Home/AI datacenters
Section
AI datacenters, beyond the GPU count
The dataset holds 280 AI training and inference datacenters and clusters. 168 of them have an accelerator count, 223 have at least one figure from the AI metrics table, and only 14 state a PUE. That gap is the finding: no independent body measures AI datacenters on a common workload, so there is no ranked list, and almost every number here is stated by the operator or estimated by someone else. This section records those numbers with who stated them, and the benchmarks page covers what is measured at system scale (MLPerf, InferenceX), what operators publish about efficiency and failures, and how far each can be trusted.
Nothing on this page is a ranking. Accelerator counts, megawatts and investment figures are not comparable unless they cover the same scope, and they rarely do: IT load, total campus power and a planned build-out are different quantities. Every figure carries its basis (operator stated, regulatory filing, third-party estimate or design target) on the system's page.
Explore
Compute deals
Who buys capacity from whom: cloud contracts, leases, chip supply and joint ventures.
Power sources
Grid, gas, nuclear, renewables and storage behind each campus.
Accelerators
Dense peak FLOPS, memory and power per chip, and derived cluster peaks.
Economics
Cost per megawatt, power per accelerator and announcement-to-first-light.
Benchmarks
Why no ranking exists and what MLPerf and others measure.
Status check
Planned and under-construction campuses whose records may be stale.
- Big tech
- AI labs
- GPU clouds
- Other operators
Table view: Stated power of AI datacenters
| System | Country | Sector | MW |
|---|---|---|---|
| Isambard-AI Phase 2 | United Kingdom | Research HPC | 5 |
| ABCI 3.0 | Japan | Research HPC | 6 |
| xAI Colossus | USA | AI labs | 150 |
| Stargate Abilene | USA | Big tech | 590 |
| Microsoft Fairwater (Atlanta) | USA | Big tech | 859 |
| Microsoft Fairwater (Wisconsin) | USA | Big tech | 442 |
| Project Rainier | USA | Big tech | 1,092 |
| Stargate UAE | United Arab Emirates | Neoclouds | 1,000 |
| Google TPU (Ironwood) capacity for Anthropic | USA | Big tech | 1,000 |
| Microsoft Azure AI supercomputer (Loughton, UK) | United Kingdom | Big tech | 90 |
| Microsoft Azure AI datacenter (Narvik, Norway) | Norway | Big tech | 230 |
| Mistral Campus AI (Bruyeres-le-Chatel) | France | AI labs | 44 |
| Meta Prometheus | USA | Big tech | 612 |
| Meta Hyperion | USA | Big tech | 2,260 |
| xAI Colossus 2 | USA | AI labs | 1,276 |
| Opera Iceland KEF-1 SuperPOD | Iceland | Big tech | 83 |
| Crusoe Abilene Campus (Microsoft) | USA | Neoclouds | 900 |
| SINES Data Campus GB300 Deployment | Portugal | Big tech | 1,200 |
| Firebird Armenian AI Factory (DC-1) | Armenia | Neoclouds | 300 |
| USIO | Uganda | Neoclouds | 100 |
| Cerebras Guyana AI Data Center | Guyana | Industry | 100 |
| Lambda Kansas City AI Factory | USA | Neoclouds | 100 |
| IREN Childress Horizons 1-4 (Microsoft) | USA | Neoclouds | 750 |
| Polaris Forge 1 (Ellendale) | USA | Neoclouds | 250 |
| Stargate Shackelford County (Frontier) | USA | Big tech | 1,400 |
| Stargate Milam County | USA | Neoclouds | 1,200 |
| Stargate Project Jupiter (New Mexico) | USA | Big tech | 900 |
| Stargate Port Washington (Lighthouse) | USA | Big tech | 1,000 |
| Taiwan AI Factory | Taiwan | Neoclouds | 25 |
| du NextGenAI Supercluster | United Arab Emirates | Big tech | 13 |
| Galaxy Helios (CoreWeave) | USA | Neoclouds | 200 |
