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

Supercomputers/Analysis/Interconnect share

Analysis 02

Interconnect share over time

The fabric layer consolidated twice in twelve months and almost nobody charted it. NVIDIA completed its acquisition of Mellanox in April 2020; HPE completed its acquisition of Cray in September 2019. Four independent fabric suppliers became two corporate owners, and because a spec string names the product rather than the parent, the change is invisible unless you resolve the entities first. As of 2026, Hewlett Packard Enterprise accounts for 44% of installed FLOPS by fabric and NVIDIA for 37%.

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.

  • Hewlett Packard Enterprise
  • NVIDIA
  • Shenzhen Cloud Computing Center
  • Fujitsu
  • Broadcom
  • Intel
  • Atos
  • Other (16 suppliers)
Interconnect share of installed FLOPS, 1993–2026 Stacked area chart of each interconnect supplier's share of installed Rmax by year, with suppliers rolled up to the corporate parent that owns them today. Hewlett Packard EnterpriseNVIDIAShenzhen Cloud Computing CenterFujitsuBroadcomIntelAtosOther (16 suppliers) 0%25%50%75%100%19931998200320082013201820232026 HPENVIDIAShenzhen Cloud Computing Center 1993 HPE 0.0% NVIDIA 0.0% Shenzhen Cloud Computing Center 0.0% Fujitsu 36.2% Broadcom 0.0% Intel 46.3% Atos 0.0% Other 17.4%1994 HPE 12.9% NVIDIA 0.0% Shenzhen Cloud Computing Center 0.0% Fujitsu 31.6% Broadcom 0.0% Intel 40.4% Atos 0.0% Other 15.2%1995 HPE 8.2% NVIDIA 0.0% Shenzhen Cloud Computing Center 0.0% Fujitsu 36.0% Broadcom 0.0% Intel 46.2% Atos 0.0% Other 9.6%1996 HPE 20.1% NVIDIA 0.0% Shenzhen Cloud Computing Center 0.0% Fujitsu 25.4% Broadcom 0.0% Intel 16.7% Atos 0.0% Other 37.8%1997 HPE 32.0% NVIDIA 0.0% Shenzhen Cloud Computing Center 0.0% Fujitsu 12.2% Broadcom 0.0% Intel 37.7% Atos 0.0% Other 18.1%1998 HPE 42.8% NVIDIA 0.0% Shenzhen Cloud Computing Center 0.0% Fujitsu 5.4% Broadcom 0.0% Intel 16.9% Atos 0.0% Other 34.9%1999 HPE 44.8% NVIDIA 0.0% Shenzhen Cloud Computing Center 0.0% Fujitsu 4.2% Broadcom 0.0% Intel 13.1% Atos 0.0% Other 37.8%2000 HPE 25.6% NVIDIA 0.0% Shenzhen Cloud Computing Center 0.0% Fujitsu 7.3% Broadcom 0.0% Intel 7.5% Atos 0.0% Other 59.7%2001 HPE 15.7% NVIDIA 0.0% Shenzhen Cloud Computing Center 0.0% Fujitsu 4.1% Broadcom 0.0% Intel 4.6% Atos 0.0% Other 75.6%2002 HPE 4.7% NVIDIA 0.0% Shenzhen Cloud Computing Center 0.0% Fujitsu 1.2% Broadcom 0.0% Intel 1.4% Atos 0.0% Other 92.7%2003 HPE 13.5% NVIDIA 7.1% Shenzhen Cloud Computing Center 0.0% Fujitsu 0.8% Broadcom 0.0% Intel 0.9% Atos 0.0% Other 77.7%2004 HPE 2.9% NVIDIA 1.5% Shenzhen Cloud Computing Center 0.0% Fujitsu 0.2% Broadcom 0.0% Intel 0.2% Atos 0.0% Other 95.2%2005 HPE 4.1% NVIDIA 1.2% Shenzhen Cloud Computing Center 0.0% Fujitsu 0.1% Broadcom 0.0% Intel 0.2% Atos 0.0% Other 94.4%2006 HPE 3.5% NVIDIA 1.7% Shenzhen Cloud Computing Center 0.0% Fujitsu 1.1% Broadcom 0.0% Intel 0.1% Atos 0.0% Other 93.6%2007 HPE 40.1% NVIDIA 3.9% Shenzhen Cloud Computing Center 0.0% Fujitsu 0.5% Broadcom 0.0% Intel 0.0% Atos 0.0% Other 55.5%2008 HPE 12.3% NVIDIA 67.4% Shenzhen Cloud Computing Center 0.0% Fujitsu 0.1% Broadcom 0.0% Intel 0.0% Atos 0.0% Other 20.2%2009 HPE 21.6% NVIDIA 62.3% Shenzhen Cloud Computing Center 0.0% Fujitsu 0.1% Broadcom 0.0% Intel 0.0% Atos 0.0% Other 16.1%2010 HPE 28.6% NVIDIA 50.1% Shenzhen Cloud Computing Center 0.0% Fujitsu 0.0% Broadcom 0.0% Intel 0.0% Atos 0.0% Other 21.3%2011 HPE 21.7% NVIDIA 37.9% Shenzhen Cloud Computing Center 0.0% Fujitsu 23.8% Broadcom 0.0% Intel 3.7% Atos 0.0% Other 12.9%2012 HPE 23.4% NVIDIA 26.4% Shenzhen Cloud Computing Center 0.0% Fujitsu 9.6% Broadcom 0.0% Intel 1.3% Atos 0.0% Other 39.4%2013 HPE 15.2% NVIDIA 25.0% Shenzhen Cloud Computing Center 0.0% Fujitsu 5.6% Broadcom 0.0% Intel 