Selling company data to AI: the complete guide
You can license company data to AI companies if you own it, can make it non-personal, and keep what would hurt you out. Reported payouts for ordinary companies run from about $10,000 to $100,000. Value comes from years of linked, written decisions. Rules depend on your industry, so check yours before talking to any buyer.
How should you use this guide?
Read the short answer under each heading, then follow the link when a topic matters to you. Every section names the detailed guide that covers it in full. If you only have five minutes, read the next four sections and the part for your industry.
| Your question | Read first |
|---|---|
| Is my data worth anything? | What data AI companies buy and how much company data is worth |
| Should I bother at all? | Is selling company data to AI worth it |
| Am I allowed to? | Who owns your company's data and is selling company data legal |
| How do I actually do it? | How to sell company data to AI |
| What deal should I sign? | Perpetual vs term, exclusive vs non-exclusive, lump sum or royalty |
| I'm in the UK or EU | Selling company data under GDPR |
Who buys company data, and why now?
AI labs and the companies that build training material for them. Public web text no longer gives labs what they need most: examples of real work, done by people, with the reasoning written down. Company archives hold exactly that.
Much of the demand comes from builders of practice environments for AI agents, which TechCrunch describes as "training grounds that simulate what an AI agent would be doing in a real software application." MIT Technology Review reported in October 2026 that buyers of startups' old data "include foundation AI labs, vendors that construct simulated workspaces for AI training", and that the records sold include source code plus Slack, Google Workspace, Notion and Figma data.
Experts are paid for the same knowledge directly. One report in October 2026 described how "a civil rights attorney can now earn $400 an hour with no client involved" by teaching AI systems, PYMNTS wrote, citing Bloomberg. Company data is the same knowledge, already written down.
What company data sells, and what doesn't?
Depth and decisions sell; volume alone doesn't. A buyer wants to follow one piece of work from request to discussion to outcome, across years and across systems.
| Raises value | Lowers value |
|---|---|
| Years of history on the same tools | A recent switch of tools that broke the record |
| Linked systems: chat, tickets, email, code, CRM | One isolated system |
| Decisions argued in writing | Work done by phone, with only outcomes written |
| A specialist field where public data is thin | Generic admin and scheduling email |
| Clean rights and a documented de-identification method | Data held for clients or under NDAs |
Small archives can beat large ones. A Slack history that "captures conversations with variety and depth, can be worth more than another's Slack, email, and calendars combined," a buyer told MIT Technology Review. More in what data AI companies buy.
How much do companies get paid?
Usually tens of thousands of dollars, sometimes millions, with no public price list. Forbes reported in April 2026 that shut-down startups' Slack, email and Jira archives sold for $10,000 to $100,000 each through one wind-down firm. Google bid $10 million for Spirit Airlines' business data in its bankruptcy, a rare case.
Other markets give reference points. AI companies have paid creators "between $1 and $4 per minute of footage" for unused video, Bloomberg reported. Reddit disclosed data licensing contracts worth $203 million in total, running two to three years. These are big public libraries, priced differently from a private archive.
Buyers value data like any asset: mostly against comparable deals, adjusted for quality and rights. How to estimate your own range is in how much company data is worth.
Is selling your data worth it?
Only when what you'd receive clearly beats the costs you can see and the ones you can't. Visible costs are legal review, de-identification and any broker's share. Hidden costs are staff or customer trust, a rival's tools learning your methods, and training that can't be undone.
It is most often worth it for companies closing down, firms with years of written specialist work, and businesses whose edge comes from relationships rather than process. The full weighing is in is selling company data to AI worth it.
Do you have the right to sell it?
Usually you own your records, but owning isn't the same as being free to license them. Software vendors' terms generally say the customer owns its data, yet their API terms can still forbid using exported data for AI training. Staff, clients and partners can all have claims.
- Client data you process under contract is usually theirs to decide on.
- Email and chat hold other companies' confidential information, often under NDAs.
- Your privacy notice at the time of collection still binds you.
- Employees' personal details stay protected even in company systems.
