# GPU Cluster List

> A sourced list of 87 named GPU clusters, from xAI’s Colossus (over 220,000 NVIDIA GPUs, per Anthropic) and the national-lab exascale machines (Aurora 63,744, El Capitan 46,080, Frontier 39,424) to the announced million-GPU builds. Each row gives owner, GPU model, installed and planned GPU counts, location, power and status, and every figure links to its source.

Every big GPU cluster: who owns it, which chips, how many, how live. 87 clusters, each field linked to its public source, last verified 2026-11-11.
When citing, link the canonical HTML page.

## Data
- [Largest GPU clusters in the world](https://gpuclusterlist.com/): the full comparison table
- [Methodology](https://gpuclusterlist.com/methodology/): what each field means and how it is verified

## Clusters
- [ABCI 3.0](https://gpuclusterlist.com/abci-3-0/): Japan’s ABCI offers open AI computing infrastructure to research and industrial users.
- [Adastra GPU partition](https://gpuclusterlist.com/adastra-gpu-partition/): CINES provides Adastra for French research teams preparing scientific software for exascale systems.
- [Alice Recoque](https://gpuclusterlist.com/alice-recoque/): France’s first exascale supercomputer, an Eviden machine on AMD Instinct MI430X GPUs, being installed at CEA’s TGCC for EuroHPC and GENCI.
- [Alps Grace Hopper partition](https://gpuclusterlist.com/alps-grace-hopper-partition/): CSCS offers Alps to Swiss and international researchers, with separate virtual clusters for research, AI and weather services.
- [Argonne Equinox](https://gpuclusterlist.com/argonne-equinox/): NVIDIA and Oracle announced Equinox as another Argonne system for scientific AI.
- [Argonne Solstice](https://gpuclusterlist.com/argonne-solstice/): NVIDIA and Oracle announced Solstice for DOE scientific AI and public research at Argonne.
- [Aurora](https://gpuclusterlist.com/aurora/): Argonne offers Aurora to the research community for scientific computing and AI.
- [Berzelius Ampere](https://gpuclusterlist.com/berzelius-ampere/): NAISS operates Berzelius Ampere for eligible Swedish academic AI research. The separate H200 system is excluded from this count.
- [ByteDance MegaScale Ampere cluster](https://gpuclusterlist.com/bytedance-megascale-ampere-cluster/): ByteDance’s engineers describe this production cluster and its use in large language model training.
- [Chan Zuckerberg Initiative biological AI cluster](https://gpuclusterlist.com/chan-zuckerberg-initiative-biological-ai-cluster/): CZI allocates GPU time to nonprofit life-sciences researchers building large AI models.
- [Colossus 1](https://gpuclusterlist.com/colossus-1/): xAI’s first Memphis supercluster, run under SpaceXAI. Anthropic agreed in May 2026 to use all of its compute capacity.
- [Colossus 2](https://gpuclusterlist.com/colossus-2/): xAI’s second Memphis supercluster, on GB200 and GB300 GPUs. Its first cluster came online in October 2025, per SpaceX’s S-1.
- [CoreWeave Barcelona H200 supercomputer](https://gpuclusterlist.com/coreweave-barcelona-h200-supercomputer/): CoreWeave announced this MERLIN Edged-hosted cluster for European AI training and inference customers.
- [DeepL Mercury](https://gpuclusterlist.com/deepl-mercury/): DeepL operates Mercury, an H100 supercomputer documented in TOP500’s June 2026 benchmark results.
- [DeepSeek-V3 H800 training cluster](https://gpuclusterlist.com/deepseek-v3-h800-training-cluster/): DeepSeek used this cluster to train DeepSeek-V3. Its paper documents the training configuration rather than a current fleet inventory.
- [Deutsche Telekom Industrial AI Cloud (Munich)](https://gpuclusterlist.com/telekom-industrial-ai-cloud-munich/): Deutsche Telekom and NVIDIA’s sovereign AI factory in Munich: up to 10,000 Blackwell GPUs in DGX B200 and RTX PRO servers, sold to German industry.
- [Doudna (NERSC-10)](https://gpuclusterlist.com/nersc-doudna/): NERSC’s next flagship system at Berkeley Lab, built by Dell on NVIDIA Vera Rubin with Quantum-X800 InfiniBand, due in 2026.
- [El Capitan](https://gpuclusterlist.com/el-capitan/): Lawrence Livermore provides this classified CORAL-2 system for national security computing.
- [Eni HPC6](https://gpuclusterlist.com/eni-hpc6/): Eni uses HPC6 for industrial research, energy-transition applications and subsurface studies.
- [Eni HPC7](https://gpuclusterlist.com/eni-hpc7/): Eni commissioned HPC7 alongside HPC6 for industrial scientific computing and energy research.
