Public Kempax report

Daily Update Global Data Center and GPU capacity · 2026-08-11

Kempax Research

Loop report · 2026-08-11

Global data-center capacity, electricity, and GPU inventory — 11 August 2026

As of 2026-08-11 UTC, public evidence shows a rapidly expanding global AI data-center pipeline, but it does not support a defensible global total. The register below keeps operational, under-construction, permitted or design-certified, announced, and speculative or proxy signals separate. IT data-center load is never added to a grid connection, receiving capacity, generation, PPA, MVA, or a campus program envelope. GPU counts are state-qualified and non-additive; Trainium2 and Ironwood are accelerators, not NVIDIA GPUs. This is a research report, not investment advice.

Evidence highlights

The clearest source-labelled capacity signals are Start Campus SINES’ 1.2 GW secured IT capacity, AirTrunk SYD3’s 320+ MW IT load, ODATA QR03’s 300 MW IT plan, Meta Richland’s 5 GW compute description, and Scala AI City’s 5 GW power connection. These figures have different bases and are not additive. Representative accelerator evidence includes AWS Project Rainier’s nearly 500,000 Trainium2 chips, xAI’s conflicting H100 counts, and NVIDIA Noetra’s planned 27,500 Rubin GPUs. Public price signals are dated rentals or secondary proxies, such as Lambda, RunPod, and CoreWeave, not verified OEM purchase prices.