COMPUTE ATLAS Supercomputer supply-chain graph
1345 systems 511 sites Sourced data

Supercomputers/Analysis/Accelerator share

Analysis 01

Accelerator share, weighted by FLOPS

Across the 1345 systems in this dataset, NVIDIA supplies 42% of installed FLOPS as of 2026, against 31% for AMD. Machine-count share (the figure normally quoted) produces a different ordering, because it counts one exascale system and one departmental cluster as one machine each. Weighting by capability is the correction, and it is the whole argument of this page.

Phase 1 seed dataset, compiled by hand. These rows were built from public operator, laboratory and vendor sources. A mechanical second-reader pass has since fetched every cited source: 1263 of 1345 systems have a readable citation that names them, and 56 are genuinely weakly sourced. Every claim carries its source and a confidence tier. Treat anything below verified as a lead, not a citation.

Read the shape, not the spikes. This is the installed base of the 1345 systems in this dataset: a curated seed chosen for supply-chain diversity, not a census of the market. At that sample size one machine entering or leaving service moves a share line by tens of points, so year-to-year jumps are composition effects rather than market events. The multi-year trends are the part worth quoting; the individual steps usually have one system's name on them, and the table view under the chart will tell you which.

  • NVIDIA
  • AMD
  • Intel
  • NEC
  • NUDT
  • PEZY Computing
  • Other (3 suppliers)
  • No accelerator (CPU-only)
Accelerator share of installed FLOPS, 1993–2026 Stacked area chart showing each accelerator supplier's share of installed Rmax by year, with a separate band for systems that have no accelerator at all. NVIDIAAMDIntelNECNUDTPEZY ComputingOther (3 suppliers)No accelerator (CPU-only) 0%25%50%75%100%19931998200320082013201820232026 NVIDIAAMDIntelCPU-only 1993 NVIDIA 0.0% AMD 0.0% Intel 0.0% NEC 0.0% NUDT 0.0% PEZY Computing 0.0% Other 0.0% CPU-only 100.0%1994 NVIDIA 0.0% AMD 0.0% Intel 0.0% NEC 0.0% NUDT 0.0% PEZY Computing 0.0% Other 0.0% CPU-only 100.0%1995 NVIDIA 0.0% AMD 0.0% Intel 0.0% NEC 0.0% NUDT 0.0% PEZY Computing 0.0% Other 0.0% CPU-only 100.0%1996 NVIDIA 0.0% AMD 0.0% Intel 0.0% NEC 0.0% NUDT 0.0% PEZY Computing 0.0% Other 0.0% CPU-only 100.0%1997 NVIDIA 0.0% AMD 0.0% Intel 0.0% NEC 0.0% NUDT 0.0% PEZY Computing 0.0% Other 0.0% CPU-only 100.0%1998 NVIDIA 0.0% AMD 0.0% Intel 0.0% NEC 0.0% NUDT 0.0% PEZY Computing 0.0% Other 