Each of these is covered in who owns your company's data, and Slack's rules in selling Slack data to AI.
What does the law say?
In the US, the main risks are your own promises and state privacy laws. California lets people ask businesses to "stop selling or sharing your personal information", the state Attorney General explains, so paid transfers of personal data trigger obligations. The federal trade regulator watches for broken privacy promises. Sector laws add more, as the industry sections below show.
In the UK and EU, GDPR applies to anything personal, and data with names swapped for codes still counts as personal. Data that is truly anonymous falls outside it. The US picture is in is selling company data legal, Europe in selling company data under GDPR, and nine countries in data licensing laws by country.
How do you make data safe to license?
By removing what identifies people, checked by someone independent, before anything leaves your systems. Removing names is not enough: job titles, dates and small details can still point to a person, and emails often describe health, family and pay.
The safest plan leaves out whole categories (HR, health, customer records) rather than trying to clean them. The methods and their limits are in how to de-identify company data.
How do you actually sell it?
Through one of four routes: a direct deal with a lab (for very large or rare datasets), a broker that works with operating companies, a wind-down platform for companies closing, or a data marketplace for datasets you can package yourself. Most companies use a broker or marketplace.
The routes, their published minimums and fees, and the questions to ask before signing are in how to sell company data to AI. Small firms below brokers' minimums should read can a small business sell its data.
What kind of deal should you sign?
| Choice | Usually best for an owner | Detail |
|---|---|---|
| Length | A 2 to 3 year term | Perpetual vs term |
| Exclusivity | Non-exclusive | Exclusive vs non-exclusive |
| Payment | A fixed fee as the floor | Lump sum or royalty |
| Scope | AI training only, written exclusions | Data licensing agreement terms |
Most AI content deals so far combine "a fixed upfront payment and a variable payout based on usage," a 2025 review of licensing deals found, and "most deals let publishers keep their options open" rather than going exclusive.
How do you negotiate it?
Prepare first, get more than one offer, and negotiate the terms as hard as the price. A law firm warns: "Do not share sample data before this review, since a sample itself could be a disclosure." Competition matters: in Spirit Airlines' auction, a $10 million winning bid was followed by a reported $12.5 million rival bid.
The full playbook, including a term sheet and the pressure tactics to watch for, is in how to negotiate a data licensing deal.
What are the long-term risks?
- Permanence: "once data is used to train a model, it cannot practically be pulled back out," a law firm notes. Price a deal as if trained models are permanent.
- Competitors: data that shows how you price, sell or work can teach a rival's tools. See could selling your data help your competitors.
- Your company's value: a one-off sale is one-off income, and exclusivity or messy rights can cut a future sale price. See does selling data affect your company's valuation.
- Trust: staff and customers may object. Spirit's flight attendants' union objected to its sale.
Which industries can sell their data, and under what rules?
Every industry has data buyers might want and rules that limit it. This table summarises both; the sections that follow explain each one.
| Industry | What buyers want | Main rule to check | Outlook |
|---|---|---|---|
| Healthcare | De-identified records and imaging | HIPAA de-identification | Strong demand, strict rules |
| Legal | Primary law and annotations | Client confidentiality | Content yes, client files rarely |
| Financial services | Licensed market and company data | Gramm-Leach-Bliley Act | Market data yes, customer data rarely |
| Software and SaaS | Code and workspace records | Software terms, contracts | Strong demand |
| Customer support | Real conversations | Call-recording consent | Interest, no confirmed deals found |
| Media and publishing | Archives and current content | Copyright, contracts | Many deals done |
| Video, images, audio | Unused and raw footage | Rights and releases | Paid per minute |
| Education | Expert-written content | FERPA for student records | Demand not established |
| Retail and ecommerce | Reviews and product data | Card data rules, state privacy | Early interest |
| HR and recruiting | Hiring workflows | FCRA for background checks | Reports can't be sold |
| Children's services | Rarely worth it | COPPA | Leave children's data out |
| Manufacturing | Factory video, technical files | Export controls | Strong physical-AI demand |
| Construction and engineering | Drawings and design files | Who owns the drawings | Early, rights first |
| Government contractors | Project data | Contract data-rights clauses | Often the government's |
Can healthcare companies sell data to AI?
