-29-06 Khazna QAJ1 AI data center, Ajman, UAE 2024-10-15; 2025-09-03; 2026-02-24 Under development design-certified; the 2026 certification still says the facility is to be constructed Khazna Data Centers Khazna; US$2.62B platform financing facility with ADCB and FAB, allocation to QAJ1 ND 100 MW total IT load across 20 halls at 5 MW each Electricity primarily supplied by Etihad Water and Electricity; separate grid capacity ND Original announcement targeted Q3 2025 operations; later design certification does not confirm completion Multi-billion-dirham project investment announced in 2024; platform facility is not QAJ1-only capex Khazna project 2024-10-15 · Khazna financing 2025-09-03 · QAJ1 certification 2026-02-24 DC-29-07 Scala AI City, Eldorado do Sul, Brazil 2025-05-14; undated portfolio context Planned; power connection authorized; full campus operation not evidenced Scala Data Centers Scala founded by DigitalBridge; R$180M BNDES funding is company-level and not allocated to AI City AI City IT capacity ND; Scala separately reports more than 100 MW operational and more than 900 MW under construction or future across its portfolio 5 GW ramp-up connection authorization ; full build-out described as 4.75 GW of capacity; boundary is not reconciled No reliable completion date in the cited release No AI City-specific capex or GPU price verified Scala power authorization 2025-05-14 · Scala company context DC-29-08 Teraco CT2 expansion, Cape Town, South Africa 2025-11-10 Operational expansion completed Teraco, a Digital Realty company Digital Realty ownership disclosed; expansion financing ND 50 MW critical IT load total , including 32 MW added across eight halls Separate utility or grid capacity ND; a 120 MW solar PV plant is being developed for data centers via wheeling Expansion completed 2025-11-10 Project capex ND Teraco 2025-11-10 DC-29-09 Africa Compute Fund Nairobi Supercluster, Kenya and broader African rollout 2025-07-14; 2025-10-16 Planned; initial engineering and power-allocation design commenced; company-reported and not independently verified here Africa Compute Fund Investor or financing sponsor ND; unnamed global or regional partners 100 MW AI data-center first phase ; broader network target is 1 GW of compute capacity by 2030 Renewable or hybrid power described; project-specific grid MW ND Deployment begins in 2026 ; broader 1 GW target by 2030 Project price ND Nairobi Supercluster 2025-07-14 · 100 MW plan 2025-10-16

GPU inventory and deployment register

ID Operator program Source date Model Quantity and state Memory, compute, interconnect, and power Company backing Possible price or proxy Evidence --- --- --- --- --- --- --- --- --- GPU-29-01 xAI Colossus, Memphis, US Live page; publication date not shown NVIDIA H100 200,000 H100 GPUs in one interconnected cluster; operating first-party claim H100 SXM reference: 80 GB memory; 3.35 TB s memory bandwidth; up to 3,958 FP8 teraFLOPS; up to 700 W configurable TDP; 900 GB s NVLink; exact xAI variant ND xAI SpaceXAI operator or owner; outside investor ND in the cited pages H100 rental index: average US$3.84 GPU-hour , range US$1.79–12.52 , prices as of 2026-09-13; not a purchase quote xAI Colossus · NVIDIA Colossus release 2024-10-28 · H100 datasheet · Mercatus index GPU-29-02 Crusoe Abilene Stargate campus, Texas, US 2025-03-18; 2025-05-21; price estimate 2024-07-28 NVIDIA GB200 NVL72 systems Up to 50,000 systems per building across an eight-building design; installed total ND; design capacity, not confirmed inventory One rack-scale system has 36 Grace CPUs and 72 Blackwell GPUs; 72-GPU NVLink domain; 130 TB s NVLink Switch System communications; 1.8 TB s GPU-to-GPU interconnect; rack power ND Crusoe, Oracle, and OpenAI workload association; project sponsors Crusoe, Blue Owl Real Assets funds, and Primary Digital Infrastructure Analyst estimate about US$3M per GB200 NVL72 rack , reported 2024-07-28; not a vendor quote Crusoe campus design · Crusoe JV · GB200 NVL72 · Price report GPU-29-03 Microsoft Fairwater, Mount Pleasant, US 2025-09-18; 2026-06-23; price context 2024-03-19 NVIDIA model mix not disclosed Hundreds of thousands of NVIDIA GPUs for the facility; exact count and model mix ND; phase 1 operational, phase 2 under construction Per-GPU specification cannot be assigned because the model mix is undisclosed Microsoft owner and operator; no separate hardware financier disclosed Media-reported US$30,000–40,000 Blackwell GPU or stack reference; no confirmed purchase price and NVIDIA sells integrated systems Fairwater GPU statement · Completion update · Blackwell pricing context GPU-29-04 Nebius Patmos data center, Kansas City, US 2024-11-19; price index 2026-09-13 NVIDIA H200 initially; Blackwell later planned Initial 5 MW cluster expandable to 40 MW about 35,000 GPUs at full potential; exact H200 count ND; planned deployment with no later commissioning verified H200 reference: 141 GB HBM3e; 4.8 TB s memory bandwidth; same power profile as H100; exact H200 SKU or TDP ND Nebius deployment at a Patmos-owned data center; investor or financing sponsor ND H200 rental index: average US$4.43 GPU-hour , range US$2.25–11.71 , prices as of 2026-09-13; not a purchase price DCD report 2024-11-19 · NVIDIA H200 · Mercatus index GPU-29-05 Africa Compute Fund Nairobi and planned African network 2025-07-14; 2025-10-16 NVIDIA H100-class systems Network designed to accommodate more than 100,000 H100-class systems ; Nairobi quantity ND; company-reported plan, not independently verified H100-class is not a precise SKU; H100 family specifications are reference only and do not prove hardware procured Africa Compute Fund and unnamed global or regional partners; investors ND Public price ND ACF 100 MW plan · ACF Nairobi node · NVIDIA H100 datasheet

Model specification reference

These manufacturer fields are reference specifications only. They are not site inventories, confirmed procurement quantities, or project power draws.