0.0% CPU-only 100.0%1999 NVIDIA 0.0% AMD 0.0% Intel 0.0% NEC 0.0% NUDT 0.0% PEZY Computing 0.0% Other 0.0% CPU-only 100.0%2000 NVIDIA 0.0% AMD 0.0% Intel 0.0% NEC 0.0% NUDT 0.0% PEZY Computing 0.0% Other 0.0% CPU-only 100.0%2001 NVIDIA 0.0% AMD 0.0% Intel 0.0% NEC 0.0% NUDT 0.0% PEZY Computing 0.0% Other 0.0% CPU-only 100.0%2002 NVIDIA 0.0% AMD 0.0% Intel 0.0% NEC 0.0% NUDT 0.0% PEZY Computing 0.0% Other 0.0% CPU-only 100.0%2003 NVIDIA 0.0% AMD 0.0% Intel 0.0% NEC 0.0% NUDT 0.0% PEZY Computing 0.0% Other 0.0% CPU-only 100.0%2004 NVIDIA 0.0% AMD 0.0% Intel 0.0% NEC 0.0% NUDT 0.0% PEZY Computing 0.0% Other 0.0% CPU-only 100.0%2005 NVIDIA 0.0% AMD 0.0% Intel 0.0% NEC 0.0% NUDT 0.0% PEZY Computing 0.0% Other 0.0% CPU-only 100.0%2006 NVIDIA 0.0% AMD 0.0% Intel 0.0% NEC 0.0% NUDT 0.0% PEZY Computing 0.0% Other 0.0% CPU-only 100.0%2007 NVIDIA 0.0% AMD 0.0% Intel 0.0% NEC 0.0% NUDT 0.0% PEZY Computing 0.0% Other 0.0% CPU-only 100.0%2008 NVIDIA 0.0% AMD 0.0% Intel 0.0% NEC 0.0% NUDT 0.0% PEZY Computing 0.0% Other 7.8% CPU-only 92.2%2009 NVIDIA 2.7% AMD 3.0% Intel 0.0% NEC 0.0% NUDT 0.0% PEZY Computing 0.0% Other 8.3% CPU-only 86.1%2010 NVIDIA 14.7% AMD 2.9% Intel 0.0% NEC 0.0% NUDT 0.0% PEZY Computing 0.0% Other 5.4% CPU-only 77.0%2011 NVIDIA 11.8% AMD 1.7% Intel 0.6% NEC 0.0% NUDT 0.0% PEZY Computing 0.0% Other 3.2% CPU-only 82.7%2012 NVIDIA 19.3% AMD 1.1% Intel 1.8% NEC 0.0% NUDT 0.0% PEZY Computing 0.0% Other 1.2% CPU-only 76.7%2013 NVIDIA 13.1% AMD 0.6% Intel 20.1% NEC 0.0% NUDT 14.4% PEZY Computing 0.0% Other 0.7% CPU-only 51.1%2014 NVIDIA 15.1% AMD 0.7% Intel 19.9% NEC 0.0% NUDT 12.8% PEZY Computing 0.0% Other 0.2% CPU-only 51.2%2015 NVIDIA 15.2% AMD 0.5% Intel 17.8% NEC 0.0% NUDT 9.9% PEZY Computing 0.2% Other 0.2% CPU-only 56.2%2016 NVIDIA 10.9% AMD 0.3% Intel 16.9% NEC 0.0% NUDT 6.2% PEZY Computing 0.3% Other 0.1% CPU-only 65.2%2017 NVIDIA 15.0% AMD 0.3% Intel 17.7% NEC 0.0% NUDT 4.8% PEZY Computing 3.3% Other 0.1% CPU-only 58.9%2018 NVIDIA 40.3% AMD 0.2% Intel 10.7% NEC 0.0% NUDT 2.9% PEZY Computing 2.0% Other 0.1% CPU-only 43.8%2019 NVIDIA 43.0% AMD 0.1% Intel 9.5% NEC 0.0% NUDT 2.6% PEZY Computing 0.3% Other 0.0% CPU-only 44.4%2020 NVIDIA 38.5% AMD 0.2% Intel 5.7% NEC 0.5% NUDT 1.6% PEZY Computing 0.2% Other 0.1% CPU-only 53.2%2021 NVIDIA 45.7% AMD 0.9% Intel 4.5% NEC 0.6% NUDT 1.2% PEZY Computing 0.1% Other 0.1% CPU-only 46.8%2022 NVIDIA 31.5% AMD 39.4% Intel 2.4% NEC 0.3% NUDT 0.7% PEZY Computing 0.1% Other 0.1% CPU-only 25.6%2023 NVIDIA 36.4% AMD 27.0% Intel 15.8% NEC 0.5% NUDT 0.4% PEZY Computing 0.1% Other 0.1% CPU-only 19.8%2024 NVIDIA 37.7% AMD 38.8% Intel 10.1% NEC 0.3% NUDT 0.3% PEZY Computing 0.0% Other 0.1% CPU-only 12.8%2025 NVIDIA 47.2% AMD 33.8% Intel 7.9% NEC 0.3% NUDT 0.2% PEZY Computing 0.0% Other 0.0% CPU-only 10.4%2026 NVIDIA 41.6% AMD 30.7% Intel 6.3% NEC 0.2% NUDT 0.2% PEZY Computing 0.0% Other 0.0% CPU-only 20.9%
