Compute·Atlas

Supercomputers/Analysis/Public-market exposure

Analysis 06

Public-market exposure by ticker

11 listed companies supply parts to the systems in this dataset. The useful number is not the total: it is the split inside it. NVDA spreads its exposure across 27 systems with only 27% in its largest; 6702 has 94% of its exposure in Fugaku alone. Those are different risks wearing the same headline.

This is a breadth measure, not revenue and not attributable share. An integrator and a memory supplier in the same machine each count its full Rmax here, because we have no public basis for splitting a system's value between roles. A large bar means a company appears in machines representing a lot of installed capability. It does not mean it captured a proportionate share of the spend, and it must not be read as a sales figure. Nothing on this site is investment advice.

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: 32 of 52 systems have a readable citation that names them, and 4 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 52 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.

  • Largest single system
  • All other systems
Aggregate FLOPS exposure by ticker Horizontal bars of aggregate installed FLOPS in systems each listed company supplies any part of, split into the largest single system and everything else. HPE Hewlett Packard Enterprise Hewlett Packard Enterprise (HPE) 21 systems 5135 PFlop/s aggregate largest: El Capitan at 1742 PFlop/s (34%) Hewlett Packard Enterprise (HPE) 21 systems 5135 PFlop/s aggregate largest: El Capitan at 1742 PFlop/s (34%) 34% in one AMD AMD AMD (AMD) 15 systems 3734 PFlop/s aggregate largest: El Capitan at 1742 PFlop/s (47%) AMD (AMD) 15 systems 3734 PFlop/s aggregate largest: El Capitan at 1742 PFlop/s (47%) 47% in one INTC Intel Intel (INTC) 18 systems 2288 PFlop/s aggregate largest: Aurora at 1012 PFlop/s (44%) Intel (INTC) 18 systems 2288 PFlop/s aggregate largest: Aurora at 1012 PFlop/s (44%) 44% in one NVDA NVIDIA NVIDIA (NVDA) 27 systems 2060 PFlop/s aggregate largest: Eagle at 561 PFlop/s (27%) NVIDIA (NVDA) 27 systems 2060 PFlop/s aggregate largest: Eagle at 561 PFlop/s (27%) 27% in one IBM IBM IBM (IBM) 12 systems 693 PFlop/s aggregate largest: Leonardo at 241 PFlop/s (35%) IBM (IBM) 12 systems 693 PFlop/s aggregate largest: Leonardo at 241 PFlop/s (35%) 35% in one MSFT Microsoft Microsoft (MSFT) 1 systems 561 PFlop/s aggregate largest: Eagle at 561 PFlop/s (100%) Microsoft (MSFT) 1 systems 561 PFlop/s aggregate largest: Eagle at 561 PFlop/s (100%) 100% in one 6702 Fujitsu Fujitsu (6702) 3 systems 472 PFlop/s aggregate largest: Fugaku at 442 PFlop/s (94%) Fujitsu (6702) 3 systems 472 PFlop/s aggregate largest: Fugaku at 442 PFlop/s (94%) 94% in one ATO Atos Atos (ATO) 5 systems 430 PFlop/s aggregate largest: Leonardo at 241 PFlop/s (56%) Atos (ATO) 5 systems 430 PFlop/s aggregate largest: Leonardo at 241 PFlop/s (56%) 56% in one DELL Dell Technologies Dell Technologies (DELL) 2 systems 34 PFlop/s aggregate largest: Frontera at 24 PFlop/s (69%) Dell Technologies (DELL) 2 systems 34 PFlop/s aggregate largest: Frontera at 24 PFlop/s (69%) 69% in one 992 Lenovo Lenovo (992) 2 systems 26 PFlop/s aggregate largest: SuperMUC-NG at 19 PFlop/s (75%) Lenovo (992) 2 systems 26 PFlop/s aggregate largest: SuperMUC-NG at 19 PFlop/s (75%) 75% in one 6701 NEC NEC (6701) 1 systems 0 PFlop/s aggregate largest: Earth Simulator at 0 PFlop/s (100%) NEC (6701) 1 systems 0 PFlop/s aggregate largest: Earth Simulator at 0 PFlop/s (100%) 100% in one 0k1k3k4k5k
Bar length is aggregate installed Rmax across every system the company supplies any part of. The first segment is its single largest system, so a bar that is mostly one colour is a company whose exposure is one programme.
Table view: Aggregate FLOPS exposure by ticker
TickerCompanySystemsAggregate RmaxConcentrationLargest systemRolesSpan
HPE · NYSEHewlett Packard Enterprise215.14 EFlop/s34%El CapitanCooling, CPU, Integrator, Interconnect, Operating system, Storage2002–2024
AMD · NASDAQAMD153.73 EFlop/s47%El CapitanAccelerator, CPU2008–2024
INTC · NASDAQIntel182.29 EFlop/s44%AuroraAccelerator, Cooling, CPU, Integrator, Interconnect, Storage1997–2023
NVDA · NASDAQNVIDIA272.06 EFlop/s27%EagleAccelerator, Cooling, CPU, Integrator, Interconnect2008–2025
IBM · NYSEIBM12693.4 PFlop/s35%LeonardoAccelerator, Cooling, CPU, Integrator, Interconnect, Operating system, Storage2000–2023
MSFT · NASDAQMicrosoft1561.2 PFlop/s100%EagleIntegrator2023–2023
6702 · TSEFujitsu3472.4 PFlop/s94%FugakuCooling, CPU, Integrator, Interconnect2011–2020
ATO · EPAAtos5430.4 PFlop/s56%LeonardoCooling, Integrator2020–2025
DELL · NYSEDell Technologies234.20 PFlop/s69%FronteraIntegrator2017–2019
992 · HKEXLenovo225.95 PFlop/s75%SuperMUC-NGCooling, Integrator2017–2018
6701 · TSENEC135.86 TFlop/s100%Earth SimulatorCooling, CPU, Integrator, Interconnect2002–2002
How this was calculated