- [ExxonMobil Discovery 6](https://gpuclusterlist.com/exxonmobil-discovery-6/): ExxonMobil’s HPE Cray system on NVIDIA GH200 superchips for 4D seismic imaging, at 164.2 PFLOP/s on HPL, one of the largest private GPU supercomputers.
- [Foxconn Taiwan AI Factory (Big Innovation Company)](https://gpuclusterlist.com/foxconn-taiwan-ai-factory/): Foxconn’s 10,000-GPU Blackwell Ultra AI factory, run by its NVIDIA Cloud Partner subsidiary Big Innovation Company for Taiwan’s NSTC, TSMC and Foxconn.
- [Frontier](https://gpuclusterlist.com/frontier/): The Oak Ridge Leadership Computing Facility provides Frontier for large-scale scientific simulation and AI research.
- [Gefion](https://gpuclusterlist.com/gefion/): Denmark’s sovereign AI supercomputer, an NVIDIA DGX SuperPOD run by the Danish Centre for AI Innovation for researchers and companies.
- [GMI Cloud Taiwan AI Factory](https://gpuclusterlist.com/gmi-cloud-taiwan-ai-factory/): GMI Cloud announced an AI factory for Taiwanese sovereign AI training, fine-tuning and inference.
- [Horizon (TACC)](https://gpuclusterlist.com/tacc-horizon/): The NSF Leadership-Class Computing Facility system at TACC: 1,000 GB200 NVL4 nodes with 4,000 Blackwell GPUs, built by Dell, for US open science.
- [HUMAIN GB300 AI factory (phase 1)](https://gpuclusterlist.com/humain-gb300-phase-1/): The first phase of Saudi Arabia’s HUMAIN AI factories: 18,000 NVIDIA GB300 Grace Blackwell systems on InfiniBand, part of a 500 MW five-year plan.
- [IBM Blue Vela](https://gpuclusterlist.com/ibm-blue-vela/): IBM uses Blue Vela to train Granite models and conduct AI research. Its published pod size is not treated as a full-cluster GPU count.
- [IREN Horizon 1](https://gpuclusterlist.com/iren-horizon-1/): IREN delivered Horizon 1 to Microsoft as the first of four liquid-cooled GPU deployments at Childress.
- [Isambard-AI](https://gpuclusterlist.com/isambard-ai/): Bristol offers Isambard-AI through the UK AI Research Resource to academic and industrial AI users.
- [Jean Zay H100 extension](https://gpuclusterlist.com/jean-zay-h100-extension/): GENCI and IDRIS provide this Jean Zay extension for academic and industrial open AI research.
- [JUPITER Booster](https://gpuclusterlist.com/jupiter-booster/): EuroHPC and Jülich provide JUPITER for scientific simulation, data analysis and large-scale AI.
- [JUWELS Booster](https://gpuclusterlist.com/juwels-booster/): Jülich provides JUWELS Booster for large-scale GPU scientific computing and AI research.
- [KISTI National Supercomputer No. 6](https://gpuclusterlist.com/kisti-6/): South Korea’s sixth national supercomputer at KISTI, built by HPE on 8,496 NVIDIA GH200 GPUs, with full service from December 2026.
- [Lambda Kansas City AI Factory](https://gpuclusterlist.com/lambda-kansas-city-ai-factory/): Lambda announced a dedicated supercomputer for one customer’s large-scale AI training and inference.
- [Leonardo Booster](https://gpuclusterlist.com/leonardo-booster/): CINECA offers Leonardo’s accelerated Booster partition for scientific research and AI workloads.
- [LUMI-G](https://gpuclusterlist.com/lumi-g/): LUMI provides GPU computing to European researchers and industry through its accelerated partition.
- [Lux (ORNL)](https://gpuclusterlist.com/ornl-lux/): DOE’s AMD AI cluster at Oak Ridge: more than 4,000 Instinct MI355X GPUs in 500+ nodes on Ultra Ethernet, open to science and industry users.
- [Magic G4 on Google Cloud](https://gpuclusterlist.com/magic-g4-on-google-cloud/): Magic announced G4 to train code models using Google Cloud GPU infrastructure.
- [Magic G5 on Google Cloud](https://gpuclusterlist.com/magic-g5-on-google-cloud/): Magic announced a next-generation Blackwell supercomputer for frontier code-model development.
- [MareNostrum 5 ACC](https://gpuclusterlist.com/marenostrum-5-acc/): BSC provides MareNostrum’s accelerated HPC partition to scientific computing users.
- [MareNostrum 5 AI inference partition](https://gpuclusterlist.com/marenostrum-5-ai-inference-partition/): BSC lists this separate Blackwell Ultra partition for AI inference.
