Every deal below has a named source and a date. Most are platform and publisher deals, which are very different from a company licensing its internal records. The second half of the page explains why.
Last checked: 7 October 2026. Figures are as reported; we have not seen the contracts.
Sorted by type. "Reported" means the figure comes from press coverage or an executive interview, not a published contract.
| Deal | What is licensed | Reported value | Source and date | Status note |
|---|---|---|---|---|
| Google and Reddit Platform | Reddit posts and comments for AI use | about $60M a year | Signed February 2024, value widely reported. Renewal talks: Wall Street Journal via CNBC, July 22, 2026 | Reported in renewal negotiation; Reddit had discussed ending Google's AI access |
| OpenAI and News Corp Publisher | Content from News Corp publications | more than $250M over five years | Announced May 2024; value reported by The Wall Street Journal, via Variety, May 2024 | Multiyear agreement |
| Shutterstock AI licensing Stock media | Images, video and metadata licensed to AI developers | about $104M (2023) | CEO interview with Bloomberg, reported by PetaPixel, June 2024 | Revenue across many buyers, not one deal |
| Shutterstock 2024 Stock media | Same business unit | about $138M (2024) | PetaPixel correction, June 7, 2024 | A projection for the whole unit that includes AI licensing |
| Google and Spirit Airlines Company archive | Internal emails, Teams messages, spreadsheets, operations files | $10M agreed | ABC News and TIME, August 2026; court status via Reuters, Aug 19, 2026 | Bankruptcy sale; final approval not confirmed as of 7 Oct 2026 |
| micro1 and Spirit Airlines Company archive | The same Spirit dataset | $12.5M offer | Competing bid notice filed Sept 3, 2026, per Stretto report, Sept 13, 2026 | An offer, not a closed deal |
| Closure market (Troveo) Shut-down startups | Slack, email and code archives of closed startups | $10K to $100K per archive; ~$5K per repo | Figures cited by Troveo in 2026 coverage (Forbes, April 16, 2026; Fast Company; Gizmodo) | Typical ranges cited by one participant |
How this tracker is kept. A deal is listed only when a named outlet, a company executive or a court filing reported a value, and we can give the source and date. We do not estimate undisclosed deals, we do not convert reported figures into per-unit prices, and we do not list rumors from social media. Offers and pending sales are labeled as such. When a status changes, such as a court ruling on the Spirit sale or a reported outcome of the Reddit renewal talks, we update the row and the "last checked" date rather than adding a new one. If a figure is later corrected by its source, as PetaPixel did for Shutterstock's 2024 number, we use the corrected wording.
Market context, both estimates: more than 50 companies sell data and RL environments to AI labs, with about $8.5B in revenue (Deedy Das market map, July 2026); labs are on a path to more than $100B a year of data spend by 2030 (Will Depue, July 2026). Our full case notes: Spirit Airlines and shut-down startups.
Patterns visible across the rows, before any comparison with your company.
Only Shutterstock's figure came from its own executive. The Reddit and News Corp numbers come from press reports about private contracts. Treat them as orders of magnitude, not prices.
Reddit's deal is reported as a yearly figure, and the July 2026 reporting shows that such licenses come up for renewal and get renegotiated. A recurring license is a different product from a one-off sale.
Spirit's sale is public because bankruptcy sales go through court. The closure-market figures are public because a participant cited them to reporters. Running-company deals are private contracts.
In the one auction on the list, the reported price moved from a $5 million opening bid to $10 million, and a later $12.5 million offer followed. More bidders gave the seller more leverage on price and terms.
What the table does not show is just as telling. There are no rows for ordinary companies of 20 to 1,000 people licensing their internal records through buyer programs, even though those programs publish six- and seven-figure ranges. Those deals are private contracts, so they rarely surface. When you compare your situation with this tracker, start with the published program ranges and treat the headline deals as background.
The same word, "licensing," covers two very different transactions.
The useful public reference for a company of 20 to 1,000 people is not the Reddit number. It is what buyer programs publish on their own pages, checked 7 October 2026: micro1 lists "$100k+ qualified," "$500k+ large-scale" and "$1M+ highly unique"; Mode lists "$100K-$5M"; Grepped lists "$20K-$5M." Those are ranges with "up to" logic, not averages, and nobody can price your data without seeing a sample. Our guide to what AI companies pay breaks down what moves a number inside those ranges.
Reddit's renewal talks show that a license has an end date and a renewal moment. Read the term, the renewal and what happens to the copy afterward. Our AI data licensing agreement outline lists the clauses.
A licensor that sells to several buyers keeps its options open. If a buyer wants exclusivity, it should be time-limited and priced. See exclusivity and resale rights.
Practitioners say raw data is the cheapest tier and evaluations built on it are worth roughly ten times more. See raw data vs evaluations vs environments.
The same facts as the tracker, arranged by date, so you can see how the market moved from platforms to company records.