| River Bend (Fluidstack) | USA | Neoclouds | 330 |
| Barber Lake (Fluidstack) | USA | Neoclouds | 244 |
| Polaris Forge 2 (Harwood) | USA | Neoclouds | 200 |
| Abernathy (Fluidstack) | USA | Neoclouds | 168 |
| CoreWeave Lancaster AI Data Center | USA | Neoclouds | 300 |
| Nebius Kansas City Cluster | USA | Neoclouds | 40 |
| Nebius Mäntsälä Cluster | Finland | Neoclouds | 75 |
| Nebius Lappeenranta AI Factory | Finland | Neoclouds | 310 |
| Nebius Independence AI Factory | USA | Neoclouds | 1,200 |
| Nebius Vineland (Microsoft) | USA | Neoclouds | 300 |
| TELUS BC Sovereign AI Cluster | Canada | Big tech | 150 |
| Campus AI Fouju | France | Neoclouds | 1,400 |
| SoftBank Tomakomai AI Data Center | Japan | Big tech | 300 |
| Yotta D2 Blackwell Ultra Cluster | India | Neoclouds | 60 |
| Nebius Modiin Cluster | Israel | Neoclouds | 8 |
| HUMAIN AI Factory Phase 1 | Saudi Arabia | Neoclouds | 500 |
| Coatlicue | Mexico | Research HPC | 20 |
| HUMAIN Qualcomm AI 100 Cluster | Saudi Arabia | Neoclouds | 200 |
| Meta Lebanon (Indiana) | USA | Big tech | 1,000 |
| Meta El Paso (Texas) | USA | Big tech | 1,000 |
| Justified Data (Anthropic) | USA | AI labs | 401 |
| Google Pryor North TPU Campus | USA | Big tech | 368 |
| Google Columbus TPU Campus | USA | Big tech | 303 |
| Google New Albany TPU Campus | USA | Big tech | 333 |
| Google Council Bluffs East TPU Campus | USA | Big tech | 237 |
| AWS Madison Mega Site (Canton) | USA | Big tech | 284 |
| AWS Ridgeland Campus | USA | Big tech | 228 |
| Microsoft Project Osmium | USA | Big tech | 190 |
| Microsoft Goodyear AI Datacenter | USA | Big tech | 202 |
| Meta Rosemount (Minnesota) | USA | Big tech | 178 |
| Meta Montgomery (Alabama) | USA | Big tech | 153 |
| Meta Cheyenne (Wyoming) | USA | Big tech | 152 |
| xAI Atlanta | USA | AI labs | 17 |
| Muskogee (CoreWeave) | USA | Neoclouds | 100 |
| CoreWeave Kenilworth (NEST) | USA | Neoclouds | 250 |
| Bell AI Fabric Saskatchewan | Canada | Big tech | 300 |
| CoreWeave Cambridge (Cohere) | Canada | Neoclouds | 64 |
| Scala AI City | Brazil | Neoclouds | 54 |
| Omnia Pecem (TikTok) | Brazil | Neoclouds | 300 |
| Project Jade (Cheyenne) | USA | Industry | 2,700 |
| Delta Forge 1 (Rapides Parish) | USA | Neoclouds | 430 |
| Cipher Black Pearl (AWS) | USA | Neoclouds | 300 |
| Nebius Birmingham AI Factory | USA | Neoclouds | 300 |
| Wonder Valley AI Campus | Canada | Neoclouds | 7,500 |
| Nscale Verne Iceland Cluster | Iceland | Neoclouds | 15 |
| Nebius Iceland Cluster | Iceland | Neoclouds | 10 |
| DataVolt Oxagon AI Campus | Saudi Arabia | Neoclouds | 360 |
| Khazna Ajman AI Data Center | United Arab Emirates | Neoclouds | 100 |
| STACKIT Lubbenau AI Data Center | Germany | Big tech | 200 |
| SK Ulsan AI Data Center | South Korea | Big tech | 40 |
| Racks Central RCJM AI Campus | Malaysia | Neoclouds | 510 |
| SoftBank Sakai AI Data Center | Japan | Big tech | 150 |
| NEXTDC S7 Sydney AI Campus | Australia | Neoclouds | 550 |
| Mila Sovereign AI Research Hub | Canada | Industry | 3 |
| Cerebras Montreal Inference Site (MTL-3) | Canada | Industry | 5 |
| GroqCloud Sydney Inference Cluster | Australia | Industry | 5 |
| GroqCloud Kamloops Inference Cluster | Canada | Industry | 7 |
| Poolside Project Horizon | USA | AI labs | 2,000 |
| HUMAIN Project Halo | Saudi Arabia | Neoclouds | 2,000 |
| Killellan AI Growth Zone | United Kingdom | Neoclouds | 600 |
| NVIDIA DGX SuperPOD (A100) | USA | Industry | 0 |
The systems
Click a header to sort. A dash means no figure has been recorded, not that none exists.
