About

Data on AI is published by many organisations in different formats. This site collects those figures into a single index and records the origin of each one. Sources include Epoch AI, Eurostat, OpenAlex, Microsoft and government publishers.

Scope

The site covers benchmark results, training compute and disclosed cost, disclosed energy use and emissions, stated token volumes and user counts, adoption by population and by enterprise, published frontier safety frameworks, research volume, reported incidents, and the law and policy that applies to AI in eleven jurisdictions.

Each term used on a data page is defined in the glossary as this site applies it.

Eight datasets are compiled here from primary documents: the Governance Readiness Index, the AI Law and Policy Index, the Frontier Safety Framework Index, the Compute Threshold Index, the AI Usage Disclosure Index, the Environmental Disclosure Index, the AI Datacentre Index and the Release Safety Disclosure Index. The methodology sets out how each is compiled.

Sourcing

Each source is declared once with its licence, attribution, update cadence and stated limitations. Each processed record carries a source identifier. Licence and attribution are resolved from that registry when a page is built. The pipeline does not write data whose licence forbids republication, and the site does not render it. Removing a source from the registry removes everything derived from it.

Every figure carries its source, licence and retrieval date. The sources page lists each dataset in use, its licence, its use here and its stated limitations.

Limits

• The site publishes no forecasts of its own. Straight-line extrapolations appear on the progress page only, and each is labelled with its fit, its window and the direction of its bias.
• Governance readiness records the machinery a jurisdiction has. It is not an assessment of quality or enforcement, and a jurisdiction may score low by choice.
• Nothing here is legal advice. See the terms.
• There are no accounts, comments, newsletter, advertising or revenue model, and no tracking beyond aggregate page counts.

Construction

A Python pipeline fetches and processes each source. A static site generator renders the result. Charts and maps are drawn to SVG when the site is built. The build ships four scripts: one applies the light or dark theme chosen in the header, one zooms and pans the datacentre map by mouse, one filters the search page as a query is typed, and one lets the period control on the home page map be dragged. The site works without any of them, in the dark theme; the search page is then the full list, and the period control works by click. The host injects one analytics script, described in the privacy policy. Data is refreshed daily. The date on a page is the last time its source was checked, or, for an index compiled here, the last time it was reviewed.

AI was used to help build this website.