- [MareNostrum 5 AI training partition](https://gpuclusterlist.com/marenostrum-5-ai-training-partition/): BSC lists this Blackwell partition for AI training, separately from its original HPC accelerator partition.
- [MeluXina accelerator module](https://gpuclusterlist.com/meluxina-accelerator-module/): LuxProvide offers MeluXina GPU computing for research and industrial workloads.
- [Meta AI Research SuperCluster (RSC)](https://gpuclusterlist.com/meta-ai-research-supercluster-rsc/): Meta uses RSC for AI research, including language, vision and speech models.
- [Meta GenAI InfiniBand cluster](https://gpuclusterlist.com/meta-genai-infiniband-cluster/): Meta operates this second cluster for internal generative AI training and research.
- [Meta GenAI RoCE cluster](https://gpuclusterlist.com/meta-genai-roce-cluster/): Meta uses this internal cluster to train Llama and other generative AI models.
- [Meta Hyperion](https://gpuclusterlist.com/meta-hyperion/): Meta is developing Hyperion as a major internal AI training cluster. The capacity cited here is a future design target.
- [Meta Prometheus](https://gpuclusterlist.com/meta-prometheus/): Meta uses Prometheus to train its Muse foundation models. The first phase is complete.
- [Microsoft Azure Eagle](https://gpuclusterlist.com/microsoft-azure-eagle/): Azure Eagle demonstrates large-scale GPU supercomputing on Microsoft cloud infrastructure.
- [Microsoft Fairwater Atlanta](https://gpuclusterlist.com/microsoft-fairwater-atlanta/): Microsoft operates Atlanta Fairwater for AI workloads and connects it to other Fairwater sites through a dedicated network.
- [Microsoft Fairwater Wisconsin](https://gpuclusterlist.com/microsoft-fairwater-wisconsin/): Microsoft designed Fairwater as a single large AI supercomputer for training and inference, supporting Microsoft and OpenAI workloads.
- [Mistral AI Bruyères-le-Châtel cluster](https://gpuclusterlist.com/mistral-ai-bruyeres-le-ch-tel-cluster/): Mistral financed this French GPU cluster for its own models and customer training and inference.
- [NAVER Sejong](https://gpuclusterlist.com/naver-sejong/): NAVER operates Sejong at GAK Sejong to support its AI technology development.
- [Nebius Kansas City GPU cluster](https://gpuclusterlist.com/nebius-kansas-city-gpu-cluster/): Nebius announced this US cluster for AI cloud customers, beginning with Hopper GPUs and a possible larger buildout.
- [Nebius Modiin Blackwell cluster](https://gpuclusterlist.com/nebius-modiin-blackwell-cluster/): Nebius provides this Israeli Blackwell deployment to startups, enterprises and researchers for training, fine-tuning and inference.
- [Nebius Vineland GPU infrastructure](https://gpuclusterlist.com/nebius-vineland-gpu-infrastructure/): Nebius is the GPU-infrastructure tenant at the DataOne Vineland AI factory, serving advanced AI workloads.
- [Nscale Loughton AI campus](https://gpuclusterlist.com/nscale-loughton/): Nscale’s planned UK supercomputer in Essex, launched with 23,040 GB300 GPUs for Microsoft Azure. Microsoft left the deal in 2026 over grid power dates.
- [Nscale Narvik (Stargate Norway)](https://gpuclusterlist.com/nscale-narvik-stargate-norway/): Nscale is developing the Narvik AI campus for Microsoft workloads. Published figures distinguish contracted Blackwell capacity from a later Rubin expansion.
- [Nscale Sines AI campus deployment](https://gpuclusterlist.com/nscale-sines-ai-campus-deployment/): Nscale supplies Microsoft with GPU infrastructure at the SINES Data Campus, with a further Rubin deployment announced for late 2027.
- [Nscale Texas Microsoft cluster](https://gpuclusterlist.com/nscale-texas-microsoft-cluster/): Nscale contracted a large Texas GPU deployment to provide AI infrastructure services to Microsoft.
- [NVIDIA Eos (10,752-GPU system)](https://gpuclusterlist.com/nvidia-eos-10752-gpu/): NVIDIA used this separate Eos DGX SuperPOD for MLPerf training benchmarks.
- [NVIDIA Eos (4,608-GPU system)](https://gpuclusterlist.com/nvidia-eos-4608-gpu/): NVIDIA uses this DGX SuperPOD for AI research and as a reference for enterprise AI infrastructure.
- [NVIDIA Israel-1](https://gpuclusterlist.com/nvidia-israel-1/): NVIDIA uses Israel-1 for research and as a Spectrum-X networking testbed. The cited count describes the planned completed system.