| System | Operator | Country | Status | Accelerators | IT MW | Facility MW | Claimed MW | PUE | WUE | Investment ($bn) |
|---|---|---|---|---|---|---|---|---|---|---|
| Google TPU (Ironwood) capacity for Anthropic | Alphabet | USA | under construction | 1,000,000 plan | - | 1,000 plan | - | - | - | - |
| IREN Sweetwater Hub | IREN | USA | under construction | 700,000 plan | - | 2,000 | 2,000 | - | - | - |
| Google New Albany TPU Campus | Alphabet | USA | operational | 576,400 est. | 333 est. | - | - | 1.07 | - | 12.6 est. |
| Project Rainier | Amazon | USA | operational | 500,000 | 910 est. | 1,092 est. | - | - | - | 11 |
| Google Columbus TPU Campus | Alphabet | USA | operational | 465,700 est. | 303 est. | - | - | 1.06 | - | 11.5 est. |
| AWS Madison Mega Site (Canton) | Amazon | USA | operational | 326,600 est. | 284 est. | - | - | - | - | 10.8 est. |
| Google Council Bluffs East TPU Campus | Alphabet | USA | operational | 305,600 est. | 237 est. | - | - | - | - | 9 est. |
| Microsoft Fairwater (Atlanta) | Microsoft | USA | operational | 304,279 est. | 636 est. | 859 est. | - | - | - | 24.1 est. |
| AWS Ridgeland Campus | Amazon | USA | operational | 261,200 est. | 228 est. | - | - | - | - | 8.6 est. |
| xAI Colossus 2 | xAI | USA | operational | 220,000 | 430 | 1,276 est. | 1,000 | - | - | 20 |
| Stargate Abilene | Oracle | USA | operational | 201,600 est. | 421 est. | 590 est. | 1,200 | 1.3 plan | - | 15 |
| Microsoft Fairwater (Wisconsin) | Microsoft | USA | operational | 176,391 est. | 369 est. | 442 est. | - | - | - | 7 |
| Meta Prometheus | Meta Platforms | USA | operational | 166,015 est. | 471 est. | 612 est. | 1,000 | - | - | 17.8 est. |
| Microsoft Goodyear AI Datacenter | Microsoft | USA | operational | 116,400 est. | 202 est. | - | - | - | - | 7.7 est. |
| Microsoft Project Osmium | Microsoft | USA | operational | 103,200 est. | 190 est. | - | - | - | - | 7.2 est. |
| Google Pryor North TPU Campus | Alphabet | USA | operational | 101,400 est. | 368 est. | - | - | 1.11 | - | 13.9 est. |
| Denton (CoreWeave) | CoreWeave | USA | operational | 100,000 est. | 260 | 367 est. | 260 | - | - | 9.9 est. |
| STACKIT Lubbenau AI Data Center | Schwarz Digits | Germany | under construction | 100,000 plan | - | 200 plan | 200 | - | - | - |
| xAI Colossus | xAI | USA | operational | 100,000 | 130 | 150 | 150 | - | - | 12.9 est. |
| Solstice | Argonne National Laboratory | USA | planned | 100,000 plan | - | - | - | - | - | - |
| JD Cloud 100,000-GPU Cluster | JD.com | China | planned | 100,000 plan | - | - | - | - | - | - |
| Meta Rosemount (Minnesota) | Meta Platforms | USA | operational | 85,400 est. | 178 est. | - | - | - | - | 0.8 |
| Firebird Armenian AI Factory (DC-1) | Firebird | Armenia | operational | 70,000 plan | - | - | 300 | - | - | - |
| Meta Montgomery (Alabama) | Meta Platforms | USA | operational | 69,100 est. | 153 est. | - | - | - | - | 1.5 |
| Meta Cheyenne (Wyoming) | Meta Platforms | USA | operational | 68,400 est. | 152 est. | - | - | - | - | 0.8 |
| Galaxy Helios (CoreWeave) | CoreWeave | USA | operational | 63,151 est. | 133 | 200 | 1,630 | - | - | 5 est. |
| TELUS BC Sovereign AI Cluster | TELUS | Canada | under construction | 60,000 plan | - | 150 plan | 150 | - | - | - |
| Nebius Mäntsälä Cluster | Nebius | Finland | operational | 60,000 plan | - | 75 plan | 75 | 1.1 | 0.02 | - |
| SK Ulsan AI Data Center | SK Telecom | South Korea | under construction | 60,000 plan | - | 40 plan | 100 | - | - | - |
| Tesla Cortex | Tesla | USA | operational | 50,000 | - | - | - | - | - | - |
| Hyundai Motor Group Blackwell AI Factory | Hyundai Motor Group | South Korea | planned | 50,000 plan | - | - | - | - | - | - |
| Polaris Forge 1 (Ellendale) | CoreWeave | USA | operational | 36,900 est. | 250 | - | 1,000 | 1.18 plan | - | 2.6 est. |
| Muskogee (CoreWeave) | CoreWeave | USA | under construction | 36,000 est. | 70 | 100 | 70 | - | - | - |
| Together AI / Hypertec Cloud GB200 Cluster | Together AI | USA | operational | 36,000 plan | - | - | - | - | - | - |
