Over $1 billion in AI data deals. Only the public ones.
Only 28 of the 107 public deals we track name a dollar price. Rightmark licenses your data to AI companies: priced against the public ones, negotiated in private.
Values from the Rightmark deal tracker. The 24 licensing deals and sales with a dollar price add up to $1.2B. Acquisitions are counted apart. Of the values in that total, 2 are estimates and the rest are as reported. The one winning bid still waiting for court approval stays out of the total. Tracker updated .
Rightmark licenses private data to AI companies for its owners: code with git history, company archives such as Slack, email and tickets, de-identified health records, audio, video, documents and robot logs. Rightmark does not take health data yet (patient records, medical imaging or veterinary records), until a HIPAA program and vendor agreements are in place. The appraisal is free and takes a minute. Owners keep ownership and pay nothing upfront; on standard licenses the buyer pays our fee.
- A large airline archive of email, chat and code drew a winning bankruptcy auction bid of $10M, subject to court approval. TIME, (opens in a new tab)
- AI video labs were reported to pay creators $1 to $4 per minute for footage that never made the edit. Bloomberg Law, (opens in a new tab)
- One publisher licensed nonfiction backlist titles for AI training at $5,000 per title for 3 years, split with authors who opt in. eWeek, (opens in a new tab)
Updated
What do you have?
Indicative ranges- Code and repositoriesPrivate repos with their git history, issues and reviews.$150 to $1,500 per repoEstimate, Medium confidence
- Company archivesSlack, email, Jira, docs and support history from real teams.$10,000 to $1M+ per companyEstimate, Medium confidence
- Health and life sciencesDe-identified imaging, notes, pathology, dental and vet records.Revenue share
- Voice and audioCalls, multilingual speech, podcasts, music and sound.$6.30 to $150 per hourMedium confidence
- Video and imageryUnused footage, first-person video, drone, dashcam and photo libraries.$1 to $4 per minuteMedium confidence
- Books, documents and expert workBacklists, news archives, translation memories, courseware and forums.$5,000 per title for 3 yearsOne reported deal, High confidence
- Robotics and sensorsTeleoperation logs and industrial sensor data.$50 to $200 per hourEstimate, Low confidence
- Something elseTell us what it is. We review every submission by hand.Priced after review
Indicative estimate, not an offer. Final offers depend on the systems you connect, data size and quality, access terms and due diligence.
- You keep ownership
- Nothing is licensed until you sign
- No upfront fees
- Anything you share is stored privately
AI labs have read the public internet. Now they pay for what is private.
- $10MOne reported deal, Medium confidenceThe winning bid for a bankrupt airline archive of email, chat logs and code, subject to court approval.Source: TIME, (opens in a new tab)
- $1 to $4 per minuteMedium confidenceWhat AI video labs were reported to pay creators for footage that never made the edit.Source: Bloomberg Law, (opens in a new tab)
- $5,000 per title for 3 yearsOne reported deal, High confidenceOne publisher’s AI training license for nonfiction backlist titles, split with authors who opt in.Source: eWeek, (opens in a new tab)
Wanted now
Indicative ranges- Private repos with full git historyAI labs and training-task builders$150 to $1,500 per repoEstimate, Medium confidence
- Company archives from teams of 30+Data programs for AI labs$185,000 to $1M+ per companyEstimate, Medium confidence
- First-person video of real tasksRobotics foundation models$15 to $50 per hourMedium confidence
- Unused 4K, drone and B-roll footageVideo model labs$1 to $4 per minuteMedium confidence
- Speech in low-resource languagesSpeech and voice AI$60 to $150 per hourMedium confidence
- CT and MRI studies paired with reportsMedical AI developersRevenue share
- Nonfiction and STM backlist titlesModel developers$5,000 per title for 3 yearsOne reported deal, High confidence
- Robot teleoperation logs with force dataRobotics labs$50 to $200 per hourEstimate, Low confidence
Ranges from public reporting and published buyer programs, with Rightmark estimates marked, Price Index Q4 2026. Indicative estimate, not an offer. Final offers depend on the systems you connect, data size and quality, access terms and due diligence.
What an hour of data licenses for
Buyers price data by the unit: the repo, the company, the title, the hour. Here is one unit across audio, video and robotics. Every row links to its source.