Model system Source date Memory Compute bandwidth Interconnect Power Evidence --- --- --- --- --- --- --- NVIDIA H100 SXM Date ND in cited datasheet 80 GB 3.35 TB s memory bandwidth; up to 3,958 FP8 teraFLOPS 900 GB s NVIDIA NVLink Up to 700 W configurable TDP NVIDIA H100 datasheet NVIDIA H200 SXM Date ND in cited product page 141 GB HBM3e 4.8 TB s memory bandwidth; up to 2x LLM inference speed versus H100 in cited workloads Cluster interconnect number ND in captured evidence Same power profile as H100; exact SKU TDP requires verification NVIDIA H200 NVIDIA GB200 NVL72 Date ND in cited product page 72 Blackwell GPUs plus 36 Grace CPUs per rack-scale system 130 TB s NVLink Switch System communications; 1.8 TB s GPU-to-GPU interconnect 72-GPU NVLink domain; liquid-cooled rack ND in cited product page NVIDIA GB200 NVL72 NVIDIA DGX B200 Date ND in cited product page 1,440 GB total across eight Blackwell GPUs 64 TB s HBM3e bandwidth; 144 PFLOPS FP4; 72 PFLOPS FP8; 14.4 TB s aggregate NVLink Fifth-generation NVLink NVSwitch Approximately 14.3 kW maximum system power NVIDIA DGX B200

Price and financing observations

These are directional references with original units and dates. They are not project quotes, hardware invoices, or comparable totals.

Item Possible price or rate Source date Price type Boundary and qualification Evidence --- ---: --- --- --- --- H100 US$30,000–40,000 per GPU at the historical peak cited 2024-03-19 Media-reported historical hardware range Not NVIDIA MSRP and not a current quote Tom's Hardware B200 Blackwell Approximately US$30,000–40,000 for the GPU or stack as described 2024-03-19 Media-reported approximation NVIDIA sells integrated systems and stacks; not a card-level price Tom's Hardware GB200 NVL72 About US$3M per rack 2024-07-28 HSBC analyst estimate reported by TechSpot Rack configuration and scope matter; not a vendor quote TechSpot H100 cloud rental US$3.84 GPU-hour average; US$1.79–12.52 range Prices as of 2026-09-13 Market-index on-demand rental Recurring access price, not capex or ownership Mercatus GPU Index H200 cloud rental US$4.43 GPU-hour average; US$2.25–11.71 range Prices as of 2026-09-13 Market-index on-demand rental Recurring access price, not capex or ownership Mercatus GPU Index B200 cloud rental US$6.39 GPU-hour average; US$3.74–16.14 range Prices as of 2026-09-13 Market-index on-demand rental Recurring access price, not capex or ownership Mercatus GPU Index

Methodology and limitations

The source packet was researched on 29 September 2026 UTC and is a representative shortlist, not an exhaustive global census. Company, operator, developer, utility, government, manufacturer, and secondary market sources are kept distinct. Primary or official sources support most project and specification fields; secondary sources are used only for the Nebius Patmos status and the GB200 and Blackwell price estimates. The visible packet contains nine project rows, five GPU inventory or deployment rows, and four reference-spec rows. Its eight-program header does not reconcile with the visible GPU register and is treated as an explicit data-quality gap.

Capacity is retained in the source-defined boundary: IT load or compute capacity is not substituted for facility power, campus power, grid connection, generation, renewable supply, or a contracted-power target. No MVA, kV, facility power, or generation figure is converted to MW without a source-backed basis. Parent programs and child sites are not added. GPU systems and racks are not multiplied into installed GPU counts. Operating, under-construction, planned, design, service-ceiling, orderable, and network-target states remain separate. Financing, project investment, supplier orders, cloud rental, market estimates, and hardware purchase proxies are not combined.

The baseline report's ten quantitative chart fences remain in their original count, order, kinds, item inventories, and display options. They retain the two distinct analytical families already established: source-labelled capacity or IT-load comparisons and status-qualified GPU quantity or hosted-rate comparisons. The current packet does not reproduce every baseline chart item, so no chart slot is replaced or expanded, and missing observations are not filled with zero. The current packet is therefore shown through the tables above, with dates and uncertainty visible beside every material value.

Limitation Current treatment --- --- Capacity comparability Meta's 5 GW is compute capacity; Crusoe's 1.2 GW is total power and grid-interconnection evidence; Khazna's 100 MW is IT load; Scala's 5 GW is a connection authorization. No global MW total is calculated. Inventory state xAI is a first-party operating claim; Crusoe is a design ceiling; Microsoft is an undisclosed model mix; Nebius is a planned full-potential estimate; Africa Compute Fund is a network target. None is a verified global installed total. Price quality No project-specific GPU purchase order, OEM quote, or electricity tariff is disclosed in the packet. Hardware values are media or analyst estimates; cloud rates are access proxies. Status refresh Khazna QAJ1's original operating target is contradicted by a later design-certification status; Nebius Patmos's Q1 2025 target has no later commissioning evidence in the packet. Backing disclosure Crusoe's JV sponsors and Khazna's platform lenders are explicit; separate project investors for Meta, Microsoft, Nebius, Teraco, and Africa Compute Fund are ND. Count reconciliation The packet header says eight GPU operator programs while the visible inventory register has five rows; the mismatch is disclosed and not resolved by inference.

Watch items

• Revalidate commissioning and energization for Crusoe phase 2, Khazna QAJ1, Nebius Patmos, Digital Edge EEC, and Africa Compute Fund deployment.
• Obtain first-party utility or interconnection records that distinguish IT load, facility power, generation, and delivered electricity.
• Seek auditable GPU manifests for Microsoft Fairwater, Nebius Patmos, and Africa Compute Fund, and a precise SKU for the H100-class network target.
• Treat all hardware and hosted-rate observations as dated market references until a project-specific quote or purchase record is disclosed.

Latest evidence update — 30 September 2026

The latest bounded packet refreshes nine representative project records and eight GPU inventory or deployment records across North America, Europe, the Middle East, and Asia-Pacific. The clearest capacity signals are Meta Hyperion's 5 GW compute-capacity statement , Crusoe Abilene's 1.2 GW total-power and grid-interconnection signals , Nebius Lappeenranta's up to 310 MW AI-factory capacity , Stargate UAE's 1 GW cluster with 200 MW expected live in 2026 , HUMAIN's up to 500 MW projected AI-factory capacity , AirTrunk TOK1's 300+ MW scalable campus , and AirTrunk JHB3 JHB4's more than 280 MW combined IT load . These values use different boundaries and are not summed. See Crusoe, Meta, Nebius, Stargate UAE, HUMAIN NVIDIA, AirTrunk TOK1, and AirTrunk Malaysia.

