ComputeFinder

Home / Data

Data downloads

Every compiled dataset we publish as JSON and spreadsheet-ready CSV, with its source basis and the date it was last checked.

Provider exports include pricing_url, the official link already recorded on the provider page. It contains no quoted price. A missing link is null in JSON and empty in CSV; it does not mean a provider has no pricing page. Links and their contents can change and are not newly verified by an export. Check the provider's current conditions before choosing a route.

Detailed-site exports include grid_connection, pue, water_use and chips as separate scoped evidence arrays, or null when not recorded (empty in CSV). Each statement retains its source conditions, dates and an actually known immutable source version or null. Source versions are not inferred from data-export versions or review dates. Design, target and measured PUE are distinct; water figures retain their measurement boundary. These fields do not establish whole-site status or available compute and are not aggregated.

Detailed-site inspection_limits describe the scope of selected checks separately from recorded values. A note can identify an uninspected field, a value not stated in one inspected source or a field not admitted. It never counts towards field coverage or proves a whole application omits information. The recorded scope date does not refresh the parent record.

Selected commercial service scopes have JSON/CSV for three provider-attributed categories. The existing programme routes retain their own evidence. National standards-catalogue JSON/CSV covers only the reviewed DIN German Part 1 record, not all standards or their normative text.

Licence and attribution

ComputeFinder's original summaries, labels and compilation rights are licensed under Creative Commons Attribution 4.0 International (CC BY 4.0). You may share and adapt these contributions, including commercially, with appropriate credit to ComputeFinder, a licence link and an indication of changes. Material is provided as-is without warranties.

This grant covers only rights ComputeFinder controls. Record-specific adaptation licences take precedence for their stated scope: the 800 VDC explainer carries CC BY-SA 4.0 attribution and modification notices in HTML and exports. Third-party material retains its source licences, attribution requirements and other applicable terms. Each record links its sources. Their reuse terms have not been comprehensively cleared: a source link is not permission to republish a source's text or database. Uses that need no copyright or database permission are unaffected.

Private correction submissions, email addresses, website code and branding are excluded. Epoch AI's separate download retains its own licence and credit. See sources and licences for dataset-specific scope.

Catalogue JSON files use the same top-level evidence envelope. CSV files repeat that envelope in cf_ columns on every row, preserve nested values as JSON text, and keep missing values empty rather than converting them to zero. No file contains prices. Single providers and countries are also available as JSON at /data/providers/<slug>.json and /data/countries/<slug>.json. Software can discover these routes from the OpenAPI description. Model Context Protocol clients can connect to https://computefinder.net/mcp to list and read the JSON datasets and entity templates. The MCP endpoint is read-only and implements the stateless 2026-07-28 Streamable HTTP protocol. Public data responses allow read-only cross-origin access.