| Nebius Kansas City Cluster | Nebius | USA | operational | 35,000 plan | - | - | 40 | - | - | - |
| Google TPU v4 ML Hub (Oklahoma) | Alphabet | USA | operational | 32,768* | - | - | - | 1.11 | - | - |
| Microsoft Azure AI datacenter (Narvik, Norway) | Microsoft | Norway | under construction | 30,000 plan | - | 230 plan | 230 | - | - | - |
| Meta GenAI cluster (RoCE) | Meta Platforms | USA | operational | 24,576 | - | - | - | - | - | - |
| Meta GenAI cluster (InfiniBand) | Meta Platforms | USA | operational | 24,576 | - | - | - | - | - | - |
| Microsoft Azure AI supercomputer (Loughton, UK) | Microsoft | United Kingdom | under construction | 23,040 plan | - | - | 90 | - | - | - |
| Yotta D2 Blackwell Ultra Cluster | Yotta Data Services | India | under construction | 20,736 plan | - | 60 plan | 60 | - | - | 2 plan |
| Project Ceiba | Amazon | USA | under construction | 20,736 plan | - | - | - | - | - | - |
| XTX Research Cluster | XTX Markets | Iceland | operational | 20,000 | - | - | - | - | - | - |
| China Mobile Hohhot Computing Center | China Mobile | China | operational | 20,000 | - | - | - | - | - | - |
| GroqCloud Dammam Inference Cluster | Groq | Saudi Arabia | operational | 19,000 | - | - | - | - | - | - |
| HUMAIN AI Factory Phase 1 | HUMAIN | Saudi Arabia | under construction | 18,000 plan | - | - | 500 | - | - | - |
| Mistral Campus AI (Bruyeres-le-Chatel) | Mistral AI | France | operational | 18,000 | - | 44 plan | 44 | - | - | - |
| China Mobile Harbin Computing Center | China Mobile | China | operational | 18,000 | - | - | - | - | - | - |
| Korea National AI Computing Center | Korea AI Computing Center Co. | South Korea | under construction | 15,000 plan | - | - | - | - | - | - |
| Coatlicue | National Polytechnic Institute (IPN) | Mexico | planned | 14,480* | - | - | 20 | - | - | - |
| Eagle | Microsoft | USA | operational | 14,400 | - | - | - | - | - | - |
| SINES Data Campus GB300 Deployment | Microsoft | Portugal | operational | 12,600 | - | - | 1,200 | 1.1 plan | 0 plan | - |
| xAI Atlanta | xAI | USA | operational | 12,448 | 17 est. | - | - | - | - | 0.7 |
| ByteDance MegaScale Cluster | ByteDance | China | operational | 12,288 | - | - | - | - | - | - |
| Cassava AI Factory | Cassava Technologies | South Africa | operational | 12,000 | - | - | - | - | - | - |
| Qingyang Enflame S60 Inference Cluster | Enflame | China | operational | 10,016 | - | - | - | - | - | - |
| Lambda Kansas City AI Factory | Lambda | USA | under construction | 10,000 plan | - | 100 plan | 100 | - | - | 0.5 plan |
| Equinox | Argonne National Laboratory | USA | planned | 10,000 plan | - | - | - | - | - | - |
| Microsoft Azure AI supercomputer for OpenAI (2020) | Microsoft | USA | operational | 10,000 | - | - | - | - | - | - |
| Fire-Flyer 2 | High-Flyer | China | operational | 10,000 | - | - | - | - | - | - |
| Industrial AI Cloud | Deutsche Telekom | Germany | operational | 10,000 | - | - | - | - | - | - |
| China Telecom Shaoguan Zhenwu Cluster | China Telecom | China | operational | 10,000 | - | - | - | - | - | - |
| Shenzhen 10,000-Card Ascend Cluster | Shenzhen Municipal Government | China | operational | 10,000 | - | - | - | - | - | - |
| Google TPU7x (Ironwood) pod | Alphabet | USA | operational | 9,216 | - | - | - | - | - | - |
| MAXIMUS-384 | Core42 | USA | operational | 9,000 | - | - | - | - | - | - |
| Google TPU v5p pod | Alphabet | USA | operational | 8,960 | - | - | - | - | - | - |
| TensorWave Tucson MI325X Cluster | TensorWave | USA | operational | 8,192* | - | - | - | - | - | - |
| Taiwan AI Factory | GMI Cloud | Taiwan | operational | 7,000* | - | - | 25 | - | - | - |
| ABCI 3.0 | AIST | Japan | operational | 6,128 | - | 6 | - | 1.15 | - | - |
| Meta Research SuperCluster (RSC) | Meta Platforms | USA | operational | 6,080 | - | - | - | - | - | - |
| Isambard-AI Phase 2 | University of Bristol | United Kingdom | operational | 5,280 | - | 5 | - | 1.08 | - | - |
| IBM Blue Vela | IBM | USA | operational | 5,000* | - | - | - | - | - | - |