GPU evidence remains state-qualified: xAI reports 100,000 NVIDIA Hopper GPUs operational with a 200,000 combined target ; Meta discloses 49,152 H100s across two clusters alongside larger targets; Yotta reports more than 4,000 H100s delivered in a dated disclosure; HUMAIN announces 18,000 GB300 in a first phase with ambiguous unit semantics; Nebius describes a 60,000-GPU site expansion ceiling ; Crusoe describes up to 50,000 GB200 NVL72 systems per building ; OpenAI describes more than 2 million chips across Stargate U.S. capacity under development; and Meta's historical RSC configuration contains 16,000 A100 GPUs . These are not an installed global total. See xAI, Meta GenAI, Yotta, NVIDIA HUMAIN, Nebius, Crusoe GB200, OpenAI Stargate, and Meta RSC.

Cost disclosures remain uneven: $15B for the Abilene joint venture, approximately $27B for Hyperion development, $6B for xAI's Series C, $3B planned for AirTrunk JHB3 JHB4, and more than $1B for Nebius's broader European program. None is a per-GPU purchase price. Site-specific energization, utility boundaries, installed counts, unit definitions, and project financing allocation remain unresolved; ND is retained rather than converted to zero.

Latest evidence update — 1 October 2026

The current evidence matrix is a representative, source-qualified update across North America, Europe, the Middle East, Africa, Asia-Pacific Oceania, and Latin America. It adds 14 project records and 7 GPU or hosted-service records. The clearest current signals are Meta Hyperion's 5 GW compute capacity , Oracle's more than 1.7 GW of wind generation supporting the ERCOT grid, NEXTDC S4's 365 MW IT-capacity statement, xAI's 200,000 NVIDIA H100 claim, and Meta's up to 6 GW AMD Instinct commitment. These measures have different boundaries and are not summed.

Current data-center projects and capacity signals

ID Region project Status and source date IT facility capacity evidence Electricity grid evidence Company; investor funder Timing and price signal Evidence status --- --- --- --- --- --- --- --- DC-01 United States — Meta Hyperion, Richland Parish, Louisiana Expansion active; ground broken Dec 2024. Meta update dated 2026-07-13. 5GW compute capacity ; Meta also describes a 4M sq ft campus. The source does not equate 5GW with IT load. No site import or grid-connection MW disclosed in the reviewed sources. Meta operates the AI campus. Hyperion JV: Blue Owl funds owns 80%; Meta owns 20%. JV announcement dated 2025-10-21: approximately $27B total development costs for buildings and long-lived power cooling connectivity; Blue Owl cash contribution approximately $7B . No completion date disclosed. Compute and investment are primary-source disclosures; IT grid equivalence remains unknown. Meta 2026 Meta JV 2025 DC-02 United States — Microsoft Fairwater, Wisconsin Active build-out; Microsoft technical overview dated 2025-09-18. 315-acre site, three buildings, combined 1.2M sq ft; designed as one AI supercomputer with hundreds of thousands of latest NVIDIA GPUs . No IT MW or exact GPU count. No facility-load or grid-connection MW disclosed. Microsoft operator and project funder; no separate backer disclosed. No project price or completion date in the reviewed source. Primary source; strong physical GPU-scale evidence, weak power and funding detail. Microsoft DC-03 United States — Oracle AI campus, Abilene, Texas Construction active; Oracle construction update dated 2026-06-25. Oracle reports nearly 20M worker hours and up to 9,000 workers on site to date; no IT MW disclosed. Oracle announced investment in wind projects delivering more than 1.7GW across ten projects to the ERCOT grid that powers Abilene. This is generation grid support, not the campus load. Oracle operator and investor; project-specific funding split not disclosed. No total project price or completion date in the reviewed sources. Primary sources; capacity type is explicitly separated. Construction Grid support DC-04 Malaysia — Google data center and cloud region, Elmina Business Park Ground broken 2024-10-01; Google announcement. Current completion status was not reverified in this run. No IT or facility MW disclosed; first Malaysia data center and cloud region. No site grid connection MW disclosed. Google Alphabet operator and announced investor; DNeX is described as a partner. Funding split not disclosed. US$2B investment announced; Google also cited an estimated $3.2B economic impact by 2030, which is not the project price. Primary source; investment is clear, status and power details need refresh. Google DC-05 United Kingdom — Yondr 100MW London campus, Slough Mixed: first second-building phases partly live; third building under construction in the latest reviewed release. Sources dated 2025-02-10 and 2025-04-10. Campus just over 100MW ; second data center first phase 10MW of 30MW live ; third building 40MW . These are campus data-center capacity figures as stated by Yondr. No separate grid-import rating disclosed. Yondr operator; project financing investor not disclosed in the reviewed releases. Third phase was expected mid-2026 in the 2025-02-10 release; current completion was not reverified. No project price. Primary operator releases; timing is stale relative to the as-of date. Second data center Third phase DC-06 Spain — Nabiax ADC3, Alcalá de Henares, Madrid Groundbreaking first-stone event 2026-09-18; secondary capacity report dated 2026-09-21. Up to 100MW IT capacity through 2029 ; ADC1 and ADC2 together reported at 22MW; ADC3 investment reported near €800M. No grid-connection MW disclosed. Nabiax operator; investor funder not disclosed in the reviewed sources. Nearly €800M project investment reported by DCD; 100MW IT target through 2029. Official regional event confirms first-stone event; DCD supplies capacity and investment detail. Madrid government DCD DC-07 United Arab Emirates — Khazna AUH4 AUH8 Abu Dhabi and QAJ1 Ajman Groundbreaking progress announcement dated 2025-04-21; planned under construction. AUH4 and AUH8 combined 60MW ; QAJ1 100MW , with an initial phase. Source frames QAJ1 as AI-optimized. No grid-connection MW disclosed. Khazna operator; project-level funder not disclosed in the reviewed announcement. AUH8 due Aug 2026; AUH4 and QAJ1 initial phase due Dec 2026. No project price. Primary operator source; current completion status and funding need refresh. Khazna DC-08 Japan — AirTrunk TOK1 A 40+MW phase broke ground 2025-06-19; under construction at source date. TOK1 planned at 300+MW total capacity ; AirTrunk says its Japan TOK1 TOK2 OSK1 portfolio totals 430+MW . IT-vs-facility basis is not further defined in the source. No grid-connection MW disclosed. AirTrunk operator; investor funder not stated in the reviewed release. No project price or completion date for the phase in the reviewed source. Primary operator source; planned capacity, not verified live load. AirTrunk DC-09 Japan — FLOW TK7 TK8, central Tokyo Construction underway; announcement dated 2025-07-29. TK7 TK8 combined 30MW IT ; TK7 6MW IT ; FLOW portfolio stated as more than 170MW current planned. No grid-connection MW disclosed. FLOW operator; platform is backed managed by PAG, which the source describes as managing more than $55B for 300 institutional investors. Project-specific funding allocation is not disclosed. Target delivery Q1 2027; no project price. Primary operator source; IT basis is explicit, investor context is not a project funding amount. FLOW DC-10 Australia — NEXTDC S4 Sydney, Horsley Park Current NEXTDC page says “in development”; page accessed 2026-10-01. Current page states 365MW IT capacity and 124,000m² technical space; 2021 announcement described a 300MW target subject to approval. Site is near a major electricity substation; no import connection MW disclosed. NEXTDC operator; funding source not disclosed in reviewed materials. No completion date or project price. The 365MW current page and older 300MW announcement require reconciliation against the latest filing. Primary current page plus older primary announcement; capacity revision is unresolved. Current S4 page 2021 announcement DC-11 Chile — AWS South America Chile Region Planned being delivered; AWS announced launch by end-2026. Three Availability Zones; no regional IT or facility MW disclosed. No grid connection MW disclosed. Amazon AWS operator and announced investor; no separate funder. AWS says Amazon plans to invest more than $4B over 15 years for construction, connection, operation, and maintenance. Primary AWS Amazon announcement; power and current construction status need refresh. Amazon AWS DC-12 Nigeria — Kasi Cloud LOS1, Lekki Lagos Commissioned and moving from construction into commissioning operational readiness in the report dated 2026-05-21. Expected scale approximately 100MW IT ; exact live building load not disclosed. No grid-connection MW disclosed. Kasi Cloud operator; funder not disclosed. No project price or final completion date. Secondary source; capacity is described as expected scale and requires operator utility confirmation. Data Center Post DC-13 Chile — TECfusions Baeza Group, Puente Alto Land secured; planned and subject to due diligence permits. Source dated 2025-06-06. Ultimate 100MW campus; initial 10MW ; source also describes an additional 10MW from the local grid . Do not treat the ultimate 100MW as committed. 10MW local-grid component disclosed; full grid capacity not disclosed. TECfusions operator developer; Baeza Group land owner and JV partner. Funding amount not disclosed. First phase expected within one year of groundbreaking, subject to permits; no project price. Secondary source with useful phased-capacity detail; no official utility confirmation in this run. DCD DC-14 Brazil — Scala Tamboré, São Paulo Power-infrastructure construction began 2024-05-06; current completion status not reverified. A 560MW power substation not IT capacity ; existing substations cited at 60MW each. Current campus IT total not used. MME authorization for connection to the transmission system; 560MW is the substation rating stated by Scala. Scala operator; funding split not disclosed. More than $80M investment in the substation; first phase was scheduled for Dec 2024. Primary operator source; current operational status and IT capacity remain open. Scala