Datasets, record counts and sources
DatasetRecordsJSONCSVSource basisLast checked
Providers84JSONCSVEach provider's own public pages, read by ComputeFinder. Link-out only.2026-10-23
Countries and territories111JSONCSVGovernment, regulator, grid and trade press pages. Reported items are labelled as the source gives them.2026-10-03
Policy register17JSONCSVPrimary legal and government sources, one record per jurisdiction with its instruments.2026-10-02
Supply chain118JSONCSVCompany and technology pages, plus trade press where a constraint is marked reported.2026-10-14
Free and funded compute programmes52JSONCSVEach programme's own page.2026-11-14
News108JSONCSVSource-linked news accounts labelled official, corroborated or source claim, plus separately labelled editorial corrections and explicitly scoped source guides. A source guide supplies reading routes rather than a verified account of every linked announcement. An editorial correction is not a newly verified source account; archive and correction dates remain distinct. page_url links the exact ComputeFinder account; source_url remains the original evidence.2026-11-12
Export controls and compute thresholds23JSONCSVOfficial gazettes and regulator texts. Information only, not legal advice.2026-10-03
Grid rules for large loads11JSONCSVGrid operator, regulator and government pages, summarised in our own words. Each record links its source.2026-10-31
Large-load demand evidence2JSONCSVTwo scoped operator statements distinguish forecast adjustments from a load queue and overlapping reservation/ordinary-queue categories. No demand-capacity total, project list or connection guarantee is supplied.2026-10-31
Data centre records141JSONCSVOperator, planning and regulator pages, plus explicitly attributed publisher records. Source-reported identity is not a verified operating stage. Capacity is per metric, stage and basis and is never to be summed.2026-11-12
Research in context7JSONCSVOriginal attributed accounts of research, methods and guidance with scope, uncertainty, reuse terms and source-specific exclusions. Guidance is not an empirical finding or modelled result by itself. No underlying dataset, Annex, chart, proprietary input or local-project inference is imported.2026-10-31
Claim reviews4JSONCSVScoped editorial reviews with exact claim, attribution, typed citations, methods, finding, reason and unresolved questions. Empty evidence lists mean no evidence recorded, not none exists.2026-11-06
Named-campus record links1JSONCSVOne precisely reviewed same-named-campus contextual inference between retained publisher and detailed records. It does not identify a shared building, phase, parcel, owner/user, metric, operation or coordinate, and does not deduplicate source records.2026-11-01
Unscored usefulness evidence samples3JSONCSVDated scoped original-source and repository observations for Trust, Evidence depth and Freshness. The remaining seven dimensions are unassessed; counts do not calibrate a score or establish visitor benefit.2026-10-31
Repair, reuse and retirement evidence3JSONCSVThree reviewed source scopes distinguish historical technical guidance, an issued named grid-asset licence condition and an operator-reported named-system retirement. The retirement report does not establish disposal, recovery or site closure/restoration; the older records are not promoted to completed outcomes.2026-11-01
Selected commercial service scopes3JSONCSVThree selected provider-attributed categories: AWS migration staffing guidance, Canonical physical-server provisioning/support and Salute commissioning. Separate from the existing programme routes; not a representative services census, ranking or endorsement.2026-10-31
Scoped national standards catalogue evidence1JSONCSVOne reviewed DIN German Part1 catalogue record, not all standards, normative clauses, every national edition or facility conformity. The same canonical object drives the scoped EN50600 entry.2026-10-31
Frontier AI data centressee fileJSONNot providedEpoch AI, Frontier Data Centers. Licence CC BY 4.0, attribution to Epoch AI required.See file

Connected project evidence

Repair, reuse and retirement evidence has JSON and CSV for three reviewed starting points. Historical guidance, an issued grid-asset condition and an operator-reported system-retirement event remain distinct. The retirement report does not establish disposal, recovery or site closure/restoration. Exact source-specific rights, locators and separate clocks accompany each record.

Inspect large-load demand evidence, with matching JSON and CSV. These two scoped accounts distinguish forecasting, queues and reservations; they contain no load-capacity total and do not renew the separately dated generation/storage dataset.

Named-campus record links JSON and CSV keep one reviewed contextual cross-reference separate from the original records. It does not merge facilities, phases, metrics or source dates; unique physical facilities remain unknown.

Research and claims in context has separate research JSON/CSV and claim-review JSON/CSV. The register contains 7 source records and 4 scoped claim reviews. Guidance, modelled methods and report scenarios retain their distinct evidence labels. Publication dates, page-update dates and evidence checks remain separate; unknown publication dates and reuse terms stay null in JSON and empty in CSV with source-specific notices. These downloads do not renew checks or grant source rights. Retain attribution, derived-work notices and reuse exclusions. No IEA Annex/data-product material, graphics or proprietary report inputs are imported.

Inspect the first unscored usefulness samples for Trust, Evidence depth and Freshness, with matching JSON and CSV. Counts retain their named sample and original clocks; all numeric ratings and the other seven dimensions remain unassessed.

Download the initial connected-evidence cohort as JSON: scoped entities, attributed observations, relationships, explicit gaps and the briefing. Its separate schema retains per-observation dates and source-use conditions. It is not a fresh status check, physical-facility census or complete change history. No CSV is provided for this linked model.

Found a wrong value? Tell us. See sources and licences and method for how records are collected.