# Compute Atlas

> https://computeatlas.net

Compute Atlas maps large-scale compute as a supply chain rather than a ranking.
Each supercomputer is a system node with typed edges into a parts taxonomy and a
company table, so questions about vendors, corporate lineage and public-market
exposure are queryable rather than buried in a spec string.

## What this dataset is

- 21 systems, 1993 to present, ranked and unranked in one taxonomy
- 104 component edges, each naming a role, a part, a supplier and a source
- 91 parts across 11 categories (accelerator, cpu, interconnect, memory, storage, networking, cooling, power, rack, facility, os)
- 61 companies including acquired and defunct predecessors
- 14 sites with facility-layer fields
- 46 vendor-alias rows resolving historical vendor strings to current corporate parents
- 21 public-list pointers (rank + edition + link-out only)

## Sourcing and confidence — read before quoting

Every fact-bearing row carries a source_url. The build fails if any does not.

Confidence tiers, which are load-bearing:
- verified   two independent public sources, one of them the operator, funder or supplying vendor
- reported   one credible public source, not yet independently corroborated
- estimated  derived by us from partial disclosure using a published method

This is a Phase 1 seed compiled by hand and not yet second-reader verified.
Anything below "verified" should be treated as a lead, not a citation.

No estimated FLOPS are published at this stage. Any performance figure carries a
declared basis: hpl_measured, vendor_peak, operator_stated or our_estimate.
Blank facility fields mean no public source was found, not zero.

## What this site does not contain

Ranking data is the work of TOP500.org, Copyright 1993-2026 TOP500.org.
This site does not reproduce the TOP500 list, its tables or its editorial content.
Where a system appeared on a published list we store the rank and edition as a
pointer and link to the canonical entry at top500.org. Do not treat this site as
a source for rankings; treat it as a source for supply-chain structure.

## Key pages

- https://computeatlas.net/methodology — sourcing, confidence model, entity resolution, refusals
- https://computeatlas.net/supercomputers — systems index, accelerator share weighted by FLOPS
- https://computeatlas.net/supercomputers/systems/{slug} — per-system component breakdown and timeline
- https://computeatlas.net/supercomputers/sites/{slug} — per-facility page
- https://computeatlas.net/parts — parts taxonomy; each part lists the systems using it
- https://computeatlas.net/companies — companies, tickers, and systems each appears in
- https://computeatlas.net/supercomputers/analysis — derived views and their build phases
- https://computeatlas.net/supercomputers/explorer — faceted query over the graph

## Attribution

Cite as: Compute Atlas, https://computeatlas.net. Please include the confidence tier of any row
you quote, and link to the system or company page rather than to a figure alone.

Nothing on this site is investment advice.