The current project values cannot be summed without a common basis. Meta's 5GW is compute capacity, Oracle's 1.7GW is new wind generation for ERCOT support, Scala's 560MW is a power-substation rating, and Yondr, AirTrunk, NEXTDC, and TECfusions values describe campus or data-center capacity on differing source bases. A global aggregate remains not established .

Current GPU inventory and hosted-service register

ID Company system Disclosed model and quantity scale Status and ownership Investor funder Price type and observed value Evidence --- --- --- --- --- --- --- GPU-01 xAI Colossus 200,000 NVIDIA H100 GPUs in one interconnected cluster; roadmap to 1M. xAI-owned operated cluster; page says built in 122 days and doubled in 92 days. Not disclosed. No acquisition price or cloud rate stated. Primary xAI page, undated and accessed 2026-10-01. xAI Colossus GPU-02 Microsoft Fairwater Hundreds of thousands of latest NVIDIA GPUs in one flat-network AI supercomputer; exact count and models not disclosed. Microsoft build-out; not a verified installed count as of this run. Microsoft; no separate project backer disclosed. No price stated. Primary Microsoft technical overview dated 2025-09-18. Microsoft Fairwater GPU-03 Meta AMD agreement Up to 6GW of AMD Instinct GPUs ; exact model, GPU count, and delivered quantity not disclosed. Multi-year commitment; first deployments planned for H2 2026. It is not an installed-inventory count. Meta buyer operator; AMD supplier; no external funder stated. Contract price not disclosed. Primary Meta announcement dated 2026-02-24. Meta AMD GPU-04 CoreWeave NVIDIA IBM MLPerf system 2,496 NVIDIA Blackwell GPUs in the reported GB200 benchmark cluster. Benchmark configuration result, not CoreWeave's total fleet. CoreWeave operator provider; NVIDIA and IBM collaborators, not reported as funders. No acquisition price stated. Primary CoreWeave investor release dated 2025-06-04. CoreWeave MLPerf GPU-05 AWS EC2 P5 family P5 uses NVIDIA H100; P5e P5en use H200. Quantity inventory and a directly retrievable P5 hourly price were not disclosed in the reviewed page text. Cloud service availability, not owned-fleet inventory. AWS Amazon. Product availability only; no price recorded here. Primary AWS instance page, accessed 2026-10-01. AWS P5 GPU-06 Lambda GPU Cloud 1-Click clusters from 16 to 2,000+ NVIDIA B200 or H100 GPUs ; offered cluster scale, not disclosed installed inventory. Lambda cloud rental service offering. Lambda; no separate funder stated. US$6.69 GPU-hour for B200 180GB and US$3.99 GPU-hour for H100 SXM 80GB; hosted rental rates. Primary Lambda pricing page, accessed 2026-10-01. Lambda pricing GPU-07 CoreWeave cloud offerings Pricing page lists GB200 NVL72, HGX H100 8 GPUs , HGX H200 8 GPUs , and B200 8 GPUs ; this is not a fleet count. CoreWeave rental service offering in the North America table. CoreWeave; no separate funder stated. US$42 instance-hour for GB200 NVL72; US$49.24 8-GPU instance-hour for HGX H100; US$50.44 8-GPU instance-hour for HGX H200; US$68.80 8-GPU instance-hour for B200. Primary CoreWeave pricing page, accessed 2026-10-01. CoreWeave pricing

Current accelerator specification anchors

These manufacturer or provider fields are reference specifications, not proof of site ownership, procurement, installation, or current stock.

