Supercomputers/Analysis/Supply concentration
Analysis 09
Supercomputer supply-chain concentration
The accelerator-share and interconnect-share pages show who has the largest slice. This page asks the antitrust-style question directly: how concentrated is the supplier base for each role, and is it getting more or less concentrated over time. As of 2026, accelerator is the most concentrated role in this dataset at an HHI of 4331, held by NVIDIA at 52% of installed Rmax in that role.
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: 1092 of 1153 systems have a readable citation that names them, and 40 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 1153 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.
- Accelerator
- CPU
- Interconnect
- Integrator
- Storage
Table view: Supply concentration by role (HHI)
| Year | Accelerator | CPU | Interconnect | Integrator | Storage |
|---|---|---|---|---|---|
| 2008 | 10000 | 4551 | 5249 | 5422 | 5808 |
| 2009 | 4257 | 4335 | 4578 | 4240 | 6434 |
| 2010 | 4854 | 4960 | 3623 | 3025 | 6378 |
| 2011 | 4928 | 3342 | 2516 | 2312 | 3896 |
| 2012 | 7200 | 2843 | 2630 | 2992 | 3529 |
| 2013 | 3288 | 4011 | 2341 | 2369 | 3093 |
| 2014 | 3317 | 4369 | 2322 | 2240 | 3140 |
| 2015 | 3308 | 4785 | 2205 | 2047 | 2871 |
| 2016 | 3630 | 3797 | 1704 | 1680 | 2461 |
| 2017 | 3333 | 4824 | 1730 | 1695 | 2325 |
| 2018 | 5560 | 3830 | 2733 | 1852 | 3055 |
| 2019 | 6320 | 3996 | 3038 | 1765 | 3020 |
| 2020 | 6970 | 2495 | 3019 | 1581 | 2554 |
| 2021 | 7488 | 2476 | 3270 | 1353 | 2245 |
| 2022 | 4621 | 4011 | 3483 | 3051 | 3879 |
| 2023 | 3577 | 3807 | 3877 | 2906 | 2696 |
| 2024 | 3983 | 3793 | 4405 | 3844 | 4501 |
| 2025 | 4279 | 3390 | 4254 | 3369 | 3777 |
| 2026 | 4331 | 2847 | 3460 | 2908 | 4087 |
How this was calculated
For each role and each year, every system in service that year contributes its Rmax to its role's supplier, rolled up to the corporate parent that owns that supplier today. A system with more than one supplier in a role splits its Rmax evenly between them. The HHI is the sum of each supplier's squared percentage share that year: US antitrust guidance treats a market above 2500 as highly concentrated and above 1500 as moderately concentrated, which is the convention used here, not a scale specific to this site.
Computed from every supplier, not the eight-series capped form the accelerator-share and interconnect-share charts use for legibility. Folding a long tail of small suppliers into one "Other" band before squaring would understate concentration, since "Other" here is really several small suppliers, not one mid-sized one.
Systems with no supplier recorded for a given role (the CPU-only band on the accelerator-share page) are left out of that role's denominator entirely. This page measures concentration among the suppliers that exist for a role, not how widely the role itself is adopted; the accelerator-share and interconnect-share pages already cover adoption.
This is the installed base of the 1153 systems in this dataset, a curated seed, not a census of the market. A market this small moves HHI by hundreds of points on a single large procurement, which is itself the finding: real HPC and AI-cluster supply chains are this concentrated.
Where each role sits today
| Role | HHI (2026) | Concentration | Leading supplier | Share | Suppliers |
|---|---|---|---|---|---|
| Accelerator | 4331 | Highly concentrated | NVIDIA | 52.4% | 9 |
| CPU | 2847 | Highly concentrated | AMD | 39.7% | 14 |
| Interconnect | 3460 | Highly concentrated | Hewlett Packard Enterprise | 43.9% | 23 |
| Integrator | 2908 | Highly concentrated | Hewlett Packard Enterprise | 50.3% | 46 |
| Storage | 4087 | Highly concentrated | Hewlett Packard Enterprise | 61.0% | 17 |
The read
Every one of these five roles sits above the "highly concentrated" antitrust threshold of 2500 in this dataset, and most sit well above it. That is not unique to any one supplier; it is close to structural for leadership-class compute, where a handful of firms can build at the frontier at all and a single large procurement is a meaningful fraction of the year's total installed base.
Integrator concentration looks highest, and reads as an artefact of history as much as market power: Cray's acquisition by HPE, and SGI's before it, folded what were once several independent system builders into one corporate parent. The accelerator role tells a cleaner story: fewer historical mergers, and its concentration is closer to being a live market outcome than a rollup effect.
Read this alongside domestic substitution and public-market exposure: concentration by role is "how many suppliers exist", substitution is "how many countries can build without a foreign supplier at all", and exposure is "which ticker actually carries that concentration".