State of AI data licensing 2027

1. Deals with a disclosed value

   11 of 25

   11 of the 25 AI data licensing deals we track have a disclosed value, and the largest is about $250 million over five years in cash and credits, paid by OpenAI to News Corp (Engadget, citing The Wall Street Journal, May 2024).

   Data: AI data deals tracker

2. Most active paying party

   10 of 25

   OpenAI is the paying party in 10 of the 25 tracked deals, more than any other company, followed by Google and Microsoft with 3 each.

   Data: AI data deals tracker

3. Published payout range by industry

   $20,000 to $1.5 million

   Published payouts for company records run from $20,000 to $1.5 million per company across 11 of the 30 industries we track, according to Mercor's published listings, as reported by ddiy.co.

   Data: AI data price index

4. Buyers that publish a payout

   11 of 61

   Only 11 of the 61 data buyers, marketplaces and brokers we profile publish a dollar figure for payouts or prices, and 3 of the 14 direct buyers do.

   Data: Data buyers directory

5. AI training data lawsuits pending

   25 of 37

   Of the 37 AI training data lawsuits and rulings we track, 25 are pending, 7 have a ruling, 2 are on appeal, 2 have settled and 1 was dismissed, as of November 2026.

   Data: AI training data lawsuits tracker

6. Largest disclosed settlement

   $1.5 billion

   The largest disclosed settlement is $1.5 billion in Bartz v. Anthropic, which received final court approval on July 20, 2026.

   Data: AI training data lawsuits tracker

7. Platforms whose API terms restrict AI training

   5 of 12

   Of the 12 SaaS platforms we reviewed, 5 restrict AI training on data pulled through their APIs (Slack, Google Workspace, Zendesk, HubSpot and Zoom), and we found no such restriction in the terms of 3.

   Data: Platform data rights

8. US state comprehensive privacy laws in force

   19 of 23

   19 of the 23 US states that have passed comprehensive privacy laws have them in force as of November 2026, 3 of them since January 1, 2026, and 3 more take effect in 2027.

   Data: Data privacy laws for AI training by country

Every number is computed from our datasets as of November 2026 and links back to its source. Published figures are ceilings, not quotes.

Who is buying data and how?

AI training data is bought by 61 companies we profile in 6 categories, from AI labs that license by direct deal to training-data companies and marketplaces that publish their terms. 14 are direct buyers such as OpenAI, Google and Microsoft, and 3 of them publish a dollar figure for payouts or prices.

Reported deals show who pays most often. OpenAI is the paying party in 10 of the 25 deals in our AI data deals tracker, followed by Google and Microsoft with 3 each.

Training-data companies are the largest group, with 21 profiles, and 5 of them publish a dollar figure. Across all 61 profiles, 6 state that sellers pay no fee. Every profile in our data buyers directory is written from the company's own pages and dated.

For a company with records to license, the practical routes are a direct deal with a large buyer, a training-data company that publishes a payout, a marketplace listing or a broker. Our guide to AI data marketplaces, brokers and direct deals compares them.

What do buyers pay?

Buyers pay $20,000 to $1.5 million per company for operational records, according to Mercor's published listings, as reported by ddiy.co, which cover 11 of the 30 industries we track. These are ceilings for the companies that fit best, not quotes.

12 of the 19 training data types we cover have at least one published price, and the pricing unit depends on the type. Video is priced per minute: average about $2.62 per minute, up to about $7.25, paid by Adobe, reported by Bloomberg. micro1 publishes tiers from $100,000 to $500,000 for a qualified enterprise data partnership to $3 million and above for highly unique data.

Estimates of the whole market vary by firm. Grand View Research, in a report published June 2026, estimates the AI training dataset market at $3.2 billion in 2025. Mordor Intelligence, in a report updated May 26, 2026, estimates $8.74 billion in 2025. The firms define the market differently, as our guide to AI training data market size explains.

Every published figure, with its source, is in the AI data price index. A free data valuation gives a company a realistic range for its own records.