Model family Specification supported by reviewed source Source --- --- --- NVIDIA H100 Hopper architecture; fourth-generation Tensor Cores and FP8 support; 900GB s GPU interconnect; 60 TFLOPS FP64 on the official product page. Lambda's rental listing identifies the priced H100 SXM configuration as 80GB. NVIDIA H100 Lambda pricing NVIDIA H200 141GB HBM3e at 4.8TB s ; NVIDIA says it has the same power profile as H100 and labels the specifications preliminary on the reviewed page. NVIDIA H200 NVIDIA Blackwell B200 family Blackwell architecture page states 208B transistors , fifth-generation NVLink supporting up to 576 GPUs, and up to 130TB s GPU bandwidth in a 72-GPU NVLink domain. NVIDIA Blackwell NVIDIA GB200 NVL72 CoreWeave's MLPerf release identifies a GB200 Blackwell cluster with 2,496 GPUs; the CoreWeave price page separately lists a GB200 NVL72 rental row at $42 per on-demand instance-hour. The price row is not used to infer the benchmark cluster's topology or acquisition cost. CoreWeave MLPerf CoreWeave pricing

Current price basis register

Price type Observed value Interpretation --- ---: --- Lambda H100 SXM 80GB $3.99 GPU-hour Cloud rental price as displayed; not hardware purchase price. Lambda B200 180GB $6.69 GPU-hour Cloud rental price as displayed; not hardware purchase price. CoreWeave GB200 NVL72 $42 instance-hour Cloud rental price in the North America table; the provider's instance row should not be converted into an acquisition value. CoreWeave HGX H100 $49.24 8-GPU instance-hour Equivalent arithmetic is $6.155 GPU-hour, calculated only as a comparison from the displayed 8-GPU configuration. CoreWeave HGX H200 $50.44 8-GPU instance-hour Equivalent arithmetic is $6.305 GPU-hour, calculated only as a comparison from the displayed 8-GPU configuration. CoreWeave B200 $68.80 8-GPU instance-hour Equivalent arithmetic is $8.60 GPU-hour, calculated only as a comparison from the displayed 8-GPU configuration. Hardware acquisition list prices Unknown No defensible manufacturer list price, transaction price, or purchase contract price was found in the reviewed evidence. Cloud rates must not be reported as purchase prices.

Methodology and unresolved risks

The current matrix was reviewed as of 1 October 2026 UTC . It is a representative evidence set, not a global census or underwriting model. Source publication dates and access dates are retained beside material claims. IT load, facility campus capacity, compute capacity, grid interconnection, generation, substation ratings, and policy or contracted capacity remain separate. Parent programs and child sites are not added, and racks, systems, servers, offered clusters, targets, aggregate chips, and cloud access do not become installed-GPU totals. Capital, debt, project investment, transaction value, hosted rental, and hardware estimates retain their original price type, currency, unit, and qualification.

Unresolved risk Current treatment --- --- No global aggregate The source set has no common denominator. Compute capacity, IT MW, campus MW, generation, substation ratings, and cloud GPU-hours remain heterogeneous. Sparse grid evidence Most projects disclose no facility import, interconnection, utility, or contracted-power MW. Oracle's 1.7GW and Scala's 560MW are electricity signals, not facility IT loads. Stale or future status Google Malaysia, Yondr, AirTrunk, FLOW, Khazna, AWS Chile, and Scala have source dates earlier than the as-of date or future delivery dates; current commissioning is not upgraded without evidence. NEXTDC reconciliation The current S4 page says 365MW IT capacity while the older announcement says a 300MW target. The difference remains unresolved. Funding disclosure Several owners and operators do not publish project-level investor or financing details. Parent-company or platform capital is not inferred as project funding. GPU definitions xAI's 200,000 H100 statement and CoreWeave's 2,496-GPU benchmark are direct scale signals; Microsoft's “hundreds of thousands” and Meta's 6GW AMD agreement are not exact installed counts. Price limitations The reviewed sources provide hosted rental rates but no reliable acquisition or list prices. Currency, region, reservations, taxes, and availability can change; rental figures are not comparable to capex. Chart continuity The baseline's ten quantitative chart fences remain in their original count, order, kinds, item inventories, and display options. This packet does not reproduce every baseline chart item, so no slot is replaced or expanded and no missing observation is filled with zero.

Latest evidence update — 2 October 2026

The 2 October packet is a time-bounded premarket overlay on the 1 October capacity and GPU ledger. It adds no data-center, electricity-capacity, GPU-inventory, investor backer, or hardware-price row. The indicative quote screen leans constructive for broad technology and AI infrastructure, but its fields are bid ask observations and derived midpoints rather than executable trades. This is analysis only, not an instruction to place a trade.

Market context and key observed levels

The strongest indicative gaps in the selected liquid screen were $AMD +2.212%, $NVDA +1.899%, $SOXX +1.701%, $SMH +1.517%, and $AVGO +1.289%; broad proxies showed $QQQ +0.662% and $SPY +0.467%. The macro proxies showed $USO -3.876%, $TLT +0.238%, and $VIXY -1.054%. $GOOGL was +0.506% and $META +0.273% on the same midpoint calculation. These are observed levels at the 08:22–08:24 ET pull, not support resistance targets or trade signals.

The daily context through 1 October was constructive but mixed: $QQQ, $SOXX, and $NVDA remained above their 10- and 20-day averages, while $SPY was below its 10-day average and its RSI14 was near 50. The four indicator series had 43 candles, so a 50-day average was unavailable. Calendar gaps were flagged as a data-quality caveat; duplicate bars, bad spikes, zero-volume bars, and invalid timestamps were not reported.

Catalysts and data-center context

• The September Employment Situation was scheduled for 08:30 ET on 2 October. U.S. Bureau of Labor Statistics
• Reuters attributed the higher-futures backdrop to easing Treasury yields and oil prices ahead of the jobs report.
• Micron search-result evidence described record fiscal Q4 and full-year 2026 results, including Q4 revenue of $54.23B versus $41.46B in the prior quarter, and AI-driven demand.
• A TipRanks search result associated higher AI-chip premarket trading with Micron’s results outlook and company-specific catalysts. This is secondary context, not proof of causality.

