To sell company data to AI, you license a de-identified copy of your business records (chat, email, files, tickets, code) to an AI lab or an AI data company, and you keep the originals. The path is: check you qualify, list what you have, confirm your rights, get written offers from more than one buyer after a sample review, sign, deliver, and get paid. Published timelines run from under 21 days from first call to first wire (Labelbox Prism) to net 30 to 60 days after approval and delivery (Polyshares), both checked October 2026.
On this page
- Who buys company data for AI?
- Step 1: Do I qualify?
- Step 2: What should I prepare before talking to buyers?
- Step 3: Do I have the right to license these records?
- Step 4: How do I get offers?
- Step 5: What happens when I sign?
- When do I get paid?
- What mistakes should I avoid?
- Can I keep using my data after I license it?
- Questions
- Sources
Who buys company data for AI?
Four kinds of companies are active in 2026. Data Licensors, the company that publishes this page, is one of the buyers. Names are sorted alphabetically.
- AI data companies license records directly and supply them to AI labs: AfterQuery, Data Licensors, G2i, Handshake AI, Labelbox Prism, Mercor, micro1, Nyne, Polyshares, Scale AI, SuperAnnotate and Turing. micro1 alone has committed to spend $1 billion over 12 months on company data (micro1, October 2026).
- Brokers and licensing agents find buyers for you. Some take a percentage of the sale; Endless instead pays you a fixed sum and licenses the records on: Endless, SellMyData, Troveo, Your Data Value.
- Marketplaces list your data for many buyers: Grepped, FileYield, DataVendor.
- Shutdown services sell records as part of closing a company: SimpleClosure. See selling startup data before shutdown.
Our AI data partnership programs guide compares the programs side by side.
Step 1: Do I qualify?
Buyers want years of real work in writing. Their published entry rules differ a lot:
| Buyer | Size or history they ask for |
|---|---|
| Endless | Roughly 30+ employees, several years, English |
| G2i | One app is enough, 1+ year of activity |
| Handshake AI | 20+ full-time staff, 3+ years |
| micro1 | 30+ full-time staff, records mainly in English |
| Nyne | 20+ US W-2 staff, 3+ years |
| Scale AI | 40+ employees preferred, case by case |
| SellMyData | No minimum headcount |
| SuperAnnotate | Any size, records built up over time |
| Your Data Value | 30+ employees |
All from each company's page, October 2026. If you are below one buyer's line, another may still take you.
Step 2: What should I prepare before talking to buyers?
You do not need to export anything yet. Buyers tell you how they want to sample. Prepare a one-page inventory:
- Each system your team uses (for example Gmail or Outlook, Slack or Teams, Google Drive or SharePoint, Jira, HubSpot or Salesforce, QuickBooks, ServiceTitan, GitHub).
- The years of history in each, and roughly how many people used it.
- What you would never include: client files, HR and payroll, health information, privileged legal material, passwords.
- Any customer or supplier contract that limits how you can use records.
Our estimate asks similar questions and gives a range in about a minute.
Step 3: Do I have the right to license these records?
Check three things before any sample moves:
- The company, not an employee, owns the records.
- Client and supplier contracts do not forbid this use.
- You have a plan for personal information about employees and customers.
As Nyne's page puts it, being able to export a record does not by itself give you permission to license it. This is general information, not legal advice; our legal guide covers it in more detail, and a lawyer should review your contracts.
Step 4: How do I get offers?
Talk to more than one buyer and ask each to quote the same scope. Most start with a short call, then review a sample under a confidentiality agreement:
- G2i agrees evaluation terms in writing before any records are shared, then pulls a single snapshot.
- AfterQuery sends a mutual NDA and a questionnaire, and replies within three business days.
- Labelbox Prism says it produces a written quote within 48 hours of mapping your systems.
- Polyshares says an offer comes within one to two weeks of intake.
- Data Licensors reviews a small de-identified sample and sends a written offer, usually within one to two weeks (Data Licensors, October 2026).
Compare offers on price, term, exclusivity, what uses are allowed, who does the de-identification, and when payment is due. Our guide to the data licensing agreement lists each term.
Step 5: What happens when I sign?
The agreement sets the scope, the license terms and the payment. Then:
- Export. Your team exports the agreed records with the buyer's guidance, or grants a limited read-only connection. Scale AI says it anonymizes inside your environment before data is extracted.
- De-identification. The buyer replaces names, emails, phone numbers, addresses and account numbers with placeholders. You review a sample. See how de-identification works.
- Acceptance. Many buyers check the data before paying. micro1's program page gives it 30 days to review; its published master terms give 30 business days to reject data, after which the data counts as accepted (micro1, October 2026). Ask which applies to you.
- Payment. By wire, on the timing in your agreement.
When do I get paid?
| Buyer | Published payment timing |
|---|---|
| Data Licensors | Within 30 days of signing |
| Labelbox Prism | Under 21 days from first call to first wire |
| Mercor | By wire on delivery, typically 2 to 4 weeks after signing |
| micro1 | Up to 30 days to approve, then within 30 days of invoice |
| Polyshares | Net 30 to 60 after approval and delivery |
| SellMyData | Aims for 7 days after a sale closes |
From each company's page, October 2026. Ask whether payment depends on acceptance, and what happens if some data is rejected.
What mistakes should I avoid?
- Deciding under a deadline. Business Insider (October 2026) reported six-figure offers that expire in 48 hours. A fair buyer will give you time to compare.
- Sending raw data before terms. Agree in writing how a sample is stored, used and deleted first.
- Giving broad system access. Prefer an export, or a scoped read-only connection you can revoke.
- Including what you do not own. Client files, a former employer's material, or anything under a contract that forbids sharing.
- Reading a ceiling as an offer. "Up to $4M" is a headline. Ask for a written number on your own scope.
- Skipping your team. Tell employees before records are shared. Data Licensors recommends this and offers a plain-language note owners can send (Data Licensors, October 2026).
- Signing without a lawyer. These are new contracts with unusual terms. micro1's published master terms, for example, grant a perpetual, irrevocable license and say deleting the data does not require deleting models already trained on it (micro1, October 2026).
Can I keep using my data after I license it?
Yes. A license gives the buyer the right to use a copy. You keep the originals and can keep running on them. Whether you can license the same records to a second buyer depends on exclusivity, which varies: Endless says its license is non-exclusive, while Labelbox Prism says its licenses are exclusive by default (buyer pages, October 2026).