Share of installed Rmax by accelerator supplier, rolled up to the corporate parent that owns each supplier today. Systems with no accelerator are a band, not an omission.
Download this chart's data (CSV) ↓
Table view: Accelerator share of installed FLOPS
YearNVIDIAAMDIntelNECNUDTPEZY ComputingOther (3 suppliers)No accelerator (CPU-only)Installed total
19930.0%0.0%0.0%0.0%0.0%0.0%0.0%100.0%398.3 GFlop/s
19940.0%0.0%0.0%0.0%0.0%0.0%0.0%100.0%472.1 GFlop/s
19950.0%0.0%0.0%0.0%0.0%0.0%0.0%100.0%786.5 GFlop/s
19960.0%0.0%0.0%0.0%0.0%0.0%0.0%100.0%1.92 TFlop/s
19970.0%0.0%0.0%0.0%0.0%0.0%0.0%100.0%3.79 TFlop/s
19980.0%0.0%0.0%0.0%0.0%0.0%0.0%100.0%8.46 TFlop/s
19990.0%0.0%0.0%0.0%0.0%0.0%0.0%100.0%10.76 TFlop/s
20000.0%0.0%0.0%0.0%0.0%0.0%0.0%100.0%20.15 TFlop/s
20010.0%0.0%0.0%0.0%0.0%0.0%0.0%100.0%34.36 TFlop/s
20020.0%0.0%0.0%0.0%0.0%0.0%0.0%100.0%104.08 TFlop/s
20030.0%0.0%0.0%0.0%0.0%0.0%0.0%100.0%162.05 TFlop/s
20040.0%0.0%0.0%0.0%0.0%0.0%0.0%100.0%769.52 TFlop/s
20050.0%0.0%0.0%0.0%0.0%0.0%0.0%100.0%1.14 PFlop/s
20060.0%0.0%0.0%0.0%0.0%0.0%0.0%100.0%1.35 PFlop/s
20070.0%0.0%0.0%0.0%0.0%0.0%0.0%100.0%2.99 PFlop/s
20080.0%0.0%0.0%0.0%0.0%0.0%7.8%92.2%13.22 PFlop/s
20092.7%3.0%0.0%0.0%0.0%0.0%8.3%86.1%18.89 PFlop/s
201014.7%2.9%0.0%0.0%0.0%0.0%5.4%77.0%29.03 PFlop/s
201111.8%1.7%0.6%0.0%0.0%0.0%3.2%82.7%48.84 PFlop/s
201219.3%1.1%1.8%0.0%0.0%0.0%1.2%76.7%127.4 PFlop/s
201313.1%0.6%20.1%0.0%14.4%0.0%0.7%51.1%212.7 PFlop/s
201415.1%0.7%19.9%0.0%12.8%0.0%0.2%51.2%239.1 PFlop/s
201515.2%0.5%17.8%0.0%9.9%0.2%0.2%56.2%309.0 PFlop/s
201610.9%0.3%16.9%0.0%6.2%0.3%0.1%65.2%492.4 PFlop/s
201715.0%0.3%17.7%0.0%4.8%3.3%0.1%58.9%646.2 PFlop/s
201840.3%0.2%10.7%0.0%2.9%2.0%0.1%43.8%1.06 EFlop/s
201943.0%0.1%9.5%0.0%2.6%0.3%0.0%44.4%1.19 EFlop/s
202038.5%0.2%5.7%0.5%1.6%0.2%0.1%53.2%1.98 EFlop/s
202145.7%0.9%4.5%0.6%1.2%0.1%0.1%46.8%2.50 EFlop/s
202231.5%39.4%2.4%0.3%0.7%0.1%0.1%25.6%4.69 EFlop/s
202336.4%27.0%15.8%0.5%0.4%0.1%0.1%19.8%7.05 EFlop/s
202437.7%38.8%10.1%0.3%0.3%0.0%0.1%12.8%11.12 EFlop/s
202547.2%33.8%7.9%0.3%0.2%0.0%0.0%10.4%14.10 EFlop/s
202641.6%30.7%6.3%0.2%0.2%0.0%0.0%20.9%17.54 EFlop/s
How this was calculated

