Three kinds of organizations buy or license company data for AI training: large AI developers, companies building AI for a specific industry, and data programs that license records from many businesses and supply AI developers. Public deals show the range. A source cited by Reuters put Google's AI licensing deal with Reddit at about $60 million a year, and in 2026 Google won a $10 million bankruptcy auction for Spirit Airlines' internal business records. The largest public deals have been with very large content owners, so an established business with years of operational records is more likely to start with a data program than with an AI lab directly.
- Buyers want records that show how work was done: the problem, the investigation, the decision and the outcome.
- Public deals so far involve large content libraries; terms for smaller companies are rarely published.
- Who buys matters less than what you can license: rights, history and de-identification decide whether a deal is possible.
Who buys company data for AI training?
Large AI developers license data to train and improve general models. Their publicly reported deals have mostly been with platforms and publishers that hold very large libraries of text.
Companies building AI for one industry need examples of real work in that industry: support conversations, service records, procedures and project histories. These records teach a model how a job is actually done.
Data programs license records from many businesses, prepare them and supply AI developers. For a company without a very large content library, a data program is often the more practical starting point. Origin Data Partners is an independent referral business that helps owners prepare for that conversation; it is not itself a buyer.
What public deals show
Most AI data licenses are private. The deals below were reported publicly; amounts that the companies did not confirm are marked as reported.
| Buyer | Seller | What was licensed | Reported terms |
|---|---|---|---|
| Reddit (February 2024) | User posts, available for training AI models | About $60 million a year, per one source cited by Reuters; neither company confirmed the amount | |
| OpenAI | News Corp (May 2024) | Current and archived content from titles including The Wall Street Journal and the New York Post | Multiyear; potentially more than $250 million over five years, per The Wall Street Journal as reported by Variety |
| OpenAI | Stack Overflow (May 2024) | Developer questions and answers through Stack Overflow's API | Value not disclosed; Stack Overflow uses OpenAI models in its own products (press release) |
| Unnamed; Bloomberg reported Microsoft | HarperCollins (November 2024) | Select nonfiction backlist titles | Authors opt in; a reported $5,000 per title split between author and publisher (Quill and Quire) |
| Spirit Airlines, in bankruptcy (August 2026) | About 100 million emails, 500 million Teams chats and other internal business records; passenger profiles excluded | $10 million winning auction bid (Axios); court approval still pending on October 8, 2026 |
What buyers look for in operational records
The Spirit sale is the clearest public example of a company's internal records being bought for AI. The buyer wanted the working history of a business: messages, documents and records of how decisions were made. Customer profiles were left out, and personal information was to be removed or de-identified before delivery.
For an established business, the useful material is usually linked records: a support ticket with its investigation and fix, a work order with its notes and outcome, a procedure with the revisions that followed real problems. See how buyers think about value and what makes records worth discussing.
- History: several years of records, with dates.
- Links: the problem, the work and the result connected in one trail.
- Rights: records the company created and can authorize, separate from customer or third-party material.
- Handling: personal information that can be removed or de-identified.
How an established business reaches a buyer
Start by describing your records at company level: what kinds exist, how many years you keep and who owns them. A data program decides whether that profile fits what its AI customers need, and proposes terms if it does.
Check the legal questions before any records move; see is it legal to sell company data for AI training. Origin helps organize a Company Licensing Brief and, with your approval, introduces you to a program that may fit. Talk to Origin. No files are needed to start.
Common questions
Do AI companies buy data from small businesses?
The largest public deals have been with big platforms and publishers. A smaller or mid-sized business more often works with a data program that licenses records from many companies. Whether a program accepts a company depends on its records, history and rights.
Do buyers want customer data?
Public deals center on content and work records. In the Spirit Airlines sale, passenger profiles and loyalty records were excluded and personal information was to be removed or de-identified before delivery.
How much do buyers pay for company data?
Published figures exist only for very large libraries, and some were never confirmed by the companies. There is no reliable public price per record for an established business. See how buyers think about value before discussing a number.
Is Origin a buyer?
No. Origin Data Partners is an independent referral business. It helps owners describe their records and, with their approval, introduces them to a program that may fit.
Sources and further reading
- Reuters via Yahoo: Reddit in AI content licensing deal with Google (February 22, 2024)
- Variety via Yahoo: News Corp inks OpenAI licensing deal (May 22, 2024)
- Stack Overflow: partnership with OpenAI (May 6, 2024)
- Quill and Quire: HarperCollins signs AI training deal (November 20, 2024)
- Axios: Google buys Spirit Airlines emails, chats, documents out of bankruptcy (August 17, 2026)
- Bloomberg Law: Spirit data sale to Google passes muster, privacy ombudsman says (October 5, 2026)