How to sell your company's data to AI companies
There is no public price list or upload form. Most companies license their data through a broker, a wind-down platform or a data marketplace, and labs deal directly only for large or unusual datasets. Check your rights first, de-identify a sample, then compare routes on fees and terms, not the headline offer.
Who is buying company data right now
In August 2026, Google won a bankruptcy auction for Spirit Airlines' business data and software code with a $10 million bid, Skift reported, with the AI recruiting firm Mercor as backup bidder at $7.5 million. The sale covered finance, operations and revenue management systems, and explicitly left out all customer and loyalty data.
Smaller deals are far more common. MIT Technology Review reported in October 2026 that SimpleClosure, which helps startups shut down, had handled over 150 deals through its asset platform. One founder got about $25,000 for his code base, with a further payment of around $15,000 for workspace data still pending. Earlier in the year Forbes reported payouts of $10,000 to $100,000 per company.
The buyers are labs and the companies that build training material for them, such as Surge AI, which builds practice environments where AI agents learn to do work. What they want is covered in what data AI companies buy.
The four routes, compared
| Route | Examples | Who it suits | Fees |
|---|---|---|---|
| Direct deal with a lab | OpenAI's data partnership programme (2023) | Very large or unusual datasets | No public terms |
| Broker for operating companies | micro1, Nyne, Troveo | micro1: 30+ employees, English documentation, strongest demand in the US, UK and Canada. Nyne: 20+ US W-2 employees and 3+ years in business. Troveo: no minimum stated | Not published by any of the three |
| Wind-down platform | SimpleClosure AssetHub, Sunset | Companies shutting down; AssetHub opened to operating companies in September 2026 | AssetHub is free for SimpleClosure's dissolution customers; other fees not published |
| Data marketplace | Datarade | Companies that can package and describe a dataset themselves | 30% commission with no yearly fee, down to 15% on a $12,000 a year plan |
Brokers quote high: micro1 shows tiers from $100,000 to over $1 million, and Nyne says partnerships are "commonly worth $100K or more". Neither publishes its fee. Treat any figure as an opening estimate until it is in a signed agreement, and read how much company data is worth for what reported deals actually paid.
Datarade's rules ban datasets with personal information that hasn't been anonymised, and the brokers de-identify before anything leaves your hands. For OpenAI specifically, see how to sell data to OpenAI.
Will anyone buy your data?
The minimums brokers publish describe the same company: an established business with a few dozen staff, several years of history, and work recorded in English across connected systems such as email, chat, tickets and a CRM. A five-person firm with a shared inbox is unlikely to interest anyone yet.
- Strong: years of discussion that ends in decisions, such as support tickets with resolutions or sales threads with outcomes.
- Weaker: finished documents with no trail of how they were made.
- Usually ruled out: data you hold for clients, health records, and anything your contracts say you can't reuse.
Take the free data check for a plain read on your own systems. It asks about systems and history, never for files.
How to sell company data to AI, step by step
- List each system, its years of history and rough record counts. Don't export anything yet.
- Check what you're allowed to license: customer contracts, privacy notice, employee handbook and each software provider's terms. See is selling company data legal and selling Slack data.
- Decide what stays out, such as customer records, HR files and anything held for clients. Spirit's sale left out every passenger and loyalty record.
- Have an independent specialist de-identify a small sample, and review it yourself. See how to de-identify company data.
- Pick a route and ask the questions in the next section before sharing the sample, under a confidentiality agreement.
- Negotiate a licence, not a sale: limited use, a set term, no resale, and deletion at the end. See data licensing agreement terms.
- Deliver through a secure transfer the buyer can't copy on to others, and keep a record of exactly what went out.
Never send raw exports to a buyer or broker to "take a look". A de-identified sample under a confidentiality agreement is enough to judge fit.
Questions to ask a broker before you sign
- What is your fee or share, and is it taken once or on every sale?
- Who are the buyers, and will you tell me before each one sees the data?
- Is the licence exclusive? Can a buyer resell it or train a model for a competitor of mine?
- Who de-identifies the data, who checks it, and who is liable if a person is re-identified?
- When am I paid, and is any of it tied to the buyer's acceptance or later use?
- What happens to every copy at the end of the term?
A broker that won't answer these in writing is asking you to take its word on the part of the deal you can't take back.
What an operating company should weigh
Most reported deals come from companies that had already closed. A business that is still trading has more to lose.
- Your people: Spirit's flight attendants' union objected to the sale on privacy grounds, MIT Technology Review reported. Emails and calendars hold medical appointments and family plans.
- The law is thin: a law professor told the same magazine there is no federal law banning these sales, so your contracts and privacy promises do the work.
- Your know-how: a dataset that teaches how you price, sell or support customers can teach a rival's tools the same.
- Permanence: you can require deletion of copies, but a model already trained on the data can't simply forget it.
None of this rules a sale out. It is why the scope, the exclusions and the licence terms matter more than the price.
Questions
How do I sell my company's data to AI companies?
List your systems, check your contracts and software terms, de-identify a sample with a specialist, then license through a broker, a wind-down platform or a data marketplace. Labs rarely buy directly from ordinary companies.
How much do AI companies pay for company data?
Reported deals range from about $25,000 for a startup's code to $10 million for Spirit Airlines' data. Forbes reported $10,000 to $100,000 per company for shut-down startups. There is no public price list.
Do I need a broker to sell company data to AI?
No, but most first-time sellers use one or a marketplace. Ask any broker its fee, its buyers and whether the licence is exclusive before sharing a sample.
What do data brokers charge?
Of the routes in this guide, only Datarade publishes fees: 30% commission with no yearly fee, down to 15% on a $12,000 a year plan. micro1, Nyne and Troveo don't publish theirs.
Can a company that is still operating sell its data?
Yes. Brokers such as micro1 and Nyne work with operating companies, and SimpleClosure opened its platform to them in September 2026. Leave out customer and HR data, and license rather than sell.
Sources
- Google scoops up Spirit Airlines' data in bankruptcy sale to train AI (17 August 2026) Skift
- 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
- SimpleClosure AssetHub on track to return $10M to venture-backed startups (3 September 2026) SimpleClosure, press release
- Provider plans and commission Datarade
- Enterprise data partnership estimate micro1
- Sell company data Nyne
- How to license company data for AI (20 August 2026) Troveo
- OpenAI Data Partnerships (November 2023) OpenAI
- Selling company data for AI training: what to know before you say yes (1 October 2026) Frankfurt Kurnit Klein & Selz, law firm