- Longer films, one data vendor's market rateVendor-reported rates, Reuters, Apr 2024$100 to $300
- Unused creator footage$1 to $4 a minute, converted, The Decoder, Jan 2025$60 to $240
- Robot teleoperation logsIndustry estimates, DreamVu, Aug 2026$50 to $200
- Low-resource language speechLow-resource languages, SpeechData.ai, Jun 2026$60 to $150
- Ready-made speech datasetsReady-made datasets, SpeechData.ai, Jun 2026$50 to $100
- First-person task videoBuyer-side estimates, Stellaris VC, Apr 2026$15 to $50
- Bulk call-center audioMarketplace asking price, Datarade via Internet Archive, Aug 2025$6.30 to $7
- First-person video, undifferentiated bulk footageBulk trade estimate, Flikforge, Jul 2026$2 to $5
Show as a table
| Data type | Price | Unit | Confidence | Source |
|---|---|---|---|---|
| Longer films, one data vendor's market rate | $100 to $300 | Vendor-reported rates | Low confidence | Reuters, Apr 2024 |
| Unused creator footage | $60 to $240 | $1 to $4 a minute, converted | Medium confidence | The Decoder, Jan 2025 |
| Robot teleoperation logs | $50 to $200 | Industry estimates | Low confidence | DreamVu, Aug 2026 |
| Low-resource language speech | $60 to $150 | Low-resource languages | Medium confidence | SpeechData.ai, Jun 2026 |
| Ready-made speech datasets | $50 to $100 | Ready-made datasets | Medium confidence | SpeechData.ai, Jun 2026 |
| First-person task video | $15 to $50 | Buyer-side estimates | Medium confidence | Stellaris VC, Apr 2026 |
| Bulk call-center audio | $6.30 to $7 | Marketplace asking price | Medium confidence | Datarade via Internet Archive, Aug 2025 |
| First-person video, undifferentiated bulk footage | $2 to $5 | Bulk trade estimate | Low confidence | Flikforge, Jul 2026 |
Per company archive
$10,000 to $1M+ per companyEstimate, Medium confidenceShut-down startups were reported to license their archives for $10,000 to $100,000 each. A bankrupt airline’s archive drew a $10M winning bid, subject to court approval.
Sources: Gizmodo, (opens in a new tab); TIME, (opens in a new tab)Per code repository
$150 to $1,500 per repoEstimate, Medium confidenceMost repos license at standing rates. The best few percent license as training tasks.
Sources: Rightmark, ; Epoch AI, (opens in a new tab)Per book title
$5,000 per title for 3 yearsOne reported deal, High confidenceAI training rights for 3 years, split with authors who opt in.
Source: eWeek, (opens in a new tab)Indicative estimate, not an offer. Final offers depend on the systems you connect, data size and quality, access terms and due diligence.
From forgotten archive to licensing income
Appraise
Answer a few questions about what you have. See an indicative range, the buyers it fits and what would raise the price. No files, no access.
About a minute
Verify
Scans for secrets, personal data and third-party rights run on your side (Repo Check runs on your machine) or in the buyer’s pipeline. We review the reports and, under NDA, scrubbed samples, then assemble the ownership pack your route needs.
A few days for company data; code scrubbing takes hours, up to 2 to 4 weeks for large histories
Match
We route each dataset to the buyer that fits it best. Code is licensed non-exclusively by default. Company-archive lump sums are usually exclusive for a term; revenue share keeps you non-exclusive.
At your pace. Code: labs review it over 2 to 16 weeks
Get paid
We never hold your passwords. Transfers use scoped, revocable connections your admins approve, or exports. We take nothing from your payout on standard licenses; the buyer pays our fee on top. You keep ownership, and on non-exclusive routes you can license again.
Company data: transfer takes about 1 to 2 weeks, payment comes from about 2 to 4 weeks after signing at the fastest, to about 13 weeks. Code: about 4 to 10 months from submission to first payment
Indicative timings, confirmed at scoping.
We work like a leasing agent for a building. We find the tenants and handle the paperwork, and the rent goes straight to you. You keep the building, and on non-exclusive routes you can lease it again.
One exchange, three sides
Data owners
Turn archives into licensing revenue.
- Free appraisal in about a minute
- Rights and privacy checks coordinated for you
- You approve the agreement and its scope before anything moves
Scouts
Get paid for introductions.
- Introduce owners you already know
- Paid from our income, never from the owner’s payout
AI teams
Private data with clean title.
- Code, company archives, health, audio, video and more
- Ownership packs with consents and scan logs
- Non-exclusive or exclusive, delivered in batches
Lots open to buyers
Each lot shows its figures, where they come from and when its round opens. Ask about any lot and we reply with scope and timing.
