Hyperscaler List

Hyperscaler AI capacity in gigawatts

Gigawatts are the new region count. Here is the AI capacity each operator has announced, in its own words, with the date and the source.

7 operators · checked 2026-11-11 · method

Microsoft AzureFairwater Wisconsin and Atlanta: hundreds of thousands of NVIDIA Blackwell GPUs per siteND GB200 v6 VMs on NVIDIA GB200 NVL72 racksUSD 45.2bn incl. finance leases in FY2027 Q1 (Jul to Sep 2026)2026-11-11
Oracle Cloud Infrastructure (OCI)Up to 4.5 GW of additional Stargate capacity for OpenAI (agreed Jul 2025)OCI Supercluster: up to 131,072 NVIDIA B200 GPUsUSD 28.5bn in FY2027 Q1 (Jun to Aug 2026)2026-11-11
Google CloudVisakhapatnam AI hub, India: gigawatt-scale, three campuses (ground broken Apr 2026)Cloud TPU v6e (Trillium): 256 chips per podUSD 51.4bn (Alphabet) in Q3 2026; USD 195bn to 205bn guided for 20262026-11-11
Amazon Web Services (AWS)3.9 GW of new power added in 2025; total power capacity to double by end-2027Trainium3 accelerators; Neuron SDK; GPU instancesUSD 77.3bn PP&E purchases in Q3 2026; ~USD 220bn cash capex guided for 20262026-11-11
CoreWeave1.7 GW+ active power; 4.6 GW+ contracted (Q3 2026)NVIDIA GPU clusters for AI training and inferenceUSD 9.4bn PP&E purchases in Q3 2026; USD 37bn to 39bn guided for 20262026-11-11
MetaPrometheus (Ohio) 1 GW from 2026; Hyperion (Louisiana) up to 5 GWPrometheus supercluster: hundreds of thousands of GPUs deployedUSD 34.9bn incl. finance-lease principal in Q3 2026; USD 135bn to 145bn guided for 20262026-11-11
NAVER CloudAI factory with NVIDIA: 55 MW in H1 2027, 100 MW later in 2027, 200 MW by 2028KVM GPU Server: NVIDIA A100 and H100—2026-11-11

Campus-by-campus megawatts for the sites behind these numbers are on AI Data Center List.

Operator profiles

  • Amazon Web Services (AWS)

    39 regions and 124 availability zones, Trainium chips for AI, and a separate sovereign cloud for Europe.

  • CoreWeave

    An AI-only GPU cloud with over 1.7 GW active and 4.6 GW contracted at the end of Q3 2026, with USD 37bn to 39bn of 2026 capex.

  • Google Cloud

    43 regions and 130 zones on Google-owned infrastructure, TPU pods for AI, and a gigawatt-scale AI hub under construction in India.

  • Meta

    Meta builds for its own apps and models: Prometheus in Ohio is a 1 GW AI cluster and Hyperion in Louisiana scales to 5 GW.

  • Microsoft Azure

    Microsoft lists 59 Azure regions with region codes, advertises 80+ regions and 500+ datacenters, and links its Fairwater AI sites into one AI superfactory.

  • NAVER Cloud

    Korea’s NAVER Cloud serves 10 countries, runs NVIDIA A100 and H100 GPU servers and is building a 200 MW AI factory by 2028.

  • Oracle Cloud Infrastructure (OCI)

    OCI advertises 50+ public regions in 28 countries, runs EU sovereign regions in Frankfurt and Madrid, and is building up to 4.5 GW of Stargate capacity for OpenAI.