Top 50
最も電力を消費する AI データセンター
メガワット単位の電力需要で並べ替え。
Colossus 2 leads with 946 MW, ahead of Anthropic-Amazon New Carlisle (910 MW) and Microsoft Fairwater Atlanta (636 MW).
The top five account for 31% of the 11.47 GW tracked across this ranking.
Power, not silicon, is increasingly the binding constraint on AI build-out: a site can only be as large as the grid connection it can secure. That is why these numbers are worth watching alongside compute — they translate directly into generation capacity, transmission and, eventually, emissions.
| # | Name | Owner | Country | Power |
|---|---|---|---|---|
| 01 | Colossus 2 | SpaceXAI | United States | 946 MW |
| 02 | Anthropic-Amazon New Carlisle | Amazon | United States | 910 MW |
| 03 | Microsoft Fairwater Atlanta | Microsoft | United States | 636 MW |
| 04 | Meta Prometheus | Meta | United States | 631 MW |
| 05 | OpenAI Stargate Abilene | Oracle | United States | 421 MW |
| 06 | Microsoft Fairwater Wisconsin | Microsoft | United States | 369 MW |
| 07 | Colossus 1 | SpaceXAI | United States | 340 MW |
| 08 | Google New Albany | United States | 339 MW | |
| 09 | Google Columbus | United States | 303 MW | |
| 10 | Amazon Madison Mega Site | Amazon | United States | 284 MW |
| 11 | Google Bristow | United States | 279 MW | |
| 12 | CoreWeave Denton TX | CoreWeave | United States | 262 MW |
| 13 | Huawei Horinger | Huawei | China | 242 MW |
| 14 | DayOne Nusajaya | — | Malaysia | 240 MW |
| 15 | QTS Richmond 1 | — | United States | 238 MW |
| 16 | Google Council Bluffs (East) | United States | 237 MW | |
| 17 | Google Papillion | United States | 237 MW | |
| 18 | Google Omaha | United States | 237 MW | |
| 19 | Amazon Ridgeland | Amazon | United States | 228 MW |
| 20 | VNET Bayin Ulanqab | VNET | China | 221 MW |
| 21 | Microsoft Goodyear | Microsoft | United States | 202 MW |
| 22 | Microsoft Project Osmium | Microsoft | United States | 190 MW |
| 23 | QTS Richmond 2 | — | United States | 180 MW |
| 24 | Meta Rosemount | Meta | United States | 178 MW |
| 25 | Meta Jeffersonville | Meta | United States | 178 MW |
| 26 | Alibaba Zhangbei | Alibaba | China | 169 MW |
| 27 | Google Storey County | United States | 161 MW | |
| 28 | Google The Dalles | United States | 154 MW | |
| 29 | Meta Montgomery | Meta | United States | 153 MW |
| 30 | Meta Temple | Meta | United States | 152 MW |
| 31 | Meta Kuna | Meta | United States | 152 MW |
| 32 | Meta Cheyenne | Meta | United States | 152 MW |
| 33 | QTS Richmond 3 | — | United States | 144 MW |
| 34 | Meta Aiken | Meta | United States | 142 MW |
| 35 | Google Lincoln | United States | 141 MW | |
| 36 | Google Lancaster | United States | 137 MW | |
| 37 | Meta Eagle Mountain | Meta | United States | 133 MW |
| 38 | Coreweave Helios | CoreWeave | United States | 132 MW |
| 39 | Google Midlothian | United States | 103 MW | |
| 40 | Fluidstack Lake Mariner | Fluidstack | United States | 100 MW |
| 41 | Google Mesa | United States | 90 MW | |
| 42 | Google Waltham Cross | United Kingdom | 88 MW | |
| 43 | Meta Gallatin | Meta | United States | 87 MW |
| 44 | Meta Los Lunas | Meta | United States | 87 MW |
| 45 | STACK Infrastructure NVA02 | — | United States | 85 MW |
| 46 | CoreWeave Chester VA | CoreWeave | United States | 82 MW |
| 47 | Google Red Oak | United States | 77 MW | |
| 48 | Google Arcola | United States | 77 MW | |
| 49 | Microsoft SAT40 | Microsoft | United States | 75 MW |
| 50 | Oracle Batam | Oracle | Indonesia | 72 MW |
SourceEpoch AI, 'AI Models'. Published online at epoch.ai. Retrieved 2026-07-29 from https://epoch.ai/data/ai-models. Licensed under CC BY 4.0.
Browse all ai データセンター →Other rankings of ai データセンター