| Eos | NVIDIA | USA | operational | 4,608 | - | - | - | - | - | - |
| Nscale Verne Iceland Cluster | Nscale | Iceland | operational | 4,600 | 15 | - | 15 | - | - | - |
| ABCI | AIST | Japan | decommissioned | 4,352 | - | - | - | - | - | - |
| Google TPU v4 supercomputer | Alphabet | USA | operational | 4,096 | - | - | - | - | - | - |
| Jane Street Texas GPU Cluster | Jane Street | USA | operational | 4,032 | - | - | - | - | - | - |
| Nebius Modiin Cluster | Nebius | Israel | operational | 4,000 | - | - | 8 | - | - | - |
| CHIE-4 | SoftBank Corp. | Japan | operational | 4,000 | - | - | - | - | - | - |
| Naver B200 4K Cluster | Naver | South Korea | operational | 4,000 | - | - | - | - | - | - |
| Nebius Longcross UK Cluster | Nebius | United Kingdom | operational | 4,000 | - | - | - | - | - | - |
| CoreWeave AI cluster for Inflection AI | CoreWeave | USA | operational | 3,584* | - | - | - | - | - | - |
| Databricks DBRX Training Cluster | Databricks | USA | operational | 3,072 | - | - | - | - | - | - |
| Tesla Dojo | Tesla | USA | decommissioned | 3,000* | - | - | - | - | - | - |
| IBM Vela | IBM | USA | operational | 2,880* | - | - | - | - | - | - |
| Kalos (Shanghai AI Lab) | Shanghai AI Laboratory | China | operational | 2,416* | - | - | - | - | - | - |
| Seren (Shanghai AI Lab) | Shanghai AI Laboratory | China | operational | 2,288* | - | - | - | - | - | - |
| Selene | NVIDIA | USA | operational | 2,240 | - | - | - | - | - | - |
| Roche AI Factory | Roche | USA | operational | 2,176 | - | - | - | - | - | - |
| Israel-1 | NVIDIA | Israel | operational | 2,048 plan | - | - | - | - | - | - |
| NIPA (Kakao) | Kakao | South Korea | operational | 2,040 | - | - | - | - | - | - |
| SDS-AI | SDS Shonfeld | Israel | operational | 2,032* | - | - | - | - | - | - |
| RIKYU | RIKEN | Japan | operational | 1,600 | - | - | - | - | - | - |
| Chervonenkis | Yandex | Russia | operational | 1,592* | - | - | - | - | - | - |
| NVIDIA DGX SuperPOD (V100) | NVIDIA | USA | operational | 1,536 | - | - | - | - | - | - |
| Gefion | Danish Centre for AI Innovation | Denmark | operational | 1,528 | - | - | - | - | - | - |
| Lyapunov | Yandex | Russia | operational | 1,096* | - | - | - | - | - | - |
| Galushkin | Yandex | Russia | operational | 1,088* | - | - | - | - | - | - |
| SK Telecom Titan | SK Telecom | South Korea | operational | 1,040 | - | - | - | - | - | - |
| HUMAIN Qualcomm AI 100 Cluster | HUMAIN | Saudi Arabia | under construction | 1,024* | 200 plan | - | - | - | - | - |
| Dawn | University of Cambridge | United Kingdom | operational | 1,024 | - | - | - | - | - | - |
| CZI GPU Cluster | Chan Zuckerberg Initiative | USA | operational | 1,024* | - | - | - | - | - | - |
| MN-2 | Preferred Networks | Japan | operational | 1,024* | - | - | - | - | - | - |
| E2E Chennai B200 Cluster | E2E Networks | India | operational | 1,024 | - | - | - | - | - | - |
| Ubilink H100 Cluster | Ubilink.AI | Taiwan | operational | 1,024* | - | - | - | - | - | - |
| Pre-Eos | NVIDIA | USA | operational | 1,024 est. | - | - | - | - | - | - |
| LillyPod | Eli Lilly | USA | operational | 1,016 | - | - | - | - | - | - |
| Nabuchodonosor | Scaleway | France | operational | 1,016 | - | - | - | 1.25 | 0.25 | - |
| Haein Cluster | SK Telecom | South Korea | operational | 1,000* | - | - | - | - | - | - |
| Moore Threads KUAE 1,000-GPU Cluster | Moore Threads | China | operational | 1,000 | - | - | - | - | - | - |
| DGX SaturnV | NVIDIA | USA | upgraded | 992* | - | - | - | - | - | - |
| NEC AI Research Supercomputer | NEC | Japan | operational | 928* | - | - | - | - | - | - |
| Berzelius | NSC | Sweden | operational | 880 | - | - | - | - | - | - |
| Alvis | C3SE | Sweden | decommissioned | 880* | - | - | - | - | - | - |
| HammerHAI | HLRS | Germany | under construction | 860 plan | - | - | - | - | - | - |
| Vulcan (Amii) | Amii | Canada | operational | 820 | - | - | - | - | - | - |
| SAKURAONE | SAKURA internet | Japan | operational | 800 | - | - | - | - | - | - |
| Helma | FAU Erlangen-Nuremberg | Germany | operational | 768* | - | - | - | - | - | - |
