THE GRADIENT AI + robotics landscape
Updated 2026-08-26 Open data

Hardware

AI compute infrastructure from Epoch AI open data

482 GPU clusters
83 Datacenters
NVIDIA H100 SXM5 80GB Top chip type

GPU clusters

Showing 30 of 482
NameOrganizationCountryChip TypeChip CountYear
xAI Colossus Memphis Phase 3xAIUnited States of AmericaNVIDIA H100 SXM5 80GB230,0002,025
Anonymized Chinese System—China—200,0002,023
xAI Colossus Memphis Phase 2xAIUnited States of AmericaNVIDIA H100 SXM5 80GB200,0002,025
Anonymized Chinese System—China—100,0002,021
xAI Colossus Memphis Phase 1xAIUnited States of AmericaNVIDIA H100 SXM5 80GB100,0002,024
Meta 100kMeta AIUnited States of AmericaNVIDIA H100 SXM5 80GB100,0002,024
OpenAI/Microsoft Goodyear ArizonaMicrosoft,OpenAIUnited States of AmericaNVIDIA H100 SXM5 80GB100,0002,024
Oracle OCI Supercluster H200sOracleUnited States of AmericaNVIDIA H200 SXM65,5362,024
Argonne NL AuroraUS Department of EnergyUnited States of AmericaIntel Data Center GPU Max 155063,7442,024
Anonymized Chinese System—China—50,0002,013
Tesla Cortex Phase 1TeslaUnited States of AmericaNVIDIA H100 SXM5 80GB50,0002,024
Lawrence Livermore NL El Capitan Phase 2US Department of EnergyUnited States of AmericaAMD Instinct MI300A44,5442,024
CoreWeave H200sCoreWeaveUnited States of AmericaNVIDIA H200 SXM42,0002,024
Anonymized Chinese System—China—40,0002,024
Anonymized Chinese System—China—40,0002,020
Anonymized Chinese System—China—40,0002,017
Oak Ridge NL FrontierUS Department of EnergyUnited States of AmericaAMD Radeon Instinct MI250X37,6322,022
Oracle OCI Supercluster A100sOracleUnited States of AmericaNVIDIA A10032,7682,023
Google Oklahoma TPU v4 PodsGoogleUnited States of AmericaGoogle TPU v432,7682,022
Anonymized Chinese System—China—30,0002,022
Gemini 1.0 Ultra training cluster AGoogleUnited States of AmericaGoogle TPU v428,6722,023
Microsoft GPT-4 clusterMicrosoft,OpenAIUnited States of AmericaNVIDIA A10025,0002,022
Meta GenAI 2024bMeta AIUnited States of AmericaNVIDIA H100 SXM5 80GB24,5762,024
Meta GenAI 2024aMeta AIUnited States of AmericaNVIDIA H100 SXM5 80GB24,5762,024
"Jupiter, Jülich"EuroHPC JU,Julich Supercomputing CenterGermanyNVIDIA GH20023,5362,025
Meta 2017 V100 ClusterMeta AIUnited States of AmericaNVIDIA Tesla V100 SXM222,0002,017
Anonymized Chinese System—China—20,0002,024
Anonymized Chinese System—China—20,0002,024
XTX Markets ClusterXTX Markets—NVIDIA A10020,0002,023
Anonymized Chinese System—China—20,0002,020

Data centers

Showing 30 of 83
NameOrganizationCountryPower (MW)StatusYear
Colossus 2SpaceXAIUnited States—Operational2,027
Anthropic-Amazon New CarlisleAmazonUnited States—Operational2,028
Microsoft Fairwater AtlantaMicrosoftUnited States—Operational2,026
Meta PrometheusMetaUnited States—Operational2,027
Google New AlbanyGoogleUnited States—Operational2,026
OpenAI Stargate AbileneOracleUnited States—Operational2,026
Microsoft Fairwater WisconsinMicrosoftUnited States—Operational2,028
Google Pryor (North)GoogleUnited States—Operational2,026
Colossus 1SpaceXAIUnited States—Operational2,026
Google ColumbusGoogleUnited States—Operational2,027
Amazon Madison Mega SiteAmazonUnited States—Operational2,028
Google BristowGoogleUnited States—Operational2,026
CoreWeave Denton TXCoreWeaveUnited States—Operational2,027
Huawei HoringerHuaweiChina—Operational2,026
DayOne Nusajaya—Malaysia—Operational2,026
QTS Richmond 1—United States—Operational2,026
Google Council Bluffs (East)GoogleUnited States—Operational2,026
Google OmahaGoogleUnited States—Operational2,028
Google PapillionGoogleUnited States—Operational2,026
Amazon RidgelandAmazonUnited States—Operational2,028
VNET Bayin UlanqabVNETChina—Operational2,027
Microsoft GoodyearMicrosoftUnited States—Operational2,026
Microsoft Project OsmiumMicrosoftUnited States—Operational2,026
Google MesaGoogleUnited States—Operational2,026
QTS Richmond 2—United States—Operational2,026
Meta-QTS Hillsboro 2Meta #likelyUnited States—Operational2,026
Meta RosemountMetaUnited States—Operational2,026
Meta JeffersonvilleMetaUnited States—Operational2,026
Alibaba ZhangbeiAlibabaChina—Operational2,026
Google Storey CountyGoogleUnited States—Operational2,027

ML accelerators

Showing 30 of 176
NameManufacturerYearMemoryTDPFP16 FLOPS
Microsoft Maia 200Microsoft2,026———
Meta MTIA 300Meta2,026———
AMD Instinct MI355XAMD2,025———
NVIDIA GB300 (Blackwell Ultra)NVIDIA2,025———
NVIDIA GB200NVIDIA2,025———
AMD Instinct MI350XAMD2,025———
Google TPU v7 IronwoodGoogle2,025———
NVIDIA B300 (Blackwell Ultra)NVIDIA2,025———
Huawei Ascend 920Huawei2,025———
Amazon Trainium3Amazon AWS2,025———
NVIDIA RTX Pro 6000 BlackwellNVIDIA2,025———
NVIDIA B200NVIDIA2,024———
Intel Habana Gaudi3Intel2,024———
AMD Instinct MI325XAMD2,024———
NVIDIA H200 SXMNVIDIA2,024———
Google TPU v6e TrilliumGoogle2,024———
Huawei Ascend 910CHuawei2,024———
Maia 100 (M100)Microsoft2,024———
Amazon Trainium2Amazon AWS2,024———
Meta MTIA v2 (MTIA 200)Meta2,024———
Cambricon Siyuan 590Cambricon2,024———
NVIDIA H100 NVLNVIDIA2,024———
AMD Instinct MI300XAMD2,023———
NVIDIA GH200NVIDIA2,023———
NVIDIA GH100NVIDIA2,023———
NVIDIA H800 SXM5NVIDIA2,023———
AMD Instinct MI300AAMD2,023———
AMD Radeon Instinct MI308XAMD2,023———
Google TPU v5pGoogle2,023———
Google TPU v5eGoogle2,023———

Data: Epoch AI (CC-BY-4.0). As of 2026-08-26.