Datasets
The data behind this site, to download.
Every table on this site is one of these datasets. Each row names the publication that stated it and the day it was read. Free to reuse under CC BY 4.0 with a link back.
Every price observation for AI training data: the figure as published, the side of the market, the figure per normalised unit (our arithmetic on the published figure), dates, source and the source's own words.
1022 rows, reviewed 17 Oct 2026
Median, middle half and range per normalised unit for each kind of data and side of the market: our arithmetic on the price observations, with the ids behind each.
115 rows, reviewed 17 Oct 2026
How AI training data prices are negotiated
Negotiation levers for holders and buyers of AI training data, each a quantified fact from the public record with sources, and our reading of what it means in a negotiation (not legal advice).
69 rows, reviewed 17 Oct 2026
Where AI training data comes from
Quantified facts about the sources of AI training data, by kind of data: who holds it, how much exists and the channel it reaches buyers through.
122 rows, reviewed 16 Oct 2026
Disclosed licences of data to AI companies since July 2023: licensor, licensee, content, use, term, value where disclosed, source.
111 rows, reviewed 17 Oct 2026
Organisations with a public way to sell them data: what each buys, its published prices, eligibility, payment terms and introduction fees.
37 rows, reviewed 16 Oct 2026
Companies that sell training data, RL environments, evaluations, expert work or annotation to AI developers: what each says it sells, the areas it works in and where its data comes from, each with the vendor's own words and the page they are on.
150 rows, reviewed 15 Oct 2026
Lawsuits and rulings over AI training data
Lawsuits, rulings, settlements and regulatory actions over data used to train AI, with outcome and source.
83 rows, reviewed 17 Oct 2026
Funding rounds, acquisitions and valuations of companies that buy, make or broker AI training data.
111 rows, reviewed 17 Oct 2026
Revenue, run rates and ARR of data and expert-work vendors to AI labs, as published.
227 rows, reviewed 17 Oct 2026
Dated records of buyer programs, transactions, price ranges, referral terms and broker economics in company-data sales.
110 rows, reviewed 17 Oct 2026
What software terms allow for AI training data
For each business system on file: whether the holder can export its data, what the vendor's API terms say about training AI on data pulled through them, and our reading of the rights risk of the content (not legal advice), with quotes.
18 rows, reviewed 11 Oct 2026
Rights checklist for selling data for AI training
The questions a buyer's counsel asks before licensing data: good answers, red flags, the kinds of data each applies to and, on our reading (not legal advice), whether it blocks a sale, with sources.
27 rows, reviewed 11 Oct 2026
Terms of an AI training data licence
The terms a data licence for AI training turns on: what buyers ask, where a holder may want to stand (our reading, not legal advice), the public evidence and its gaps.
15 rows, reviewed 11 Oct 2026
Organisations in the AI data market
Every organisation the datasets refer to, by id, with kind and description.
452 rows, reviewed 17 Oct 2026
How the data is collected
- A row is a figure or an event that someone else published: a court record, a filing, a company's own page, or named press.
- Each row keeps the source link, the publication date and a verbatim quote where one exists, so it can be checked.
- Press figures from unnamed sources are marked as reported, and a vendor's price guide as a vendor estimate. Nothing is our own estimate.
- Dates are the date of the sale, ruling or publication, never the day we read it.
- Rows are re-read against their sources and the review date on each dataset says when.
Reusing it
The compiled datasets are licensed under CC BY 4.0: use them for anything, including commercially, with credit to this site and a link to the page the data comes from. Quotes and figures remain the words and numbers of their original publishers.
To cite a fixed version, use a dated snapshot of every table kept in a research repository: Figshare (DOI 10.6084/m9.figshare.34415514), taken 13 Oct 2026, Zenodo (DOI 10.5281/zenodo.23346240), taken 13 Oct 2026, Data Marketplace, taken 16 Oct 2026, OpenML, taken 16 Oct 2026 and Internet Archive, taken 12 Oct 2026.
To ask questions of the data from an AI assistant instead of downloading it, connect Claude, ChatGPT or any MCP client to the MCP server.
Found an error, or a deal or price we are missing? Describe it through the request form and we will check it against its source.