| Alex (NHR@FAU) | FAU Erlangen-Nuremberg | Germany | operational | 752* | - | - | - | - | - | - |
| Killarney | Vector Institute | Canada | operational | 752 | - | - | - | - | - | - |
| Cambridge-1 | NVIDIA | United Kingdom | operational | 640* | - | - | - | - | - | - |
| fortytwo | hessian.AI | Germany | operational | 636* | - | - | - | - | - | - |
| RAIDEN | RIKEN | Japan | operational | 624* | - | - | - | - | - | - |
| DeltaAI | NCSA | USA | operational | 608 | - | - | - | - | - | - |
| TX-GAIN | MIT Lincoln Laboratory | USA | operational | 600 | - | - | - | - | - | - |
| Kempner AI Cluster | Kempner Institute | USA | operational | 528 | - | - | - | - | - | - |
| alpha ONE | Aleph Alpha | Germany | operational | 512* | - | - | - | - | - | - |
| BioHive-2 | Recursion | USA | operational | 504* | - | - | - | - | - | - |
| HiPerGator AI | University of Florida | USA | operational | 504* | - | - | - | - | - | - |
| Skipjack (JHU) | Johns Hopkins University | USA | operational | 492 | - | - | - | - | - | - |
| Mimer AI Factory | NAISS | Sweden | under construction | 400 plan | - | - | - | - | - | - |
| AI-Farabium | Kazakhtelecom | Kazakhstan | operational | 400* | - | - | - | - | - | - |
| AIST AI Cloud | AIST | Japan | operational | 400* | - | - | - | - | - | - |
| AICR | MGHPCC | USA | operational | 400 | - | - | - | - | - | - |
| Voyager | SDSC | USA | operational | 352* | - | - | - | - | - | - |
| PARAM Siddhi-AI | C-DAC | India | operational | 336* | - | - | - | - | - | - |
| ASPIRE 2A+ | National Supercomputing Centre Singapore | Singapore | operational | 320* | - | - | - | - | - | - |
| Cerebras Oklahoma City | Cerebras | USA | operational | 300* | - | - | - | - | - | - |
| fortythree | hessian.AI | Germany | operational | 280* | - | - | - | - | - | - |
| Redtail | University of Utah | USA | operational | 264* | - | - | - | - | - | - |
| DAIS | MPCDF | Germany | operational | 264* | - | - | - | - | - | - |
| GMO GPU Cloud | GMO Internet Group | Japan | operational | 256* | - | - | - | - | - | - |
| NeXXt AI Factory | Fastweb | Italy | operational | 248* | - | - | - | - | - | - |
| Freyja | Amgen | Iceland | operational | 248* | - | - | - | - | - | - |
| Marlowe (Stanford) | Stanford University | USA | operational | 248* | - | - | - | - | - | - |
| Kyber | InstaDeep | France | operational | 224* | - | - | - | - | - | - |
| Isambard-AI | University of Bristol | United Kingdom | operational | 168 | - | - | - | - | - | - |
| Gautschi AI | Purdue University | USA | operational | 160* | - | - | - | - | - | - |
| Grete | GWDG | Germany | operational | 144* | - | - | - | - | - | - |
| Bede | N8 CIR | United Kingdom | operational | 137* | - | - | - | - | - | - |
| MN-3 | Preferred Networks | Japan | operational | 128* | - | - | - | - | - | - |
| Mayo Clinic DGX SuperPOD | Mayo Clinic | USA | operational | 128* | - | - | - | - | - | - |
| Tokyo-1 | Xeureka | Japan | operational | 128* | - | - | - | - | - | - |
| ai& Tenstorrent Galaxy Supercluster | ai& | Japan | operational | 120 | - | - | - | - | - | - |
| Zhores | Skoltech | Russia | operational | 104* | - | - | - | - | - | - |
| SEIRAN | Tokyo University of Technology | Japan | operational | 96* | - | - | - | - | - | - |
| Cannon2-H200 | Harvard FAS Research Computing | USA | operational | 96* | - | - | - | - | - | - |
| Jairu | USP CIAAM | Brazil | operational | 96* | - | - | - | - | - | - |
| Serbia National AI Platform | Office for IT and eGovernment of Serbia | Serbia | operational | 80* | - | - | - | - | - | - |
| Condor Galaxy 1 | Cerebras | USA | operational | 64 | - | - | - | - | - | - |
| Condor Galaxy 3 | Cerebras | USA | operational | 64 | - | - | - | - | - | - |
| PakSupercomputer | Namal Institute | Pakistan | operational | 40* | - | - | - | - | - | - |
| IARA | SiDi | Brazil | operational | 25* | - | - | - | - | - | - |
| Cerebras Andromeda | Cerebras | USA | operational | 16* | - | - | - | - | - | - |
| Neocortex | PSC | USA | operational | 2* | - | - | - | - | - | - |