- [Perlmutter GPU partition](https://gpuclusterlist.com/perlmutter-gpu-partition/): NERSC provides Perlmutter for open scientific computing, including GPU-accelerated simulation and AI.
- [Polaris](https://gpuclusterlist.com/polaris/): Argonne provides Polaris for scientific simulations, AI and data analysis.
- [Project Ceiba](https://gpuclusterlist.com/project-ceiba/): AWS and NVIDIA co-built Ceiba to support NVIDIA AI research and development through DGX Cloud architecture.
- [Recursion BioHive-2](https://gpuclusterlist.com/recursion-biohive-2/): Recursion uses BioHive-2 for AI-based drug discovery and biological model development.
- [Sakura Ishikari H100 deployment](https://gpuclusterlist.com/sakura-ishikari-h100-deployment/): Sakura completed this Ishikari GPU deployment to supply its Koukaryoku generative AI cloud service.
- [Scaleway Nabuchodonosor 2023](https://gpuclusterlist.com/scaleway-nabuchodonosor-2023/): Scaleway offers Nabu GPU supercomputing to European businesses developing AI models.
- [Setonix AMD GPU partition](https://gpuclusterlist.com/setonix-amd-gpu-partition/): Pawsey provides Setonix GPU resources for Australian scientific research.
- [Shaheen III GPU partition](https://gpuclusterlist.com/shaheen-iii-gpu-partition/): KAUST provides Shaheen III’s GPU partition to accelerate scientific research.
- [SoftBank Blackwell DGX SuperPOD](https://gpuclusterlist.com/softbank-blackwell-dgx-superpod/): SoftBank deployed this Blackwell SuperPOD initially for SB Intuitions to develop Japanese language models.
- [Stargate Abilene (Crusoe / Oracle)](https://gpuclusterlist.com/stargate-abilene-crusoe-oracle/): Crusoe built Stargate’s Abilene campus, where Oracle runs AI training and inference infrastructure for OpenAI.
- [Stargate UK (Nscale for OpenAI)](https://gpuclusterlist.com/stargate-uk/): OpenAI, NVIDIA and Nscale’s UK sovereign-compute build: up to 8,000 NVIDIA GPUs for OpenAI from Q1 2026, with room to grow to 31,000.
- [TensorWave MI325X training cluster](https://gpuclusterlist.com/tensorwave-mi325x-training-cluster/): TensorWave offers this AMD-based cluster for large AI training workloads.
- [Tesla Cortex 1](https://gpuclusterlist.com/tesla-cortex-1/): Tesla uses Cortex 1 for autonomy and robotics training. Tesla’s H100-equivalent figure is a performance measure, not a physical GPU count.
- [Tesla Cortex 2](https://gpuclusterlist.com/tesla-cortex-2/): Tesla operates Cortex 2 for vehicle and humanoid-robot autonomy development.
- [Together AI–Hypertec Blackwell cluster](https://gpuclusterlist.com/together-ai-hypertec-blackwell-cluster/): Together AI and Hypertec announced a shared build to provide AI training, fine-tuning and inference infrastructure.
- [TSUBAME4.0](https://gpuclusterlist.com/tsubame4-0/): Institute of Science Tokyo offers TSUBAME4.0 for university research and shared scientific computing.
- [Tuolumne](https://gpuclusterlist.com/tuolumne/): Lawrence Livermore provides Tuolumne as the unclassified sibling of El Capitan for open science.
- [Venado](https://gpuclusterlist.com/venado/): Los Alamos uses Venado for national-security science, including AI reasoning models on a classified network.
- [Yandex Chervonenkis](https://gpuclusterlist.com/yandex-chervonenkis/): Yandex documents this A100 system for machine-learning workloads. These specifications describe its October 2021 configuration.
- [Yandex Galushkin](https://gpuclusterlist.com/yandex-galushkin/): Yandex documents this A100 system for machine-learning workloads. These specifications describe its October 2021 configuration.
- [Yandex Lyapunov](https://gpuclusterlist.com/yandex-lyapunov/): Yandex documents this A100 system for machine-learning workloads. These specifications describe its October 2021 configuration.
- [Yotta Greater Noida D2 Blackwell deployment](https://gpuclusterlist.com/yotta-greater-noida-d2-blackwell-deployment/): Yotta announced this Greater Noida deployment to expand Indian sovereign AI capacity.
- [Yotta Shakti Cloud NM1](https://gpuclusterlist.com/yotta-shakti-cloud-nm1/): Yotta provides sovereign AI compute from NM1 to Indian enterprise and government customers.

## Optional
- [All records in one file](https://gpuclusterlist.com/llms-full.txt)
- [Sitemap](https://gpuclusterlist.com/sitemap.xml)