COMPUTE ATLAS Supercomputer supply-chain graph
1153 systems 482 sites Sourced data

Data

Download the dataset

Every page on this site is rendered from these 22 tables at build time, so this is not a special extract: it is the same CSVs, served as they are. 11,710 rows total.

What this is and is not. This is our own compiled dataset: systems, parts, companies and the edges between them, each row carrying its own source. It is not a copy of the TOP500 list; list_appearances holds pointers only (system, list, edition, rank, a link to the canonical entry), which is the same thing every system page already shows. See methodology for the sourcing policy. Free to use with attribution back to computeatlas.net; nothing here is investment advice.

TableWhat it isRows
Systems One row per supercomputer: operator, site, dates, performance, confidence and sources. 1,153 Download
System components The edge table into the parts graph: which system uses which part, from which supplier, in which role. 4,258 Download
Companies Suppliers, operators and integrators, with parent-company chains for tracking acquisitions. 714 Download
Parts The parts taxonomy: accelerators, CPUs, interconnect, storage, cooling and more, each with its own sources. 275 Download
Part categories The category tree parts.csv hangs off. 11 Download
Sites Facilities: location, utility, substation, PUE and cooling, where disclosed. 482 Download
List appearances Our stored pointers only: system, list, edition and rank, each linking to the canonical entry. Not a copy of any ranking. 810 Download
Vendor aliases Name variants the ingest pipeline resolves to a canonical company id. 89 Download
Changelog The dated log behind /changes: what was added, upgraded, corrected or flagged, and when. 1,354 Download
Open questions Where two credible sources disagree on a specific field, both sides linked. 51 Download
Expected windows Planned systems and the delivery window their operator states. 16 Download
Company sectors Which of eight sectors each operator belongs to, with a one-line reason: the table behind the sector view. 543 Download
Capacity claims Operator- or developer-stated power capacity in megawatts, by scope and stage, each tied to a page the system already cites. 132 Download
Quantum systems Named quantum processors and machines: modality, qubit count, operator, status and any link to a co-located supercomputer. 100 Download
Quantum specs Per-machine spec sheet for quantum computers: qubit type, topology, native gates, cooling, gate times, software, access and lineage, each with its source and confidence. 889 Download
Quantum benchmarks Benchmark numbers and milestone claims published by quantum-hardware builders, each with its source and whether it is vendor reported, peer reviewed or independent. 93 Download
AI datacenter metrics Facility and cluster figures beyond a GPU count for AI datacenters: IT and facility power, PUE, WUE, stated investment, rack density, utilisation and goodput, each with its basis and source. 371 Download
AI benchmark results System-scale AI benchmark records (MLPerf rounds, InferenceMAX, HPL-MxP) stored as pointers with the submitter, accelerator count, result and a link to the published result. 37 Download
Compute deals Announced compute capacity, datacenter lease, chip supply and joint-venture agreements between AI labs, clouds and datacenter developers, with stated capacity, accelerators and headline value. 76 Download
AI campus power sources How AI campuses are powered: grid, on-site gas, utility gas plants, nuclear, renewables, hydro and storage, with capacity, status and basis. 70 Download
Accelerator specifications Dense peak FLOPS by precision, memory, bandwidth, TDP and scale-up domain for the accelerators used in AI datacenters, with the vendor source for each figure. 160 Download
System events Point-in-time events on a system (an upgrade, a rebuild) that do not fit the system row itself. 26 Download

Excluded on purpose: our internal source-verification log, and the curated news feed (already fully readable at /news, and written in our own editorial voice rather than structured data). Prefer JSON? /facts.json is machine-readable already; the rest of these are CSV because that is how the site itself stores them.