Top 50
The most expensive AI data centers ever built
Ranked by total capital cost in 2025 dollars — compute, construction and everything around it.
Colossus 2 leads with $36B, ahead of Anthropic-Amazon New Carlisle ($34B) and Microsoft Fairwater Atlanta ($24B).
Most of this money is not the building. On large AI sites the chips typically dominate the capital cost, which is why capex tracks compute far more closely than it tracks floor area.
| # | Name | Owner | Country | Capital cost |
|---|---|---|---|---|
| 01 | Colossus 2 | SpaceXAI | United States | $36B |
| 02 | Anthropic-Amazon New Carlisle | Amazon | United States | $34B |
| 03 | Microsoft Fairwater Atlanta | Microsoft | United States | $24B |
| 04 | Meta Prometheus | Meta | United States | $24B |
| 05 | OpenAI Stargate Abilene | Oracle | United States | $16B |
| 06 | Microsoft Fairwater Wisconsin | Microsoft | United States | $14B |
| 07 | Colossus 1 | SpaceXAI | United States | $13B |
| 08 | Google New Albany | United States | $13B | |
| 09 | Google Columbus | United States | $11B | |
| 10 | Amazon Madison Mega Site | Amazon | United States | $11B |
| 11 | Google Bristow | United States | $11B | |
| 12 | CoreWeave Denton TX | CoreWeave | United States | $9.9B |
| 13 | Huawei Horinger | Huawei | China | $9.2B |
| 14 | DayOne Nusajaya | — | Malaysia | $9.1B |
| 15 | QTS Richmond 1 | — | United States | $9.0B |
| 16 | Google Council Bluffs (East) | United States | $9.0B | |
| 17 | Google Papillion | United States | $9.0B | |
| 18 | Google Omaha | United States | $9.0B | |
| 19 | Amazon Ridgeland | Amazon | United States | $8.6B |
| 20 | VNET Bayin Ulanqab | VNET | China | $8.4B |
| 21 | Microsoft Goodyear | Microsoft | United States | $7.7B |
| 22 | Microsoft Project Osmium | Microsoft | United States | $7.2B |
| 23 | QTS Richmond 2 | — | United States | $6.8B |
| 24 | Meta Rosemount | Meta | United States | $6.7B |
| 25 | Meta Jeffersonville | Meta | United States | $6.7B |
| 26 | Alibaba Zhangbei | Alibaba | China | $6.4B |
| 27 | Google Storey County | United States | $6.1B | |
| 28 | Google The Dalles | United States | $5.8B | |
| 29 | Meta Montgomery | Meta | United States | $5.8B |
| 30 | Meta Temple | Meta | United States | $5.8B |
| 31 | Meta Kuna | Meta | United States | $5.8B |
| 32 | Meta Cheyenne | Meta | United States | $5.8B |
| 33 | QTS Richmond 3 | — | United States | $5.5B |
| 34 | Meta Aiken | Meta | United States | $5.4B |
| 35 | Google Lincoln | United States | $5.3B | |
| 36 | Google Lancaster | United States | $5.2B | |
| 37 | Meta Eagle Mountain | Meta | United States | $5.0B |
| 38 | Coreweave Helios | CoreWeave | United States | $5.0B |
| 39 | Google Midlothian | United States | $3.9B | |
| 40 | Fluidstack Lake Mariner | Fluidstack | United States | $3.8B |
| 41 | Google Mesa | United States | $3.4B | |
| 42 | Google Waltham Cross | United Kingdom | $3.3B | |
| 43 | Meta Gallatin | Meta | United States | $3.3B |
| 44 | Meta Los Lunas | Meta | United States | $3.3B |
| 45 | STACK Infrastructure NVA02 | — | United States | $3.2B |
| 46 | CoreWeave Chester VA | CoreWeave | United States | $3.1B |
| 47 | Google Red Oak | United States | $2.9B | |
| 48 | Google Arcola | United States | $2.9B | |
| 49 | Microsoft SAT40 | Microsoft | United States | $2.8B |
| 50 | Oracle Batam | Oracle | Indonesia | $2.7B |
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 data centers →Other rankings of ai data centers