0.8% Atos 0.0% Other 53.3%2014 HPE 16.6% NVIDIA 29.3% Shenzhen Cloud Computing Center 0.0% Fujitsu 5.0% Broadcom 0.0% Intel 1.0% Atos 0.0% Other 48.1%2015 HPE 21.1% NVIDIA 33.8% Shenzhen Cloud Computing Center 0.0% Fujitsu 7.0% Broadcom 0.0% Intel 1.1% Atos 0.0% Other 37.0%2016 HPE 19.2% NVIDIA 26.7% Shenzhen Cloud Computing Center 0.0% Fujitsu 4.4% Broadcom 0.0% Intel 6.2% Atos 0.0% Other 43.5%2017 HPE 24.2% NVIDIA 27.3% Shenzhen Cloud Computing Center 0.0% Fujitsu 3.3% Broadcom 0.0% Intel 9.7% Atos 1.9% Other 33.6%2018 HPE 16.0% NVIDIA 47.8% Shenzhen Cloud Computing Center 0.0% Fujitsu 2.0% Broadcom 0.0% Intel 10.5% Atos 1.1% Other 22.6%2019 HPE 14.2% NVIDIA 51.4% Shenzhen Cloud Computing Center 0.0% Fujitsu 1.9% Broadcom 0.0% Intel 11.3% Atos 1.0% Other 20.2%2020 HPE 8.4% NVIDIA 47.7% Shenzhen Cloud Computing Center 0.0% Fujitsu 24.3% Broadcom 0.0% Intel 6.8% Atos 0.6% Other 12.2%2021 HPE 12.8% NVIDIA 51.6% Shenzhen Cloud Computing Center 0.0% Fujitsu 19.8% Broadcom 0.0% Intel 5.3% Atos 1.4% Other 9.0%2022 HPE 46.7% NVIDIA 34.3% Shenzhen Cloud Computing Center 0.0% Fujitsu 10.5% Broadcom 0.0% Intel 2.8% Atos 0.8% Other 4.8%2023 HPE 46.8% NVIDIA 40.3% Shenzhen Cloud Computing Center 0.0% Fujitsu 7.4% Broadcom 0.0% Intel 1.5% Atos 0.5% Other 3.5%2024 HPE 56.1% NVIDIA 35.0% Shenzhen Cloud Computing Center 0.0% Fujitsu 4.7% Broadcom 0.0% Intel 0.9% Atos 1.1% Other 2.2%2025 HPE 49.1% NVIDIA 42.7% Shenzhen Cloud Computing Center 0.0% Fujitsu 3.7% Broadcom 1.2% Intel 0.6% Atos 0.9% Other 1.8%2026 HPE 43.7% NVIDIA 37.1% Shenzhen Cloud Computing Center 12.6% Fujitsu 3.0% Broadcom 1.0% Intel 0.4% Atos 0.7% Other 1.4%
Share of installed Rmax by fabric supplier. Mellanox counts toward NVIDIA from 2020 and Cray toward HPE from 2019, because the question is who owns the business now.
Download this chart's data (CSV) ↓
Table view: Interconnect share of installed FLOPS
YearHewlett Packard EnterpriseNVIDIAShenzhen Cloud Computing CenterFujitsuBroadcomIntelAtosOther (16 suppliers)Installed total
19930.0%0.0%0.0%36.2%0.0%46.3%0.0%17.4%342.3 GFlop/s
199412.9%0.0%0.0%31.6%0.0%40.4%0.0%15.2%392.9 GFlop/s
19958.2%0.0%0.0%36.0%0.0%46.2%0.0%9.6%618.9 GFlop/s
199620.1%0.0%0.0%25.4%0.0%16.7%0.0%37.8%1.71 TFlop/s
199732.0%0.0%0.0%12.2%0.0%37.7%0.0%18.1%3.59 TFlop/s
199842.8%0.0%0.0%5.4%0.0%16.9%0.0%34.9%8.01 TFlop/s
199944.8%0.0%0.0%4.2%0.0%13.1%0.0%37.8%10.31 TFlop/s
200025.6%0.0%0.0%7.3%0.0%7.5%0.0%59.7%18.10 TFlop/s
200115.7%0.0%0.0%4.1%0.0%4.6%0.0%75.6%29.50 TFlop/s
20024.7%0.0%0.0%1.2%0.0%1.4%0.0%92.7%98.80 TFlop/s
200313.5%7.1%0.0%0.8%0.0%0.9%0.0%77.7%144.79 TFlop/s
20042.9%1.5%0.0%0.2%0.0%0.2%0.0%95.2%670.00 TFlop/s
20054.1%1.2%0.0%0.1%0.0%0.2%0.0%94.4%859.77 TFlop/s
20063.5%1.7%0.0%1.1%0.0%0.1%0.0%93.6%1.02 PFlop/s
200740.1%3.9%0.0%0.5%0.0%0.0%0.0%55.5%2.19 PFlop/s
200812.3%67.4%0.0%0.1%0.0%0.0%0.0%20.2%11.07 PFlop/s
200921.6%62.3%0.0%0.1%0.0%0.0%0.0%16.1%15.77 PFlop/s
201028.6%50.1%0.0%0.0%0.0%0.0%0.0%21.3%25.85 PFlop/s
201121.7%37.9%0.0%23.8%0.0%3.7%0.0%12.9%44.26 PFlop/s
201223.4%26.4%0.0%9.6%0.0%1.3%0.0%39.4%122.7 PFlop/s
201315.2%25.0%0.0%5.6%0.0%0.8%0.0%53.3%208.0 PFlop/s
201416.6%29.3%0.0%5.0%0.0%1.0%0.0%48.1%234.3 PFlop/s
201521.1%33.8%0.0%7.0%0.0%1.1%0.0%37.0%303.1 PFlop/s
201619.2%26.7%0.0%4.4%0.0%6.2%0.0%43.5%485.9 PFlop/s
201724.2%27.3%0.0%3.3%0.0%9.7%1.9%33.6%638.5 PFlop/s
201816.0%47.8%0.0%2.0%0.0%10.5%1.1%22.6%1.05 EFlop/s
201914.2%51.4%0.0%1.9%0.0%11.3%1.0%20.2%1.17 EFlop/s
20208.4%47.7%0.0%24.3%0.0%6.8%0.6%12.2%1.97 EFlop/s
202112.8%51.6%0.0%19.8%0.0%5.3%1.4%9.0%2.48 EFlop/s
202246.7%34.3%0.0%10.5%0.0%2.8%0.8%4.8%4.68 EFlop/s
202346.8%40.3%0.0%7.4%0.0%1.5%0.5%3.5%7.03 EFlop/s
202456.1%35.0%0.0%4.7%0.0%0.9%1.1%2.2%11.10 EFlop/s
202549.1%42.7%0.0%3.7%1.2%0.6%0.9%1.8%14.04 EFlop/s
202643.7%37.1%12.6%3.0%1.0%0.4%0.7%1.4%17.42 EFlop/s
How this was calculated