Yes, if it is properly de-identified, and there is real demand. Protege, which raised funding in January 2026, aggregates "de-identified clinical health records and medical imaging" for AI developers, SiliconANGLE reported.
The rule is HIPAA. Under 45 CFR 164.514, "a covered entity may determine that health information is not individually identifiable health information only if" it meets one of two methods: removing a set list of identifiers (Safe Harbor) or having an expert certify that the risk of identifying anyone is very small (Expert Determination). Anything else is still protected health information. Business associates also need their agreements to allow any use.
Verdict: one of the strongest markets, but only through certified de-identification and specialist partners. Never license identified records.
Can law firms and legal publishers sell data to AI?
Legal content, yes; client files, rarely. Demand for legal information is intense: when Clio bought vLex for about $1 billion in 2025, its CEO said "data is one of the only long-term defensible competitive moats a company can have in the space," TechCrunch reported. Publishers have licensed primary law content to legal AI firms.
Client information is another matter. Summarising the American Bar Association's 2024 opinion on generative AI, one law firm notes that "lawyers are required to obtain informed client consent" before putting client information into self-learning AI tools. A firm's archive of client work is privileged and confidential, so it is very hard to license.
Verdict: publishers and legal-information businesses can license content; law firms should assume client files are off limits without each client's informed consent.
Can financial companies sell data to AI?
Licensed market and company data, yes; customer data, rarely. In October 2025 the London Stock Exchange Group announced a collaboration with Anthropic "granting access to data licensed through LSEG products, like Workspace and Financial Analytics." A trade publication put it bluntly: large language model providers are "coming for this industry."
Customer data is restricted by the Gramm-Leach-Bliley Act. The FTC's guide tells financial institutions what they must do "before you share NPI with nonaffiliated third parties": notice and an opt-out, unless an exception applies. Selling customers' nonpublic personal information to an AI company is very unlikely to fit.
Verdict: research, analytics and market data can be licensed; customer records should stay out.
Can software and SaaS companies sell data to AI?
Yes, and this is where most company deals happen. Stack Overflow announced a partnership with OpenAI in May 2024. Shut-down startups sell code and workspace records through wind-down platforms, and buyers build agent practice environments from exactly this material.
The limits are contracts, not a sector law: your customers' data inside your product is usually theirs, open-source licenses on included code still apply, and vendor terms govern how you export chat and tickets. Code you still rely on is also a trade secret; see could selling your data help your competitors.
Verdict: strong demand for internal records and old code; keep customer data and current core code out.
Can customer support and contact centers sell data to AI?
Possibly, but we found no confirmed deal by a major buyer, only broker listings. Real conversations are what AI customer service agents learn from, so interest is plausible.
Recorded calls raise consent questions. In California, recording a confidential call without "the consent of all parties" is a crime under Penal Code 632, and several other states require all-party consent too. Telephone carriers face separate FCC rules on customer network information. Customers' personal details in tickets and calls must come out.
Verdict: written tickets with outcomes are the safer asset; treat call recordings as high risk unless consent was clearly given.
Can media and publishing companies sell content to AI?
Yes; this is the most established market. News publishers, book publishers and image libraries have signed deals with AI companies since 2023, usually non-exclusive and for two to six years.
The issues are rights: what your contracts with writers, photographers and contributors allow, and whether you can license archives you didn't create. Publishers have also negotiated attribution and reset terms; see how to negotiate a data licensing deal.
Verdict: a working market, where contributor contracts decide what you can include.
Can you sell video, images and audio to AI?
Yes, often by the minute. AI companies have paid creators "between $1 and $4 per minute of footage" for unused video, Bloomberg reported in January 2025, with more for high-quality material.
You need rights to every frame: releases from people shown, permission for music and locations, and nothing recorded without consent. Faces and voices can be biometric data: Texas law, for example, says a business "may not sell, lease, or otherwise disclose the biometric identifier to another person" except in listed cases such as consent.
