Data and API

Everything on the site is available as static files: no keys, no rate limits beyond ordinary politeness, and nothing rendered by a server. The files are rebuilt nightly and sit next to the pages they describe. Cite and link freely with attribution; the downloadable catalog is for personal and research use, as set out in the terms.

Endpoints

Path	What it returns	Example
/api/parts/{id}.json	One part: name, summary, description, layer, children, bottleneck status and every company placed on it with role, confidence tag, share estimate and source.	/api/parts/5.05.json
/api/companies/{id}.json	One company: ownership, headquarters, listings, subsidiaries and every part it is placed on, each with its confidence tag.	/api/companies/vertiv.json
/api/supercomputers.json	Index of the supercomputer breakdowns, each with its own JSON endpoint.	/api/supercomputers.json
/api/supercomputers/{id}.json	One machine: specs and every line of the breakdown with the companies and part ids it names.	/api/supercomputers/el-capitan.json
/graph.json	The whole part-to-company graph in one compact file: parts, companies and edges as index pairs with a confidence initial. The explorer at /graph reads it.	/graph.json
/stress.json	Every part with its suppliers and confidence tags, every company with its parts and disclosed customers, and the machines naming each. The stress-test tool reads it.	/stress.json
/changelog.xml	The weekly data changelog as RSS.	/changelog.xml
/{page}.md	A Markdown version of every part, company, supercomputer and teardown page, for tools and language models.	/companies/vertiv.md
/supercomputers/lines.csv	Every supercomputer breakdown line as CSV.	/supercomputers/lines.csv
/feed.json · /rss.xml	The blog as a JSON Feed and an RSS feed.	/rss.xml
/companies/{id}/feed.xml · /parts/{layer}/{part}/feed.xml	News about one company or one part as RSS.	/companies/vertiv/feed.xml
/llms.txt · /llms-full.txt	A guide to the site for language models, and the full text of the catalog.	/llms.txt
/data/catalog.sqlite	The whole catalog as one SQLite file, rebuilt nightly. Tables: layers, parts, companies, part_company, listings, quotes, teardown_lines, scale_tiers.	/data/catalog.sqlite

Reading the confidence field

Every company-to-part link carries a tag: disclosed when the company says so itself, reported when a credible outlet or filing says so, and inferred when the placement follows from the company's products but nobody has said it. Treat inferred links as leads, not facts. Prices are a nightly snapshot with an as_of date, never real-time, and are never converted between currencies. The method is in sources & method.

Query it yourself

Download catalog.sqlite and run SQL. For example, the companies disclosed or reported on coolant distribution units:

SELECT c.name, pc.role, pc.confidence
FROM part_company pc JOIN companies c ON c.id = pc.company_id
WHERE pc.part_id = '5.05' AND pc.confidence IN ('disclosed','reported')
ORDER BY c.name;

In JavaScript, fetch('/api/parts/5.05.json').then(r => r.json()) returns the same part with its supplier list. Part ids are stable taxonomy numbers (5.05, children 6.05a); company ids are stable slugs.

For language models and crawlers

The Markdown and JSON mirrors are marked noindex so search engines keep the HTML pages as the canonical results, but nothing blocks agents from fetching them. llms.txt describes what is where, and robots.txt allows the major AI crawlers.