AI lab and enterprise GPU clusters: xAI, Mistral, DeepSeek, DeepL and more
12 clusters. 8 are running or partly running. Installed GPUs on record add up to 251,384, with a further 453,800 planned across 2 clusters. The largest is Colossus 2 (440,000 GPUs planned).
Frontier labs and the companies that train their own models run the largest single-purpose clusters. Their figures come from company statements, filings and press reports, and each row says which.
Last verified
12 clusters
Full sortable table| Cluster | Owner / operator | GPU model | GPUs installed | GPUs planned | Location | Status |
|---|---|---|---|---|---|---|
| Colossus 1 | xAI (SpaceXAI) | NVIDIA H100, H200 and GB200 | 220,000 | — | Memphis, Tennessee, United States | operational |
| Eni HPC7 | Eni | AMD Instinct MI300A APU | 13,920 | — | Ferrera Erbognone, Italy | operational |
| Eni HPC6 | Eni | AMD Instinct MI250X | 13,888 | — | Ferrera Erbognone, Italy | operational |
| DeepSeek-V3 H800 training cluster | DeepSeek | NVIDIA H800 | 2,048 | — | — | not stated |
| Chan Zuckerberg Initiative biological AI cluster | Chan Zuckerberg Initiative | NVIDIA H100 | 1,024 | — | — | operational |
| Recursion BioHive-2 | Recursion | NVIDIA H100 | 504 | — | Salt Lake City, Utah, United States | operational |
| Colossus 2 | xAI (SpaceXAI) | NVIDIA GB200 and GB300 | — | 440,000 | Memphis, Tennessee, United States | operational |
| Mistral AI Bruyères-le-Châtel cluster | Mistral AI | NVIDIA GB300 | — | 13,800 | Bruyères-le-Châtel (Eclairion campus), Île-de-France, France | announced |
| DeepL Mercury | DeepL SE | NVIDIA H100 | — | — | Falun (EcoDataCenter 1), Dalarna, Sweden | operational |
| ExxonMobil Discovery 6 | ExxonMobil | NVIDIA GH200 Grace Hopper | — | — | — | operational |
| Magic G4 on Google Cloud | Google Cloud (for Magic) | NVIDIA H100 | — | — | — | announced |
| Magic G5 on Google Cloud | Google Cloud (for Magic) | NVIDIA Grace Blackwell | — | — | — | announced |