Verdict: a paying market for footage you fully own; anything with identifiable people needs releases first.
Can education companies sell data to AI?
Expert-written content possibly; student records no. We found no confirmed AI training deal by a major textbook or course company, although academic publishers have licensed content.
Student records are protected by FERPA, which requires that a parent or eligible student "shall provide a signed and dated written consent" before records are disclosed. Vendors can receive records only as "school officials" under the school's "direct control", which doesn't fit selling them for a buyer's own training.
Verdict: license your own course content if contributors' contracts allow; never student data.
Can retail and ecommerce companies sell data to AI?
Product and review data is drawing interest as AI shopping agents grow. Trustpilot's CEO has said "the most effective systems will rely on datasets like those held by Trustpilot," AI News reported in March 2026.
Customer data is the limit. Card data must stay within the PCI standard, which "provides a baseline of technical and operational requirements designed to protect payment account data", and state privacy laws give shoppers the right to stop the sale of their personal information.
Verdict: catalogs and anonymous product content may sell; customer and payment data should stay out.
Can HR, recruiting and background-check firms sell data to AI?
Hiring workflows and job content possibly; consumer reports no. The Fair Credit Reporting Act lets a consumer report be furnished only "under the following circumstances and no other", and selling it to an AI company is not one of them.
Candidate and employee data is personal data under state privacy laws, and biometric data (face or voice) carries extra rules. Recruiters' value is usually in their process, which also makes it competitively sensitive.
Verdict: leave candidate data out; consider only de-identified process records, if at all.
Can companies serving children sell data to AI?
No, in practice. Under the FTC's children's privacy rule, an operator "must obtain verifiable parental consent before collecting any personal information from a child", and the 2025 changes require operators to obtain "separate verifiable parental consent to disclose children's personal information" to third parties.
Verdict: leave any data about children out of every deal.
Can manufacturers sell data to AI?
Physical AI is hungry for data. The Financial Times reported in October 2026 that robotics and physical AI firms had raised close to $48 billion that year, with one calling real-world data "Physical AI's scarcest resource." Factory-floor video is already being collected: one company described its dataset as "the backbone for AI/robotics training globally," Scroll reported.
Export controls are the main limit. Technical data under US arms export rules covers information "required for the design, development" of defense articles, and controlled technology under the Export Administration Regulations stays controlled unless it has been "made available to the public without restrictions upon its further dissemination." Defense suppliers also hold controlled information under their contracts.
Verdict: real demand for footage and process data; check export classification and contracts before anything technical leaves.
Can construction and engineering firms sell drawings to AI?
Possibly, if they own them, and the market is early: no prices for companies' engineering files have been published. Under standard AIA contracts the architect owns the drawings, and the building owner only has a license to build with them.
Verdict: your own layouts and shop drawings may be licensable; architects' plans and client NDA work need permission. The full picture is in can you sell CAD files and drawings to AI.
Can government contractors sell project data to AI?
Often not without checking the contract. Under the standard federal data rights clause, "the Government shall have unlimited rights in" data "first produced in the performance of this contract." Defense contracts distinguish data developed at private expense, which the contractor keeps more control over.
Verdict: separate privately funded work from contract deliverables, and have counsel read the data rights clauses first.
Can consultancies, agencies and accounting firms sell data to AI?
Their own internal records, sometimes; client work, rarely. Expert knowledge is exactly what AI buyers want, which is why AI data companies pay professionals by the hour. But most of a services firm's archive is client work under engagement terms and NDAs.
Verdict: start from internal playbooks and non-client records, check every engagement letter, and expect clients to own much of the rest.
What should you do in your situation?
Which describes you?
An operating company with years of records
- License, don't sell: a short, non-exclusive term.
- Keep customer data, HR files and pricing logic out.
- Read is it worth it first.
Winding up or about to
- Start before software subscriptions lapse.
- Expect creditors and investors to be paid first.
- Read selling your data when you close.
A trustee or court is involved
- Customer data sales may need a privacy ombudsman's review.
- The trustee decides, not you.
An exit in the next year or two
- Avoid exclusivity and licenses that can't pass to a buyer.