Identical construction to the accelerator chart: each in-service system contributes its last reported Rmax to every year it was live, split evenly where a system has more than one fabric supplier.

Every system in this dataset has at least one interconnect edge, so unlike the accelerator chart there is no “none” band: a machine without a fabric is not a machine.

The rollup is the whole point of this chart. Suppliers are recorded as the company that shipped the fabric at build time and then resolved through vendor_aliases to today's parent. Read without that step, Summit's dual-rail EDR belongs to a company that no longer exists independently, and the concentration this chart shows does not appear.

One caveat that cuts against the chart: vertically integrated systems understate their supplier here. IBM's BlueGene fabrics and Fujitsu's Tofu are on the processor package, and appear in no spec string as a third-party product. They are recorded as IBM and Fujitsu edges, but a dataset built only from fabric-vendor mentions would miss them entirely.

Installed base of the 1345 systems in this dataset, not a census of the market.

The read

Three structural facts sit in this chart. The first is that the fabric layer has always been more concentrated than the accelerator layer, and for most of the period it was concentrated in vendors who also supplied the machine: Cray's SeaStar, Gemini and Aries; IBM's torus networks; Fujitsu's Tofu. The independent merchant fabric is a relatively recent and relatively brief phenomenon.

The second is that the 2019–2020 acquisitions removed the last two large independent fabric businesses in the same twelve months. After them, the two companies that supply most accelerator FLOPS also supply most fabric FLOPS. Accelerator share and fabric share stopped being independent variables, which matters to anyone modelling concentration risk from either one alone.

The third is Omni-Path, which is the counterexample worth keeping. Intel discontinued it in 2019 and it did not die: Cornelis Networks took it over in 2020 and continues it. It is the only case in this dataset of a fabric outliving its vendor's interest in it, and it is why the parts table records Cornelis as the current supplier of record while the edges from Stampede2, SuperMUC-NG and TSUBAME3.0 still correctly name Intel as the supplier at build.