How much is company data worth to AI labs?
There is no price list. Reported deals give a range: Forbes reported shut-down startups' Slack, email and Jira archives selling for $10,000 to $100,000 each through one wind-down firm, and Google bid $10 million for Spirit Airlines' internal data. Price follows depth of history, linked systems, rarity, clean rights and exclusivity, not raw volume.
What reported deals show
| Deal | What was licensed | Reported value | Reported by |
|---|---|---|---|
| Shut-down startups, through one wind-down firm | Slack, email and Jira archives | $10,000 to $100,000 per company, about 100 deals in a year | Forbes, via Gizmodo (April 2026) |
| Spirit Airlines, in bankruptcy | Internal operational data | A $10 million bid from Google | Frankfurt Kurnit Klein & Selz (October 2026) |
| Reddit and Google | Reddit's public posts | About $60 million a year | Reuters, via Engadget (February 2024) |
| Dotdash Meredith and OpenAI | Magazine and web content | At least $16 million a year | Adweek, from company filings |
The first two rows are the closest comparisons for an ordinary company. The last two are deals for huge public content libraries, which buyers price differently.
What moves the price
- Depth: years of continuous history on the same tools.
- Linkage: work that can be followed across email, chat, tickets and code.
- Reasoning: decisions discussed in writing, including the options turned down.
- Rarity: a field where public data is thin and buyers are looking right now.
- Clean rights: data that is clearly yours, with contracts and notices that allow licensing.
- Preparation: de-identification a buyer can trust, with the method documented.
- Exclusivity: a promise not to license the same data to others usually costs the buyer more.
Size helps only through these. A large archive of short, disconnected messages is worth less than a smaller one where each decision can be traced. Our guide on what data AI companies buy explains why.
How deals are usually paid
Publishers' deals with AI companies, the best-documented part of the market, usually combine a fixed upfront payment with a variable payout based on usage, according to a 2025 analysis of AI content licensing. Most of those deals are not exclusive. Smaller company deals reported so far have been one-off payments.
Costs come off the top: legal review, de-identification and any broker or advisory fee. Those costs are largely fixed, which is why very small datasets often don't justify a deal. Our guide to data licensing agreement terms covers exclusivity, length and payment in more detail.
Treat any large figure offered before someone has seen a de-identified sample with care. A serious buyer prices data after reviewing it.
Questions
How much is my company's data worth?
It depends on depth, linked systems, rarity, clean rights and exclusivity. Reported deals for ordinary companies' archives range from about $10,000 to $100,000, with rare larger ones.
Why did Reddit get far more than a startup?
Reddit licensed a huge public content library to a major buyer. Company archives are smaller and private, and are priced on their depth and reasoning instead.
Is data from a live company worth more than a closed one?
It can be, because a live company can keep adding data and answer buyers' questions. Exclusivity and terms then matter more.
Are data licensing deals paid upfront?
Many small deals are one-off payments. Larger publisher deals often combine a fixed payment with a usage-based payout.
Should I accept a price before a buyer sees a sample?
Be careful. Serious buyers price after reviewing a de-identified sample, and early headline numbers often change.
Sources
- Failed companies are selling old Slack chats and email archives to train AI (17 April 2026) Gizmodo, reporting Forbes
- Selling company data for AI training: what to know before you say yes (1 October 2026) Frankfurt Kurnit Klein & Selz, law firm
- Reddit is licensing its content to Google to help train its AI models (February 2024) Engadget
- OpenAI's payment to Dotdash Meredith (2024) Adweek
- The 7 deal points of AI content licensing (9 May 2025) Media & the Machine