19 of the 30 industries we track have no published range.

Which deals set the market?

A few large publisher and platform licenses set the market. The largest reported value is about $250 million over five years in cash and credits, paid by OpenAI to News Corp, and the 3 deals reported with an annual US dollar figure add up to at least $96 million a year.

14 of the 25 deals we track were reported without terms, so the table covers only 11. The tracker holds 3 deals from 2023, 12 deals from 2024, 7 deals from 2025 and 3 deals from 2026.

The 3 deals reported in 2026 are Anthropic with The Associated Press (terms undisclosed), an unnamed buyer with a shut-down crypto startup (about $25,000 for the codebase, with about $15,000 likely for workspace data) and Google with Spirit Airlines ($10 million winning auction bid).

Values are as reported by the cited source. Some were reported by one news outlet and never confirmed by either party, and the table says so where the source does.

What did the courts decide in 2025 and 2026?

Courts in 2025 and 2026 separated how training data was obtained from how it was used. 11 of the 37 cases we track reached a ruling, settlement or dismissal dated 2025 or 2026, while 25 remain pending as of November 2026.

In Bartz v. Anthropic, Judge William Alsup ruled on June 23, 2025 that training Claude on books Anthropic had bought and scanned was fair use, but that downloading and keeping pirated copies was not. Anthropic then agreed to pay $1.5 billion, about $3,000 per book, and the court gave final approval on July 20, 2026.

In Kadrey v. Meta, Judge Vince Chhabria granted Meta partial summary judgment on June 25, 2025, finding that training Llama on 13 authors' books was fair use on the evidence those authors presented. He stressed that the authors lost because they had not shown harm to the market for their books.

In Thomson Reuters v. Ross Intelligence, the Third Circuit ruled on September 29, 2026 that Ross's copying of Westlaw headnotes to train a competing legal search tool was not fair use. It is the first US appeals court ruling on fair use in AI training.

Every case, with its status and sources, is in our AI training data lawsuits tracker. This section describes what courts decided and is not legal advice.

What do platform terms say about training on your data?

Platform terms leave ownership of stored data with the customer and put AI limits in their API and developer terms. Of the 12 platforms we reviewed, 5 restrict AI training on data pulled through their APIs, 4 have terms we rate as unclear and 3 have no restriction we found.

Native admin exports carry fewer conditions. 22 of the 27 non-API export routes we reviewed have no AI-training clause in the terms. How you extract the data can change which terms apply, so choose the route with counsel before anyone writes code.

Owning the data against the vendor does not settle whether a license is lawful. Privacy notices, customer contracts and third-party content usually decide it. Each platform page in our platform data rights section quotes the clause and links to it.

What changed in US privacy law for AI training?

US privacy law for AI training changed mostly at state level. Indiana, Kentucky and Rhode Island brought comprehensive privacy laws into force on January 1, 2026, taking the total in force to 19, and California began requiring risk assessments before a business sells personal information and made AI developers disclose licensed training data.

23 US states have passed comprehensive consumer privacy laws, according to our guide to data privacy laws for AI training, which leaves out Florida's narrower law. Under the CCPA, a transfer of personal information for monetary or other valuable consideration is a sale, so a paid license of identifiable records triggers notice and opt-out duties.

There is still no comprehensive federal privacy law. The federal changes that matter to training data came through the FTC and the Justice Department, listed below with the state changes. Our CCPA and state privacy law guide covers what each rule means for a license.

What should you watch in 2027?

In 2027, watch the US state calendar first: 3 state privacy laws, Colorado's SB26-189 and California's automated decisionmaking rules start on dates set in statute or regulation. Court rulings in pending US cases have no set date, so we list them with their last dated step.

Every item below has a date from a cited source. We do not predict outcomes or rulings.

EU dates are listed separately because they apply to models and AI systems offered in the European Union, not to US law. A US company that sells or licenses data to a model provider serving EU users may be asked to support that provider's training content summary.