| EPCC Cerebras CS-2 Cluster | EPCC | United Kingdom | operational | 2* | - | - | - | - | - | - |
| LRZ Cerebras CS-2 | LRZ | Germany | operational | 1* | - | - | - | - | - | - |
| HOLL-I | NCSA | USA | operational | 1* | - | - | - | - | - | - |
| Wonder Valley AI Campus | O'Leary Ventures | Canada | planned | - | - | 7,500 plan | 7,500 | - | - | 70 plan |
| Project Jade (Cheyenne) | Tallgrass Energy | USA | under construction | - | - | 2,700 | 2,700 | - | - | - |
| Meta Hyperion | Meta Platforms | USA | under construction | - | 5,000 plan | 2,260 | 5,000 | - | - | 50 |
| Poolside Project Horizon | Poolside | USA | planned | - | - | 2,000 plan | 2,000 | - | - | - |
| HUMAIN Project Halo | HUMAIN | Saudi Arabia | planned | - | - | 2,000 plan | 2,000 | - | - | - |
| Stargate Shackelford County (Frontier) | Oracle | USA | under construction | - | - | 1,400 plan | 1,400 | - | - | 25 |
| Campus AI Fouju | Campus AI | France | under construction | - | - | - | 1,400 | - | - | - |
| Stargate Milam County | SB Energy | USA | under construction | - | - | 1,200 plan | 1,200 | - | - | - |
| Nebius Independence AI Factory | Nebius | USA | under construction | - | - | 1,200 plan | 1,200 | - | - | - |
| Stargate UAE | G42 | United Arab Emirates | under construction | - | - | 1,000 plan | 1,000 | - | - | - |
| Stargate Michigan (Saline Township) | Oracle | USA | under construction | - | - | 1,000 plan | 1,000 | - | - | 16 |
| Stargate Port Washington (Lighthouse) | Oracle | USA | under construction | - | - | - | 1,000 | - | - | 15 |
| Meta Lebanon (Indiana) | Meta Platforms | USA | under construction | - | - | 1,000 plan | 1,000 | - | - | 10 |
| Meta El Paso (Texas) | Meta Platforms | USA | under construction | - | - | 1,000 plan | 1,000 | - | - | 14 |
| Crusoe Abilene Campus (Microsoft) | Crusoe | USA | under construction | - | - | 900 plan | 900 | - | - | - |
| Stargate Project Jupiter (New Mexico) | Oracle | USA | under construction | - | - | - | 900 | - | - | 50 plan |
| IREN Childress Horizons 1-4 (Microsoft) | IREN | USA | operational | - | 200 plan | 750 plan | 750 | - | - | - |
| Killellan AI Growth Zone | Argyll Data Development | United Kingdom | planned | - | - | 600 plan | 2,000 | - | - | - |
| NEXTDC S7 Sydney AI Campus | NEXTDC | Australia | planned | - | 550 plan | - | 550 | - | - | - |
| Racks Central RCJM AI Campus | Racks Central | Malaysia | under construction | - | - | 510 plan | 510 | - | - | 6.7 plan |
| Delta Forge 1 (Rapides Parish) | Applied Digital | USA | under construction | - | 300 | 430 | 430 | - | - | - |
| Justified Data (Anthropic) | Anthropic | USA | planned | - | 401 | - | 401 | - | - | - |
| DataVolt Oxagon AI Campus | DataVolt | Saudi Arabia | under construction | - | - | 360 plan | 1,500 | - | - | - |
| River Bend (Fluidstack) | Fluidstack | USA | under construction | - | 245 plan | 330 plan | 330 | - | - | - |
| Nebius Lappeenranta AI Factory | Nebius | Finland | under construction | - | - | 310 plan | 310 | - | - | - |
| CoreWeave Lancaster AI Data Center | CoreWeave | USA | planned | - | - | - | 300 | - | - | 6 plan |
| Nebius Vineland (Microsoft) | Nebius | USA | operational | - | - | 300 plan | 300 | - | - | - |
| SoftBank Tomakomai AI Data Center | SoftBank Corp. | Japan | under construction | - | - | 300 plan | 300 | - | - | - |
| Bell AI Fabric Saskatchewan | Bell Canada | Canada | under construction | - | - | 300 | 300 | - | - | - |
| Omnia Pecem (TikTok) | Omnia Data Centers | Brazil | under construction | - | 200 | 300 | 200 | - | - | 9.8 |
| Cipher Black Pearl (AWS) | Cipher Mining | USA | under construction | - | 216 | 300 | 300 | - | - | - |
| Nebius Birmingham AI Factory | Nebius | USA | under construction | - | - | 300 | 300 | - | - | - |
| CoreWeave Kenilworth (NEST) | CoreWeave | USA | under construction | - | - | 250 | 250 | - | - | 1.8 |
| Barber Lake (Fluidstack) | Fluidstack | USA | under construction | - | 168 plan | 244 plan | 300 | - | - | - |
| Polaris Forge 2 (Harwood) | Applied Digital | USA | under construction | - | 200 plan | - | 200 | 1.18 plan | - | - |