The 1 October project and GPU register remains the decision-useful capacity view: project MW GW, grid or generation, IT load, compute capacity, GPU state, company role, backing, and price type remain separate. The current market overlay does not convert those records into a global capacity or installed-GPU total.

Risks and confidence

Confidence in the directional premarket read is low. The sampled fields are bid ask indications, not prints; returned last and previous-close values were flat; no dedicated extended-hours trade-print feed or numeric index-futures level was separately verified; and the cited headlines do not establish quote causality. $DLR and $EQIX had materially wide or older quote fields and were excluded from directional interpretation. The employment release was still a scheduled catalyst at observation time, so no post-release reaction is included.

Source list

• Dated bid ask snapshot observed 2 October 2026 at 12:22–12:24 UTC; midpoint, gap, spread, and quote-quality labels are derived fields.
• U.S. Bureau of Labor Statistics — October 2026 release schedule
• Reuters — Wall St futures gain as yields, oil prices ease ahead of jobs report
• Micron — fiscal Q4 and full-year 2026 results
• TipRanks — AI chip stocks in premarket on 2 October

Latest evidence update — 5 October 2026

The current primary-source packet contains 12 project or platform rows and 7 GPU or hosted-service rows across North America, Europe, Asia-Pacific, the Middle East, Latin America, and Africa. The clearest new operating and financing signal is Crusoe’s corporate disclosure of more than 6 GW of gross contracted capacity, 1 GW delivered and operational, a US$3.9B Series F, and a US$30.9B post-money valuation. That corporate record is separate from the Abilene project row. The packet still does not support a comparable global capacity, electricity, installed-GPU, invested-capital, or hardware-purchase total.

Current project and construction register

Region Project Status Capacity signal Capacity basis Electricity grid boundary Company role Investor backer role Capital price object Evidence --- --- --- --- --- --- --- --- --- --- North America Microsoft Pecos datacenter campus Announced planned About 2 GW of added global datacenter capacity Microsoft-described addition; not separately identified as IT load Dedicated onsite energy planned; generation and grid MW ND Microsoft developer, operator, and funder of its onsite energy support External project financier ND Multibillion-dollar investment over five to seven years; exact amount ND Microsoft North America Crusoe Abilene Stargate flagship campus First phase operating reported; full buildout in progress Eight-building campus designed for hundreds of thousands of GPUs at completion; campus MW ND Campus design signal; not an installed-GPU count First two buildings energized; grid, generation, and IT-load MW ND Crusoe and Oracle Cloud Infrastructure; OpenAI named principal customer Project equity allocation ND; customer role is not project financing Project capex and customer pricing ND Crusoe live campus North America Galaxy Helios, West Texas Construction underway, first phase More than 1.6 GW of approved and utility-contracted power ERCOT-approved and utility-contracted power; nested approvals AEP Texas service agreement; WETT transmission provider; IT-load and generation split ND Galaxy Digital developer; CoreWeave long-term lessee customer Project equity allocation ND; utility is a counterpart, not an investor Project capex ND Galaxy Helios North America Monarch Compute Campus, West Virginia Equipment delivery September 2026–August 2027; online target 2027 2 GW dedicated generation order Caterpillar generator-set order; not IT load or grid capacity Fast-response natural gas plus battery storage; delivery and commissioning remain future checkpoints American Intelligence & Power; Caterpillar and Boyd CAT supply alliance Capital provider ND Purchase-order value ND Caterpillar Monarch Europe Altarea–Vantage AI and cloud campus, France Planned partnership 400 MW secured power connection Power-connection signal; IT-load split ND French utility connection secured; generation, grid detail, and annual energy ND Vantage developer; Altarea land development partner Project financing ND Project capex ND Vantage Altarea Europe Digital Realty FRA20, Frankfurt Construction started 16 MW installed IT capacity; 200 MW park is a larger nested scope FRA20 facility IT load; Digital Park Fechenheim envelope not added 100% renewable-energy matching described; grid rating and annual energy ND Digital Realty operator developer Project financing ND Project capex ND Digital Realty FRA20 Europe Mistral AI compute cluster at Digital Realty Paris South Partnership deployment announced 10 MW compute cluster Hosted compute-cluster signal; not a disclosed GPU count Power source and grid rating ND Digital Realty host; Mistral AI customer operator Project financing ND Project capex and customer pricing ND Digital Realty Mistral Asia-Pacific AirTrunk JHB3 JHB4, Malaysia Announced development More than 280 MW new IT load; 700 MW+ Malaysia platform is nested Source-labelled IT load for the two new campuses Energy-efficiency and recycled-water cooling described; grid rating and annual energy ND AirTrunk developer operator Cloud and AI customers are demand sources; equity backer ND US$3B new development; about US$6.8B total Malaysia investment AirTrunk Malaysia Asia-Pacific AirTrunk TOK1, East Tokyo Financed and under construction More than 100 MW additional IT load under construction; scalable to more than 300 MW Near-term additional IT load; Japan platform about 530 MW is nested Green-loan-funded development capex; grid rating and annual energy ND AirTrunk developer operator 12-bank lending group led by SMBC, MUFG, Crédit Agricole CIB, and Société Générale; equity allocation ND US$1.24B green loan; Japan investment greater than US$8B AirTrunk TOK1 Middle East Stargate UAE UAE–US AI Campus Construction reported; current status unverified 1 GW AI infrastructure cluster inside a 5 GW campus Nested cluster campus figures; IT-load basis ND Generation source, grid interconnection rating, and annual energy ND G42 and Khazna Data Centers; OpenAI, Oracle, NVIDIA, Cisco, and SoftBank named partners Project equity allocation ND; partner roles are not funding amounts Capital amount ND G42 search evidence — search-result-only in this run Latin America Scala AI City, Brazil Letter of intent planned background 54 MW initial IT capacity First-phase IT-load signal Abundant energy cited; grid interconnection and annual energy ND Scala Data Centers; Rio Grande do Sul government enabling partner Coatue and IMCO provided US$250M each in structured financing; Scala described as backed by DigitalBridge US$500M first phase; US$90B is potential full realization, not committed capex Scala project · Scala financing Africa Tscale Project Endeavour, Lagos Developer says building; first-building target H2 2026 1.4 GW “AI computing power”; ten buildings; 140 MW per phase Developer headline; phase unit is not explicitly labelled IT load Dedicated renewable generation and onsite natural gas claimed; grid interconnection and annual energy ND Tscale developer Backers not disclosed on the project page Project capex and customer pricing ND Tscale Project Endeavour