For each calendar year, every system in service that year contributes its last publicly reported Rmax. A system is in service from its first-operational date until its decommissioning date, or to the present if it has not been retired.

Where a system has more than one accelerator supplier, which happens when a machine is rebuilt with different silicon, as Tianhe-2 was after the 2015 export controls, its FLOPS are split evenly between them rather than counted twice.

Systems with no accelerator are kept in the denominator as their own band. Fugaku, Hawk and ARCHER2 are real machines representing real installed capability, and dropping them is the most common way this chart is drawn wrong elsewhere: it silently redefines the question from “share of installed FLOPS” to “share of accelerated FLOPS” and inflates every supplier.

Suppliers are the company that shipped the part at build time, rolled up through vendor_aliases to the parent that owns that business today. Beyond eight suppliers the tail folds into a single “Other” band rather than being given a ninth colour.

469 systems carry no Rmax and are excluded from this chart entirely: Blue Waters, Meta GenAI cluster (RoCE), Meta GenAI cluster (InfiniBand), xAI Colossus, Stargate Abilene, Microsoft Fairwater (Atlanta), Microsoft Fairwater (Wisconsin), Project Rainier, Tesla Cortex, Google TPU v5p pod, Google TPU7x (Ironwood) pod, DeltaAI, Doudna, Horizon, Discovery (ORNL), Lux (ORNL), Solstice, Equinox, Condor Galaxy 1, Stargate UAE, Alice Recoque, Blue Lion, KISTI-6, Stampede3, Bridges-2, Meta Research SuperCluster (RSC), Microsoft Azure AI supercomputer for OpenAI (2020), Google TPU v4 supercomputer, Cerebras Andromeda, Condor Galaxy 3, Pengcheng Cloudbrain-II, Comet, Serafin, CoreWeave AI cluster for Inflection AI, Google TPU (Ironwood) capacity for Anthropic, GroqCloud Dammam Inference Cluster, Microsoft Azure AI supercomputer (Loughton, UK), Microsoft Azure AI datacenter (Narvik, Norway), Topaz, Bridges, Mistral Campus AI (Bruyeres-le-Chatel), Together AI / Hypertec Cloud GB200 Cluster, Databricks DBRX Training Cluster, Met Office Azure Supercomputer, Meta Prometheus, Meta Hyperion, xAI Colossus 2, Supek, Devana, Pangea 4, Bubba, Cray-1, Cray X-MP, Cray-2, Cray Y-MP, CM-2, ETA10, Sherlock, Euler, Della, Savio, Anton 2, Anton 3, Cyclone, Rudens, farm22, LillyPod, Roche AI Factory, BMS AI Factory, Double Helix, BA-HPC C2, Kabre, ClusterUY, MMS1, Australis, Crusoe Abilene Campus (Microsoft), Vultr AMD GPU Supercompute Cluster (Chicago), SINES Data Campus GB300 Deployment, Sesterce Marseille SuperComputer, AWE Vulcan, PARAM Pravega, Alem.Cloud, DOST Blue Gene/P, El Khawarizmi Supercomputer, Tesla Dojo, Fuyao, Mana Oasis, Xingrui Intelligent Computing Center, Firebird Armenian AI Factory (DC-1), AzInTelecom Supercomputer Center, USIO, XTX Research Cluster, Jane Street Texas GPU Cluster, Fire-Flyer 2, BP CHPC, BUZZ AI Cloud (Paraguay), Digital Government AI