| Abernathy (Fluidstack) | Fluidstack | USA | under construction | - | 168 plan | - | 168 | - | - | - |
| SoftBank Sakai AI Data Center | SoftBank Corp. | Japan | planned | - | - | 150 plan | 400 | - | - | - |
| USIO | Synectics Technologies | Uganda | planned | - | - | - | 100 | - | - | - |
| Cerebras Guyana AI Data Center | Cerebras | Guyana | planned | - | - | 100 plan | 100 | - | - | - |
| Khazna Ajman AI Data Center | Khazna Data Centers | United Arab Emirates | under construction | - | 100 plan | - | 100 | - | - | - |
| Opera Iceland KEF-1 SuperPOD | Opera | Iceland | operational | - | - | 83 | - | 1.2 plan | - | - |
| CoreWeave Cambridge (Cohere) | CoreWeave | Canada | operational | - | - | 64 | 64 | - | - | - |
| Scala AI City | Scala Data Centers | Brazil | planned | - | 54 plan | - | 4,750 | 1.2 plan | 0 plan | 0.5 |
| du NextGenAI Supercluster | du Telecom | United Arab Emirates | operational | - | - | - | 13 | - | - | - |
| Nebius Iceland Cluster | Nebius | Iceland | operational | - | 10 | - | 10 | - | - | - |
| GroqCloud Kamloops Inference Cluster | Groq | Canada | under construction | - | - | 7 plan | 7 | - | - | - |
| Cerebras Montreal Inference Site (MTL-3) | Cerebras | Canada | operational | - | 5 | - | 5 | - | - | - |
| GroqCloud Sydney Inference Cluster | Groq | Australia | planned | - | - | 4.5 plan | 4.5 | - | - | - |
| Mila Sovereign AI Research Hub | Hypertec Cloud | Canada | planned | - | 3 plan | - | 3 | - | - | - |
| NVIDIA DGX SuperPOD (A100) | NVIDIA | USA | operational | - | 0.1 | - | - | - | - | - |
A figure marked "est." is a third-party estimate and one marked "plan" is a design or announced target; unmarked figures are stated by the operator or taken from a filing, and where several exist the operator's is shown. * Accelerator count taken from the component graph's quantity, not a metric row. Claimed MW comes from the build pipeline table of operator-stated capacities.
Estimates marked "est." for several US campuses are from Epoch AI's AI data centers dataset, licensed CC BY 4.0.
What can be compared, and what it costs to
| Quantity | What it means | Caveat |
|---|---|---|
| Accelerators | GPUs or other AI accelerators in the cluster or campus. | Different accelerators are not interchangeable: a count says nothing about their speed or how they are connected. |
| IT power | Electrical load of the computing equipment itself, before cooling and conversion losses. | Often conflated. A campus capacity is a design or contract figure, not what is drawn today. |
| Facility power | Total power the site is built or contracted to draw, including cooling and losses. | Often conflated. A campus capacity is a design or contract figure, not what is drawn today. |
| PUE | Power usage effectiveness: facility energy divided by IT energy. 1.0 is the floor; lower is better. | Operator-defined, often a fleet average, sometimes a design figure. Boundaries differ between companies. |
| WUE | Water usage effectiveness: litres of water consumed per kWh of IT energy. | Operator-defined, often a fleet average, sometimes a design figure. Boundaries differ between companies. |
| Stated investment | Investment the operator or developer states for the campus or cluster. | A stated programme size, frequently covering land, power and several phases. |
| Rack power | Power per rack of accelerators, a measure of density and of what cooling is required. | Rarely disclosed; design values dominate. |
| Model FLOPs utilisation | Share of the hardware's peak FLOPS that a reported training run actually used. | Published for individual training runs in papers, not as a standing series per cluster. |
| Training goodput | Share of wall-clock time spent making training progress, after failures, restarts and checkpointing. | Published for individual training runs in papers, not as a standing series per cluster. |
37 system-scale benchmark records (MLPerf, InferenceX and others) are on the benchmarks page. Both tables download as CSV: ai_datacenter_metrics, ai_benchmarks.