Corporate capacity and financing kept outside site totals

Company scope Capacity and operating signal Capital valuation signal Investor backer role Boundary Evidence --- --- --- --- --- --- Crusoe corporate vertically integrated platform More than 6 GW gross contracted capacity; 1 GW gross capacity delivered and operational today; US$140B+ total contracted value US$3.9B Series F initial close; US$30.9B post-money valuation Co-led by Atreides Management, Mubadala Capital, and Valor Equity Partners; named additional backers include Founders Fund, GIC, NVIDIA, QIA, Radical Ventures, and TPG Corporate platform record; not an Abilene site total or installed-GPU count Crusoe Series F

Current GPU and hosted-service register

GPU model service State and quantity Specification Company role Investor backer role Price capital object Evidence --- --- --- --- --- --- --- NVIDIA H100 CoreWeave catalogued instance: 8 GPUs; global fleet ND Hopper; 80 GB HBM3; NVLink plus 400G NDR InfiniBand NVIDIA OEM; CoreWeave cloud operator CoreWeave is public; project-level investor allocation ND AWS P5 Capacity Block at US$5.191 per H100-hour is a comparable reservation anchor, not a CoreWeave or hardware price NVIDIA H100 · AWS Capacity Blocks NVIDIA H200 Model and cloud offering; fleet count ND Hopper; 141 GB HBM3e; 4.8 TB s bandwidth NVIDIA OEM; AWS and CoreWeave offerings Project-level investor or fleet-owner role ND No exact H200 hardware price verified; AWS P5e P5en pricing varies by region and billing mode NVIDIA H200 · AWS P5 NVIDIA B200 CoreWeave catalogued instance: 8 GPUs; global fleet ND Blackwell; 180 GB HBM3e; NVLink NVLink Switch plus 400G NDR InfiniBand NVIDIA OEM; CoreWeave cloud operator CoreWeave is public; project-level investor allocation ND AWS P6-B200 Capacity Block at US$12.355 per B200-hour is a comparable reservation anchor, not a CoreWeave or hardware price CoreWeave B200 · AWS Capacity Blocks NVIDIA GB200 NVL72 Rack reference: 72 GPUs and 36 Grace CPUs; installed count ND Liquid-cooled rack-scale design; 130 TB s NVLink domain NVIDIA OEM; AWS and Crusoe Lancium platform references Rack-level investor or buyer count ND AWS P6e GB200x72 at US$10.582 per accelerator-hour is a related rental reference, not an OEM quote NVIDIA GB200 · AWS Capacity Blocks NVIDIA Vera Rubin NVL72 One fully validated operational system announced by CoreWeave; global fleet ND 72 Rubin GPUs; 36 Vera CPUs; 260 TB s NVLink NVIDIA OEM; CoreWeave cloud operator CoreWeave is public; project-level investor allocation ND Rental and purchase price ND CoreWeave Vera Rubin AMD Instinct MI350X Product catalog entry; shipped deployed fleet count ND 4th Gen AMD CDNA; 288 GB HBM3E; 8 TB s; 4.6 FP16 BF16 matrix PFLOPS and 9.2 FP8 matrix PFLOPS AMD OEM Fleet owner and project backer ND No reliable current OEM purchase or per-GPU rental price verified AMD MI350X AWS EC2 P5 P5e P5en Service ceiling up to 20,000 H100 or H200 GPUs per UltraCluster; not deployed inventory Up to 3,200 Gbps EFA and up to 20 exaflops per AWS description AWS Amazon operator; NVIDIA supplier Amazon funds the service; separate project investor ND Exact on-demand rate varies by region, OS, and billing mode AWS P5

Electricity, energy, capital, and rental checkpoints

Benchmark object Observed value Type and date Boundary Evidence --- --- --- --- --- AWS P5 Capacity Block US$5.191 per H100-hour; US$5.970 effective 7 October 2026 Dynamic reservation rate; current table and scheduled update captured 5 October 2026 Hosted reservation, not on-demand, hardware purchase, or project capex AWS Capacity Blocks pricing AWS P6-B200 Capacity Block US$12.355 per B200-hour; US$14.208 effective 7 October 2026 Dynamic reservation rate; current table and scheduled update captured 5 October 2026 Hosted reservation, not hardware price or installed inventory AWS Capacity Blocks pricing AWS P6e GB200x72 Capacity Block US$10.582 per accelerator-hour; 72 accelerators per ultraserver Rack-scale reservation reference captured 5 October 2026 Access price, not OEM rack price or site electricity cost AWS Capacity Blocks pricing United States industrial electricity 9.77¢ kWh = US$97.70 MWh July 2026 preliminary EIA average industrial retail revenue Retail-revenue proxy, not data-center tariff, wholesale node, PPA, or connection fee EIA national table Texas industrial electricity 7.07¢ kWh = US$70.70 MWh July 2026 preliminary EIA average industrial retail revenue Retail-revenue proxy, not ERCOT nodal price or data-center tariff EIA state table Louisiana industrial electricity 6.08¢ kWh = US$60.80 MWh July 2026 preliminary EIA average industrial retail revenue Retail-revenue proxy, not site-specific delivered cost EIA state table South Carolina industrial electricity 7.84¢ kWh = US$78.40 MWh July 2026 preliminary EIA average industrial retail revenue Retail-revenue proxy, not site-specific delivered cost EIA state table Hardware purchase price ND for H100, H200, B200, GB200, Vera Rubin, and MI350X No current OEM purchase price or executed quote verified Rental rates and secondary estimates are not purchase prices NVIDIA H100 · AMD MI350X