Cluster, JD Cloud 100,000-GPU Cluster, Haein Cluster, Industrial AI Cloud, NeXXt AI Factory, ILM Death Star, Cerebras Guyana AI Data Center, Mayo Clinic DGX SuperPOD, St. Jude HPC Research Cluster, Menlo, Airbus HPC6, Naver B200 4K Cluster, SenseTime AIDC (Lingang), Baidu Kunlun P800 Cluster, Reliance Intelligence Jamnagar AI Compute Platform, Lambda Kansas City AI Factory, TensorWave Tucson MI325X Cluster, IREN Childress Horizons 1-4 (Microsoft), Polaris Forge 1 (Ellendale), Stargate Shackelford County (Frontier), Stargate Michigan (Saline Township), COSMA8, wICE, BNY DGX SuperPOD, China Merchants Bank AI Compute Platform, Shell Houston HPC Cluster, Hyundai Motor Group Blackwell AI Factory, Samsung AI Megafactory, Hon Hai Kaohsiung Super Computing Center, Moordiup, Leftraru Epu, Rocket Cluster, Serbia National AI Platform, PakSupercomputer, University of Nottingham Malaysia DGX A100, HPC Saulėtekis, Kathmandu University Supercomputer, Kestrel, Gaea, Stargate Milam County, Stargate Project Jupiter (New Mexico), Stargate Port Washington (Lighthouse), Tokyo-1, Freyja, CZI GPU Cluster, Kyber, Roihu, HammerHAI, Mimer AI Factory, IT4LIA, PARAM Ganga, Xiuhcoatl, Plasma Simulator Sosei, Ursa, Mary Coombs, NASA Discover, ALMA Correlator, Project Ceiba, IBM Vela, IBM Blue Vela, Tesla Cortex 2, Azure NDv6 GB300 Cluster, Google TPU v4 ML Hub (Oklahoma), Zhangbei Intelligent Computing Center, HKUST SuperPOD, DelftBlue, Tinkercliffs, Alex (NHR@FAU), Grete, Nabuchodonosor, China Mobile Hohhot Computing Center, China Mobile Harbin Computing Center, China Telecom Shaoguan Zhenwu Cluster, Taiwan AI Factory, du NextGenAI Supercluster, NEC AI Research Supercomputer, Galaxy Helios (CoreWeave), River Bend (Fluidstack), Barber Lake (Fluidstack), Polaris Forge 2 (Harwood), Abernathy (Fluidstack), CoreWeave Lancaster AI Data Center, Denton (CoreWeave), Nebius Kansas City Cluster, Nebius Mäntsälä Cluster, Nebius Lappeenranta AI Factory, Nebius Independence AI Factory, Nebius Vineland (Microsoft), Killarney, Vulcan (Amii), Isambard 3, Bede, Sulis, Ngarrgu Tindebeek, Korea National AI Computing Center, FugakuNEXT, UK Next National Supercomputer, TELUS BC Sovereign AI Cluster, Campus AI Fouju, Paducah AI Campus, SoftBank Tomakomai AI Data Center, Yotta D2 Blackwell Ultra Cluster, Gaia AI Factory, LUMI-AI, HOKUSAI BigWaterfall 2, Earth Simulator 3, MN-2, Wuhan AI Computing Center, Wuhan Supercomputing Center, China Mobile Wuhan Computing Center, Chengdu Intelligent Computing Center, Xi'an Future AI Computing Center, Sugon 8000 Dengfeng, Sunway OceanLight, Qingdao AI Computing Center, Chongqing AI Innovation Center, ACES (Texas A&M), Neocortex, Ookami, Vista (TACC), Jetstream2, SDS-AI, Nebius Modiin Cluster, HUMAIN AI Factory Phase 1, Ibex, Jubail, ALTAY, IKARUS, Quriosity (2023), PARAM Shakti, PARAM Rudra (IUAC), E2E Chennai B200 Cluster, SAGA-220, Viettel DGX B200 System, Govorun, Zhores, Coatlicue, CMCC Juno, CMCC Cassandra, NASA