- Read does selling data affect your valuation.
Under 20 staff
- Look at marketplaces, pooling or a niche slice.
- Read can a small business sell its data.
Data about people in Europe
- Only truly anonymous data is outside GDPR.
- Read selling company data under GDPR.
A collective agreement covers your workforce
- Check whether it restricts using staff communications.
- Consider leaving staff messages out entirely.
Agency, outsourcer or software provider
- Client data is usually theirs: get written permission or leave it out.
- Read who owns your company's data.
One more situation catches owners out: your vendors hold your data too. A law firm advises reviewing those contracts, because a vendor may try to sell what it holds if it is acquired, shuts down or goes bankrupt.
The complete checklist, from first thought to signed deal
- List every system, its years of history and rough record counts. Export nothing yet.
- Check ownership: vendor terms, client contracts, NDAs, privacy notices, staff policies.
- Find your industry above and check its rule.
- Decide what stays out: personal, client, HR, health, children's, pricing and export-controlled data.
- Estimate a fair range from comparable deals, and set your walk-away point.
- Get legal review of the plan before any sample exists.
- Have an independent specialist de-identify a small sample.
- Talk to more than one buyer or route, keeping everything non-exclusive.
- Agree a short term sheet: length, exclusivity, use, payment, exclusions, deletion.
- Turn it into a contract with your lawyer, including reports and audit rights for any variable fee.
- Deliver securely and keep a record of exactly what left.
- Diary the renewal date, and charge for any fresh data separately.
For a quick read on where you stand before any of this, see if your data can sell. It asks about your systems and history, never for files.
What do the terms mean?
| Term | Meaning |
|---|---|
| Data licensing | Letting a buyer use a copy of your data on agreed terms while you keep it |
| Outright sale | Transferring the data so it becomes the buyer's |
| Training data | Material an AI model learns from |
| Practice environment | A simulated workplace where AI agents rehearse real tasks; built partly from company records |
| De-identification | Removing what could identify people, ideally checked by an independent expert |
| Anonymous data | Data from which nobody can be identified, even by linking; outside GDPR |
| Pseudonymised data | Names replaced by codes that could be linked back; still personal data under GDPR |
| Term license | A license for a fixed period, such as two or three years |
| Perpetual license | A license with no end date |
| Exclusive license | A promise that nobody else gets the data, for an agreed period and use |
| Reset clause | A right to renegotiate if a later buyer gets better terms |
| Minimum payment | A fixed amount paid whatever the usage |
| Revenue share or royalty | A percentage of what the buyer earns from your data |
| Term sheet | A short list of key terms agreed before the full contract |
| Data broker | A firm that finds buyers and prepares data, for a fee or share |
| Data marketplace | A platform where you list a dataset for buyers, usually for a commission |
| Wind-down platform | A service that sells closing companies' assets, including data |
| Consumer privacy ombudsman | A court-appointed reviewer for sales of customer data in US bankruptcy |
| Trade secret | Valuable information kept secret with reasonable measures |
| Work made for hire | Work an employee creates in their job, owned by the employer under US copyright law |
Questions
Can I sell my company's data to AI companies?
Yes, if you own it, can make it non-personal and leave out what would hurt you. Most companies license through a broker, marketplace or wind-down platform rather than directly.
How much is company data worth to AI companies?
Reported payouts for ordinary companies run from about $10,000 to $100,000; rare cases reach millions, such as Google's $10 million bid for Spirit Airlines' data. There is no public price list.
What company data do AI companies want most?
Years of linked, written work where decisions can be followed: chat, tickets, email, code, and specialist records. Volume alone matters little.
Is it legal to sell company data to AI?
Often, for data you own that is de-identified and within your privacy promises. Industry laws such as HIPAA, GLBA, FERPA, COPPA and export controls add limits, and GDPR applies in Europe.
Which industries can sell data to AI?
Software, media, video and healthcare (de-identified) have the clearest demand. Legal, finance and retail can license content but rarely customer data. Children's data should never be sold.
Can healthcare data be sold to AI companies?
