WhatsApp and Telegram +44 7563 115980
[email protected]
What listed companies have actually disclosed about licensing data for AI, taken from their own filings, and the factors that decide what a dataset is worth to the companies that train AI models.
Since 2013 we have helped firms with due diligence and data services, in more than 60 countries.
The short answer
There is no public price list for AI training data. The most reliable evidence is what listed companies disclose in their own filings. Reddit disclosed data licensing arrangements with an aggregate contract value of $203.0 million over two to three years. Wiley reported $49 million of AI revenue in its 2026 financial year. Most other agreements do not disclose financial terms. What a dataset is worth depends on how scarce it is, how clear the rights are, how clean and well described it is, whether it keeps growing, and whether the buyer gets it exclusively.
On this page
The table lists deals and figures that the companies themselves reported, in filings with the US Securities and Exchange Commission, or that the US Copyright Office recorded in its 2025 report on AI training. Where a company did not publish a figure, none is given.
| Company | What was disclosed | Source |
|---|---|---|
| Data licensing arrangements entered into in January 2024 with an aggregate contract value of $203.0 million and terms of two to three years, delivered through continuous access to its data and quarterly transfers. At least $66.4 million was expected to be recognised as revenue in 2024. | Reddit prospectus, March 2024 | |
| John Wiley and Sons | $49 million of AI revenue in the financial year to 30 April 2026, up 23%, with lifetime AI revenue above $110 million. | Wiley results, June 2026 |
| Shutterstock | Revenue from data, distribution and services of $137.3 million in 2023, of which growth in the data offering accounted for $84.9 million of the increase. The same line reached $175.3 million in 2024 and $203.3 million in 2025. | Shutterstock annual reports |
| Dotdash Meredith (IAC) | A 2024 licensing agreement with OpenAI under which attributed content appears in ChatGPT responses and is used to improve OpenAI's models, with a partnership on advertising products. No figure was given in the results release. | IAC results, May 2024 |
| News Corp | A multi-year global partnership with OpenAI, described as expected to be lucrative. No figure was given in the results release. | News Corp results, August 2024 |
| Associated Press | A July 2023 licensing deal allowing OpenAI to train on part of the AP text archive, as recorded by the US Copyright Office. No figure recorded. | US Copyright Office, May 2025 |
The Copyright Office also recorded licensing activity for images, film and television, academic publishing and news, and the growth of collective licensing, where one organisation licenses works on behalf of many rights holders.
Headline numbers are useful as evidence that the market pays, but they are not prices per record. Each one bundles different things.
The sound way to use these figures is as proof of demand and as a guide to which factors buyers pay for, not as a rate card.
| Factor | Why it raises value | Evidence |
|---|---|---|
| Scarcity | Data that exists only inside one company cannot be gathered from the public web by any competitor | Nature, 2024: data about genuine human interactions will become increasingly valuable |
| Rights certainty | One owner who can grant all the rights in one agreement removes legal risk for the buyer | US Copyright Office: direct licensing is most common with catalogues of high-quality, easily identified content, where centralised rights make licensing easier |
| Quality and curation | Accurate, complete and well-organised data trains better models and needs less cleaning | US Copyright Office: curated content likely has higher training value |
| Description and labels | Descriptions, categories and ratings attached to records save the buyer the cost of labelling | Shutterstock: 2025 growth driven mainly by the sale and delivery of metadata |
| Freshness | Data that keeps growing supports ongoing training and justifies multi-year terms | Reddit: continuous access and quarterly transfers over two to three years |
| Privacy cleanliness | Data that is anonymous can be used without data protection obligations attaching to the buyer | ICO and GDPR Recital 26: anonymous information is outside data protection law |
| Exclusivity | Data reserved for one buyer gives that buyer an edge its competitors cannot buy | Standard licensing practice; priced as a premium in the licence |
| Documentation | A data sheet, rights record and exclusion record let a buyer approve the purchase and report it | EU AI Act template, Section 2.2.1: licensed datasets must be reported |
We assess a dataset against the factors buyers price, clear it for licensing and prepare it for competing offers.
Speak to an expertAI training dataMost of these are fixed before a dataset goes to market. Our guide on data provenance and rights clearance explains how the rights side is settled.
Pricing structures vary by deal. The common building blocks are set out below, and many licences combine several of them.
Owners preparing a first sale can start with our guide on how to sell data to AI companies. Buyers sourcing data can see how we supply it on our AI training data page.
There is no public price list. Value depends on scarcity, rights certainty, quality, labels, freshness, privacy cleanliness, exclusivity and documentation. The disclosed deals show the market pays substantial sums, but each figure bundles different terms.
Reddit's March 2024 prospectus disclosed data licensing arrangements entered into in January 2024 with an aggregate contract value of $203.0 million and terms of two to three years, with at least $66.4 million expected to be recognised as revenue in 2024.
Most agreements do not disclose financial terms. IAC and News Corp announced licensing agreements with OpenAI in their results releases without stating figures.
Yes. Exclusivity keeps the data from competing buyers for a field of use or a period, and it is priced as a premium. A non-exclusive licence earns less per deal but lets the owner license the same data again.
Done well, it raises value, because anonymous data can be used without privacy obligations. Value is lost only when de-identification strips out the context a model needs, which careful methods avoid.
Yes. Buyers pay for scarcity and quality, not company size. A smaller company with unique, well-documented records and clear rights can be a strong supplier, alone or alongside other owners.
Under a non-exclusive licence the same data can be licensed to several buyers. Under an exclusive licence it is reserved for one buyer within the agreed field and period.
Data that is scarce, created by people doing real work, accurate, labelled, refreshed over time and cleared for use. Buyers also value full documentation, because they must report licensed data under the EU AI Act.
Each point of law and each reported figure on this page was checked against these sources on 27 October 2026. This page is general information, not legal advice.
Tell us about the data you hold, its volume and how often it grows.