Current dashboard analytics views

View Evidence markers Unit basis Interpretation --- --- --- --- Approved or generation power Galaxy Helios 1,600; Monarch 2,000; Vantage Altarea 400; Stargate UAE 1,000 MW; approved power, generation order, secured connection, and AI-cluster signal kept separate Scale markers only; do not sum unlike power boundaries or treat them as IT load Source-labelled IT-load signals Digital Realty FRA20 16; Mistral Paris South 10; AirTrunk JHB3 JHB4 280; AirTrunk TOK1 near-term 100; Scala phase one 54 MW; installed IT, compute cluster, new IT load, under-construction IT load, and first-phase IT capacity Non-additive project markers with explicit source bases Global energy context IEA data-center consumption 415 in 2024 and around 945 in 2030; supply requirement over 1,000 in 2030 and 1,300 in 2035 TWh; system-level consumption or supply scenarios Context only; not project MW, grid capacity, or firm supply GPU rental checkpoints AWS H100 5.191; B200 12.355; GB200 B200 rack-scale 10.582 US$ per accelerator-hour; reservation rates Dynamic access prices, not hardware prices, installed inventory, or project economics

The analytics views are evidence markers rather than totals. Parent programs, child sites, corporate capacity, power approvals, generation orders, IT-load observations, global energy scenarios, rack references, service ceilings, and rental rates remain separate.

Method and unresolved risks

• No current source discloses a defensible global deployed-GPU fleet total or common-denominator global capacity total.
• The UAE figure is search-result-only in this run; Tscale’s capacity and energy wording is developer-reported and does not establish IT-load, grid, generation, or annual-energy equivalence.
• Current energization and commissioning were not independently rechecked for every project; announced, financed, under-construction, energized, and operating states remain distinct.
• Hardware purchase prices remain ND. AWS rates are dynamic capacity-block reservations; EIA values are preliminary retail-revenue averages rather than site tariffs or delivered cost.
• Existing ten quantitative chart fences remain in their baseline count, order, kinds, item inventories, and display options; no missing observation was filled with zero and no new chart slot was introduced.

Latest evidence update — 6 October 2026

The 6 October overlay refreshes market context only; it adds no project or GPU rows. The accepted ledger remains 12 project platform records and 7 GPU or hosted-service records across six regions. At 08:19:38–08:20:43 EDT, 26 of 30 sampled symbols had indicative bid ask midpoints above their returned previous closes; two were lower and two were excluded for stale or very wide quotes. This is a low-confidence sentiment screen, not an executable return or a trading instruction.

Market context and key observed levels

The broad premarket backdrop was constructive: S&P 500 futures were indicated at +0.50%, Dow futures at +0.67%, Nasdaq futures at +0.65%, and Russell 2000 futures at +0.59%. November crude was indicated at $87.06, down 2.65%; VIX was 15.32, down 1.29%; and the 10-year Treasury yield was 5.31%, up 0.03 percentage points. Lower oil and volatility supported risk appetite, while the higher yield remained a valuation risk.

Instrument 1 Oct daily close SMA10 SMA20 RSI14 2-month change --- ---: ---: ---: ---: ---: $SPY 763.99 767.13 764.91 49.7 +0.83% $QQQ 742.03 739.35 726.18 67.75 +5.99% $SOXX 576.33 564.29 539.36 66.81 +13.52% $NVDA 230.86 226.90 223.51 66.07 +11.72%

These are daily reference levels through 1 October, not support or resistance targets. The series contains 43 candles, so a 50-day average is unavailable; calendar gaps remain a data-quality caveat.

The strongest current indicative gaps in the sampled basket were:

Instrument Indicative midpoint gap Quote quality --- ---: --- $CEG +8.947% Moderate $VST +4.386% Moderate $APLD +2.713% Tight $GLXY +2.075% Tight $AMD +1.852% Moderate $CRWV +1.659% Tight $SMCI +1.482% Tight $STX −2.941% Wide $WDC −3.076% Moderate

The midpoint gap is derived from bid and ask against the returned previous close. It is not a trade return. $DLR and $EQIX were excluded from directional interpretation because of stale or unusually wide quote fields.

Catalysts and data-center context

• The 5 October project and GPU register remains the decision-useful capacity view: project MW GW, grid or generation, IT load, compute capacity, GPU state, company role, backing, and price type remain separate. No 6 October project, construction, corporate-capacity, or GPU-inventory addition was accepted.
• A Yahoo Finance premarket article reported AI and semiconductor leadership, Broadcom-related financing for Anthropic’s Google-TPU lease, and AMD’s expectation that compute demand will outpace supply with capacity expansion by 2027. These are attributed article claims, not independent evidence of project capacity or installed inventory.
• A PreMarketPrice mover screen separately displayed $CEG +7.87%, $VST +4.35%, and $WDC −2.64%; this secondary screen is directionally consistent with the accepted quote snapshot but is not used as a substitute for it.
• A Reuters market report and a TipRanks result provided additional premarket context. The TipRanks item remains lower-confidence secondary evidence, and neither source establishes quote causality.

Risks and confidence

Confidence in the directional premarket read is low. Bid ask midpoints are indicative rather than executed prices; post-open confirmation was not available at the observation time; and no source establishes that the observed moves were caused by any one headline. The 5.31% 10-year yield remains a rates and valuation risk. The data-center evidence still does not support a global installed-GPU total or a common-denominator global MW total, and the project, generation, grid, IT-load, compute, corporate-capacity, and rental-price fields remain non-additive.

Source list

• Dated bid ask snapshot observed 6 October 2026 at 08:19:38–08:20:43 EDT; midpoint, gap, spread, and quote-quality labels are derived fields.
• Yahoo Finance — market live
• Yahoo Finance — premarket context
• PreMarketPrice — premarket movers
• TipRanks — AI chips in premarket
• Reuters — futures, yields and oil

Report details

Run date: 6 October 2026 UTC · Evidence cutoff read: 6 October 2026 UTC for the premarket overlay; 5 October 2026 UTC for the current project, GPU, electricity, and rental packet; 2 October 2026 UTC for the retained premarket overlay · Status: bounded publication candidate awaiting evaluation; not published or promoted · Owner: General Codex Agent · Run: looprun-72cab2ab-4663-4949-82a0-808edcb925b1 · Session: loopsession-5690b416-70f0-4762-b865-a9d8d9cda9bb · Continuity mode: incremental; baseline report read from commit f8a1ccf642b82e95ed371fdc210cca3aa69a5850 · Evidence dependency: accepted current research and data ledger plus retained historical sections · Closure: 6 October premarket overlay added; existing ten quantitative chart fences preserved; publication remains outside this session.