Only when de-identified under HIPAA, through Safe Harbor or an expert's determination. Identified patient records cannot be licensed without authorization.
Can a law firm sell client files to AI?
Rarely. Client information is confidential, and lawyers are expected to get informed client consent before putting it into self-learning AI tools.
Should I sell or license my data?
License, usually for two to three years and non-exclusive, so you can charge again and sell to others. Sell outright mainly when closing down.
How long does it take to sell company data?
It depends mostly on preparation: rights review, exclusions and de-identification take longer than finding a buyer. Starting early matters most for companies closing down.
Do I need a broker to sell data to AI?
No, but most first-time sellers use a broker, marketplace or wind-down platform. Get every fee in writing and keep the deal non-exclusive.
Can selling data hurt my business?
It can: trained models can't unlearn your data, rivals' tools may learn your methods, and staff or customers may object. Leaving sensitive data out limits the damage.
Does GDPR stop me selling company data to AI?
Not if the data is truly anonymous. Personal or pseudonymised data needs a lawful basis, a compatible purpose and notice to the people concerned.
Sources
- AI labs are buying data from dead startups to train their models (13 October 2026) MIT Technology Review
- Failed companies are selling old Slack chats and email archives to train AI (17 April 2026) Gizmodo, reporting Forbes
- Silicon Valley bets big on 'environments' to train AI agents (September 2025) TechCrunch
- Report on AI companies paying experts for training work (20 October 2026, citing Bloomberg) PYMNTS
- Google scoops up Spirit Airlines' data in bankruptcy sale to train AI (17 August 2026) Skift
- Fighting Spirit's data sale to Google: court filings show the list getting longer (12 October 2026) Skift
- YouTubers are selling their unused video footage to AI companies (10 January 2025) Bloomberg, via Bloomberg Law
- Reddit says it's made $203M so far licensing its data (22 February 2024) TechCrunch
- The 7 deal points of AI content licensing agreements (9 May 2025) Rob Kelly, Media & the Machine
- Selling company data for AI training: what to know before you say yes (1 October 2026) Frankfurt Kurnit Klein & Selz, law firm
- California Consumer Privacy Act California Attorney General
- Protege raises $30M to grow governed marketplace for AI training data (8 January 2026) SiliconANGLE
- 45 CFR 164.514, de-identification of protected health information Electronic Code of Federal Regulations
- Legal software company Clio drops $1B on law data giant vLex (30 June 2025) TechCrunch
- American Bar Association issues formal opinion on generative AI (14 August 2024) JD Supra, law firm summary
- LSEG announces collaboration with Anthropic (27 October 2025) London Stock Exchange Group
- LLM firms come for finance (31 October 2025) WatersTechnology
- How to comply with the Privacy of Consumer Financial Information Rule of the Gramm-Leach-Bliley Act Federal Trade Commission
- Stack Overflow and OpenAI partnership (6 May 2024) Stack Overflow
- California Penal Code § 632 FindLaw
- Texas Business and Commerce Code § 503.001, biometric identifiers Texas Public Law
- FERPA US Department of Education, Student Privacy Policy Office
- 34 CFR 99.31, disclosure without consent Legal Information Institute, Cornell Law School
- AI in ecommerce: Trustpilot partnerships (17 March 2026) AI News
- PCI Data Security Standard PCI Security Standards Council
- 15 U.S. Code § 1681b, permissible purposes of consumer reports Legal Information Institute, Cornell Law School
- Complying with COPPA: frequently asked questions Federal Trade Commission
- FTC finalizes changes to children's privacy rule (16 January 2025) Federal Trade Commission
- Robotics and physical AI labs raise ~$48B YTD on data hunt (11 October 2026) AI Weekly, reporting the Financial Times
- How Big Tech is harnessing the data of Indian factory workers to train robots (20 May 2026) Scroll
- 22 CFR § 120.33, technical data Legal Information Institute, Cornell Law School
- 15 CFR § 734.7, published Legal Information Institute, Cornell Law School
- 48 CFR 52.227-14, rights in data, general Legal Information Institute, Cornell Law School