A company is counted as present in a system if it supplies any component edge in any role: integrator, CPU, accelerator, interconnect, storage, cooling, power or operating system. Aggregate Rmax is the sum of those systems' last reported performance figures.

Exposure follows the acquisition chain. Systems shipped by Cray count toward HPE, and fabrics shipped by Mellanox count toward NVIDIA, because the question this page asks is which listed entity carries the exposure today. The build-time supplier is preserved on every system page; it is only the rollup that moves.

Rows are grouped by ticker rather than by company, so a subsidiary and its parent sharing a listing appear once. Eviden and Atos are the case here: they share ATO, and listing them separately would double-count European exposure.

Concentration is the largest single system's Rmax as a share of the aggregate. It is a deliberately crude measure and reads correctly at both ends: 100% means the exposure is one machine, and a low figure means it is spread.

What this page will not do. §6.6 asks for FLOPS “attributable” to each ticker, which implies weighting a system's value across the roles in it. We do not publish that, because we have no public basis for the weights and an invented weighting would be indistinguishable from a real one once it had been quoted. Role-weighted attribution ships when a defensible method does, and not before.

The read

The first thing the split shows is that concentration and scale are not the same property, and the largest exposures are not the most concentrated ones. NVIDIA appears in more systems than anyone here and has the lowest concentration of the large names, it is in almost everything, in several different roles, across two decades. HPE has the largest aggregate but a third of it sits in a single machine.

The second is that role breadth matters as much as system count. A company present as integrator, interconnect, storage, cooling and operating system in the same machine has five relationships with that customer; a company present only as the accelerator has one, and it is the one most easily substituted at the next procurement. The Roles column in the table is the part of this analysis that a share chart cannot show.

The third is the tail. Several listed suppliers carried in this graph appear in exactly one system, and one appears in a machine that has been retired for years. Those rows are not noise: they are what a supplier's position looks like on the way in or on the way out, and the span column says which.

A caution about what is missing. This page can only see companies that appear as a supplier edge in our data, which is heavily weighted toward compute, fabric and integration. Memory, foundry and facility suppliers are carried in the parts taxonomy but have few system edges, so TSMC, which fabricates essentially all the leading-edge silicon on this site: does not appear here at all. The most concentrated dependency in the entire graph is invisible on this chart, and that is a limitation of the edge data, not a finding.

All listed suppliers

TickerCompanySystemsAggregate Rmax ConcentrationLargest systemRoles
HPENYSE Hewlett Packard Enterprise 21 5.14 EFlop/s 34% El Capitan cooling, cpu, integrator, interconnect, os, storage
AMDNASDAQ AMD 15 3.73 EFlop/s 47% El Capitan accelerator, cpu
INTCNASDAQ Intel 18 2.29 EFlop/s 44% Aurora accelerator, cooling, cpu, integrator, interconnect, storage
NVDANASDAQ NVIDIA 27 2.06 EFlop/s 27% Eagle accelerator, cooling, cpu, integrator, interconnect
IBMNYSE IBM 12 693.4 PFlop/s 35% Leonardo accelerator, cooling, cpu, integrator, interconnect, os, storage
MSFTNASDAQ Microsoft 1 561.2 PFlop/s 100% Eagle integrator
6702TSE Fujitsu 3 472.4 PFlop/s 94% Fugaku cooling, cpu, integrator, interconnect
ATOEPA Atos 5 430.4 PFlop/s 56% Leonardo cooling, integrator
DELLNYSE Dell Technologies 2 34.20 PFlop/s 69% Frontera integrator
992HKEX Lenovo 2 25.95 PFlop/s 75% SuperMUC-NG cooling, integrator
6701TSE NEC 1 35.86 TFlop/s 100% Earth Simulator cooling, cpu, integrator, interconnect