Code and company records from one software company
Round opens Tuesday 27 October 2026
- about 290 Repositories
- about 5 million Original lines of code
- about 70,000 Support tickets
- about 4 TB Documents and email
Find your data
Each page explains why buyers want it, what sets the price and what blocks a deal, with sources.
Company data9
Health5
Audio4
Video and images5
Documents5
Five things that change the price
History is the product.
Git history often carries more than half of a repo’s standing-rate value. Squash or migrate it and much of the price goes with it.Source: Rightmark,
Offcuts sell.
AI companies have paid creators for footage that never made the edit. Rejected tickets and failed robot attempts are unused material too, so count them before you clear them out.Source: Bloomberg Law, (opens in a new tab)
Deleting is destroying.
Shut-down companies have licensed their old chat and email archives for AI training. Export a workspace or a software plan before you cancel it, then decide.Source: Gizmodo, (opens in a new tab)
Consent is part of the product.
For voice and likeness we look for consent that names AI training. Read what your recording notices and consents say before you plan a license.
Your vendors may already hold rights.
We ask whether any software vendor claims training rights over your data, because data you cannot authorize cannot be licensed. Check those contracts before you promise anyone exclusivity.
Updated
Nothing is licensed until you sign.
Owners license, they do not sell. We check title, flag what cannot be licensed so it is excluded before anything is shared, and check buyers against sanctions lists before any agreement. Anything you share with us is stored privately and deleted when you ask.
How we protect owners- Answers, not files. The appraisal needs no data and no access to your systems.
- NDA before samples. Scans run on your side or in the buyer pipeline. We see reports and scrubbed samples under NDA. Verification documents are stored privately and deleted 12 months after a deal closes or on request.
- You approve the agreement and its scope. On code routes, signing covers future non-exclusive sales, and you can withdraw code that has not been licensed yet.
- Use limits set per route. Each license states its scope, its term and what happens to the data at the end.
Reporting on the market we serve
Reporting on AI data licensing in general. None of it is about Rightmark.
TIME (opens the publisher's site)
On the winning bid for a bankrupt airline archive of email, chat and code, subject to court approval.
Fortune (opens the publisher's site)
On a rival bidder’s later, higher offer for the same airline archive.
Forbes (opens the publisher's site)
On AI companies paying for old work chat messages and email.
Bloomberg Law (opens the publisher's site)
On creators licensing unused video footage to AI companies.
What owners ask first
What does Rightmark do?
We find private data that AI teams need, check who owns it, check that secrets and personal details are removed (scans run on your side or in the buyer pipeline), and route it to the buyer that fits it best. Owners keep ownership, and anything they share with us is stored privately and deleted when they ask. Scouts who introduce owners earn a share of our income, never of the owner’s payout.
What kinds of data can I license?
Code with its history, company archives such as Slack, email, tickets and docs, de-identified health data, audio, video, images, books and other documents, and robot and sensor logs. Rightmark does not take health data yet (patient records, medical imaging or veterinary records), until a HIPAA program and vendor agreements are in place. If it is something else, tell us in the appraisal.
Do I give up ownership?
No. You grant a license and keep the data, and you can keep using it. Code is licensed non-exclusively by default, so a repo can be licensed to several buyers. Company-archive lump sums are usually exclusive for a term; non-exclusive revenue share is the alternative. Read each license’s scope, because some also allow the platform its own use or derived datasets.
What does it cost?
Nothing upfront. Buyers cover extraction, scrubbing and transfer. We take nothing from your payout on standard licenses: the buyer pays our fee on top. On premium and bespoke deals you keep 75% to 85%, agreed in writing before any sale.
How long does it take?
The appraisal takes a minute. For company archives, screening takes a few days and signing at the fastest about 3 to 6 weeks from your first email; payment arrives from about 2 to 4 weeks after signing at the fastest, to about 13 weeks. Code takes longer because AI labs choose what to license: about 4 to 10 months from submission to first payment, then again on each further license. Health data takes months, because de-identification and contracts come first. Rightmark does not take health data yet (patient records, medical imaging or veterinary records), until a HIPAA program and vendor agreements are in place. Indicative timings, confirmed at scoping.
Is this legal?
Licensing data you own is legal. What makes a deal safe is the work around it: confirming title, removing personal data, honoring consent and privacy law, and checking buyers against sanctions lists. That work is what we do.
Find out what your data is worth.
About a minute, no files, no obligation. You will see a range, the buyers it fits and what would raise the price.