| Date | Event | Source |
|---|---|---|
| February 2024 | Google and Reddit sign an AI content deal, widely reported at about $60 million a year | Widely reported; recapped by CNBC, July 22, 2026 |
| May 2024 | News Corp and OpenAI announce a multiyear licensing deal, reported at more than $250 million over five years | The Wall Street Journal, via Variety |
| June 2024 | Shutterstock's CEO says AI licensing brought in $104 million in 2023; about $138 million projected for 2024 for the unit including AI licensing | Bloomberg interview, reported by PetaPixel (correction June 7, 2024) |
| April 16, 2026 | Forbes reports closed startups selling old Slack and email archives; Troveo's prices circulate | Forbes; also Fast Company, Gizmodo |
| July 2026 | Market map counts 50+ companies selling data and RL environments to labs, about $8.5 billion in revenue | Deedy Das (estimate) |
| July 2026 | Estimate that labs are on a path to more than $100 billion a year of data spend by 2030 | Will Depue (estimate) |
| July 22, 2026 | Reddit and Google reported to be negotiating a renewal; Reddit had discussed ending Google's AI access | The Wall Street Journal, via CNBC |
| August 14, 2026 | Google named winning bidder for Spirit Airlines' de-identified enterprise data at $10 million | Stretto report, Sept 13, 2026; ABC News, TIME |
| August 19, 2026 | Spirit approval hearing delayed after a union privacy objection | Reuters |
| September 3, 2026 | micro1 files a $12.5 million competing bid notice for the Spirit data | Stretto report, Sept 13, 2026 |
| October 7, 2026 | Final outcome of the Spirit sale not confirmed when we checked | Our check |
Two shifts stand out. First, the subject of the deals moved from public content (forum posts, news articles, stock media) to private records of how work gets done (Slack, email, operations files). Second, the scrutiny grew. The 2024 deals were negotiated quietly between large companies with legal teams. The 2026 company-record sales drew employee objections, privacy reviews and press attention.
For a company considering a deal now, that history cuts both ways. Demand for internal work records is visible and public. So is the expectation that sellers handle employees, clients and privacy carefully. Buyers know it too, which is why the programs publish privacy statements on their own pages.
Use these before you quote a deal to your board, or let a buyer quote one to you.
A company's own executive (Shutterstock's CEO), a newspaper citing people familiar with the deal (the Reddit and News Corp figures), or a court filing (Spirit). Each carries a different level of certainty.
About $60 million a year (Reddit) and more than $250 million over five years (News Corp) are different shapes. Convert to the same basis before comparing.
Shutterstock's $104 million is revenue across its AI licensing business, not a single contract. Aggregates look larger than any one deal.
Shutterstock's 2024 figure was a projection for a whole business unit, as PetaPixel's correction made clear. Projections are plans, not payments.
micro1's $12.5 million for the Spirit data was an offer filed with the court, not a closed purchase. Offers tell you about demand; only closed deals tell you about price.
Google's $10 million Spirit bid needed court approval, which we could not confirm as of 7 October 2026. Quote pending deals as pending.
Licenses end. The July 2026 reporting on Reddit and Google shows that even a well-known deal can be renegotiated or ended at renewal.
Is the asset in the headline anything like yours? A forum, a newspaper archive and an airline's records are each very different from the work history of a 100-person company.
None of these companies will sign a Reddit-style deal. Each can learn something different from the table.
Ignore the platform figures entirely. Your reference points are the published program ranges and the published minimums: Mode lists 20+ full-time US office employees (10+ for accounting firms, 6+ for law firms) and micro1 lists 30+. The lesson from the tracker is about terms: even large licensors negotiate term length and renewal, and so can you.
You have enough systems and history for buyers to value different parts differently, which is what made the Spirit auction competitive. The lesson is to collect more than one offer against the same manifest, and to compare them on exclusivity, scope of use and liability as well as price.
Your records may be large or unique enough for direct lab intake as well as data company programs. The lesson from platform deals is structure: ask whether a recurring license (new records each year) fits better than a one-off copy, and price exclusivity separately if a buyer asks for it.
Per-user or per-post arithmetic on a platform deal produces a number that does not apply to private company records. The assets, buyers and terms are different.
Program ranges such as "$100K-$5M" are bands with a top end, not averages. Most companies will not be at the top.
An offer filed in a court case or made in a negotiation is not a closed price. Ask what was actually paid, and under what conditions.
Export, review, de-identification, legal time and employee communication all cost something. A headline number never shows the seller's costs.
The Google and Reddit deal, signed in February 2024, has been widely reported at about $60 million a year. On July 22, 2026, CNBC reported a Wall Street Journal story that the two companies were negotiating a renewal and that Reddit had discussed ending Google's access to its content for AI use.
Among the deals on this tracker, the largest reported headline value is News Corp's multiyear agreement with OpenAI, announced in May 2024 and reported to be worth more than $250 million over five years. Many other deals have undisclosed terms, so a public tracker is never complete.
Shutterstock's CEO told Bloomberg that licensing to AI developers brought in $104 million in 2023, as reported in June 2024. The company projected about $138 million for 2024 for the business unit that includes AI licensing, according to a correction PetaPixel published on June 7, 2024.
Not directly. Platform and publisher deals license huge, mostly public content libraries, often with ongoing access. A company of 20 to 1,000 people licenses a private, scoped copy of internal work records. Use the ranges each buyer program publishes as your anchor instead.
Very few company-level deals are public, because they are private contracts under NDA. The visible exceptions are court-run sales like Spirit Airlines and the 2026 closure-market reporting on shut-down startups. The published payout ranges of buyer programs are the most useful public reference.
On July 22, 2026, CNBC reported a Wall Street Journal story that the deal was ending soon, that the companies were negotiating a renewal, and that Reddit had discussed ending Google's access to its content for AI use. We have not confirmed the outcome of those talks.
They are private contracts, usually under NDA, between a company and a data buyer. The ones that become public do so for special reasons: a bankruptcy sale that must go through court, like Spirit Airlines, or a market participant citing prices to reporters, like Troveo in the 2026 closure-market coverage.
Estimates only. Deedy Das's market map from July 2026 counts more than 50 companies selling data and RL environments to AI labs, with about $8.5 billion in revenue. Will Depue said in July 2026 that labs are on a path to more than $100 billion a year of data spend by 2030.
See which buyer programs your company fits, by their own published minimums and ranges. It runs in your browser and stores nothing.