Cabeus, SCINet Atlas, Cerebras Oklahoma City, Hala Point, Voyager, LRZ Cerebras CS-2, EPCC Cerebras CS-2 Cluster, HOLL-I, Sunspot, HUMAIN Qualcomm AI 100 Cluster, Herder, rzVernal, Jade (LLNL), Crusher, Spock, Andes (OLCF), Chicoma, Firmus Southgate Melbourne, Firmus Southgate Launceston, Firmus Batam AI Factory, Meta Lebanon (Indiana), Meta El Paso (Texas), Beacon Point (Hut 8), YTL AI Cloud (Kulai), Google India AI Hub (Vizag), Lake Mariner (Fluidstack), Justified Data (Anthropic), mdx I, Marlowe (Stanford), Seren (Shanghai AI Lab), Kalos (Shanghai AI Lab), LG Paju AI Data Center, Corona, Iridis 6, Triton, Stanage, Pelle, Saga, Imperial HX1, Imperial CX3, Kelvin2, Computerome 2.0, Explorer (Northeastern), Discovery (USC), Hoffman2, Alpine (CU Boulder), DARWIN, Oscar (Brown), Resnick HPC Cluster, Afton (UVA), Artemis (Sydney), Katana, Bunya, Spartan (Melbourne), Petrichor, Mahuika, PARAM Sanganak, OzSTAR (2018), Hortense, Barbora, Galileo100, FinisTerrae III, CoolMUC-4, bwUniCluster 3.0, JUSTUS 2, bwForCluster Helix, Magerit-3, NIC5, Google Pryor North TPU Campus, Google Columbus TPU Campus, Google New Albany TPU Campus, Google Council Bluffs East TPU Campus, AWS Madison Mega Site (Canton), AWS Ridgeland Campus, Microsoft Project Osmium, Microsoft Goodyear AI Datacenter, Meta Rosemount (Minnesota), Meta Montgomery (Alabama), Meta Cheyenne (Wyoming), xAI Atlanta, Muskogee (CoreWeave), Pangea 5, CoreWeave Kenilworth (NEST), Bell AI Fabric Saskatchewan, CoreWeave Cambridge (Cohere), Scala AI City, Omnia Pecem (TikTok), Project Jade (Cheyenne), IREN Sweetwater Hub, Delta Forge 1 (Rapides Parish), Cipher Black Pearl (AWS), Nebius Birmingham AI Factory, Wonder Valley AI Campus, Lobo Carneiro, TUPAC, Technion Zeus, Irgetas, ISSAI Cluster, EthERNet HPC Facility, Jairu, Ada Lovelace, cHARISMa, Guane-1, HPC@UCR, Alfahd, Nscale Verne Iceland Cluster, Nebius Iceland Cluster, Nebius Longcross UK Cluster, DataVolt Oxagon AI Campus, Khazna Ajman AI Data Center, STACKIT Lubbenau AI Data Center, SK Ulsan AI Data Center, Racks Central RCJM AI Campus, SoftBank Sakai AI Data Center, NEXTDC S7 Sydney AI Campus, Laurel 3, Lonestar6, Casper, Neo, RZHound, AICR, Bouchet, Hopper (Yale), Gilbreth (Purdue), Engaging (MIT), Puma (Arizona), Lynx (Arizona), Ginsburg (Columbia), Insomnia (Columbia), Skipjack (JHU), Zaratan, Klone (UW), Quest (Northwestern), Ascend (OSC), FASTER (Texas A&M), MeluXina-AI, KarolAIna, AI:AT, SLAIF, Discoverer++, Levente, MareNostrum 5 AI, Zenith, Mila Sovereign AI Research Hub, RIKYU, Cerebras Mikkeli AI Data Centre, Cerebras Montreal Inference Site (MTL-3), GroqCloud Sydney Inference Cluster, GroqCloud Kamloops Inference Cluster, Lambda Chicago AI Factory, Poolside Project Horizon, HUMAIN Project Halo, Killellan AI Growth Zone, Mistral Borlange AI Data Centre, ai& Tenstorrent Galaxy Supercluster, DESY Maxwell, CSF3, Myriad, Kathleen, Apocrita, Vera (Chalmers), UBELIX, BlueBEAR, Oxford ARC, Aire (Leeds), sciCORE, GenomeDK, Hawk (Cardiff), Sunbird, DEGIMA, China Unicom Sanjiangyuan AI Data Center, China Mobile Beijing Computing Center, Huawei Cloud Wuhu CloudMatrix Supernodes, Huawei Cloud Ulanqab CloudMatrix Supernode, iFlytek Feixing-1, Zhongke Horinger Intelligent Computing Center, China Datang Ulanqab Computing Center, ByteDance MegaScale Cluster, Moore Threads KUAE 1,000-GPU Cluster, Tianjin AI Computing Center, Shenzhen 10,000-Card Ascend Cluster, Qingyang Enflame S60 Inference Cluster, Guacolda, Nebius Béthune AI Factory, Nscale Texas GB300 Campus (Microsoft), CoreWeave Taloja AI Campus, New Lebanon Campus (Fluidstack), Crusoe Childress AI Campus, Samsung SDS Gumi AI Data Center, Stargate UK (Cobalt Park), Voltage Park H100 Cloud, CoreWeave Hammond AI Data Center, evroc Arlandastad AI Data Center, Monarch Compute Campus, OpenAI Project Camellia, PORTS-Pike Technology Campus, Microsoft Pecos Campus, Meta Sturgeon Data Centre, Pure DC Seinajoki Campus, IREN Bundey Campus, BDx Jatiluhur AI Campus, TCS HyperVault Hyderabad Campus, NFD Korea Dangjin Campus, Galaxy McGregor Campus, Fermi America Project Matador, SoftBank Hauts-de-France AI Campus, OCTOPUS, Osaka SX-ACE, NCAR CDC 6600, NCAR CDC 7600, NCAR Cray-1A S/N 3, NCAR Cray-1A S/N 14, Graywolf (NCAR Cray-3), Littlebear (NCAR CM-5), NCAR Cray T3D, Antero 2 (NCAR C90), UCT Hex, Kebnekaise, CoolMUC-3, Genius (KU Leuven), Bem 2, Fran, Ruth, Barfoot, Wheat, Makau, Coral, Flyer, Raven, Bluecrab, Hábrók, Monash M3, Niflheim, Rackham. Their component graphs are still on their own pages.

This is the installed base of the systems in this dataset, a curated seed of 1345 machines, not a census of the market. The shape is real; the absolute totals are a sample.

The read

The chart has three regimes. Until roughly 2010 there is no accelerator band at all worth speaking of: leadership compute was CPUs, and the only question was whose. The middle period, 2012 to 2020, is NVIDIA establishing that accelerators are how you buy FLOPS, first through Titan and then through the CORAL systems. The most recent regime is the one that machine-count share hides completely: AMD's share is built out of a very small number of very large machines, and a single procurement moves the line several points.

That concentration is the finding, not the ordering. A share this sensitive to individual contracts is not a market position in the sense a reader might assume from a supplier chart; it is a statement about who won the last two exascale procurements. The company pages carry the same caution in a different form, as a concentration percentage.

The CPU-only band is worth watching on its own. It does not decay to zero, and the systems in it are not legacy: Fugaku was the fastest machine in the world with no accelerator at all, and ARCHER2 and Hawk were both bought deliberately without one because their users run codes that do not vectorise onto GPUs. This chart measures supply, not suitability.