Who buys company data for AI training?

Q: How do I find buyers looking for my kind of data?

Browse data requests and wanted posts on Highford, or list your dataset for free so buyers can find you and request an intro.

Your data is training someone's model. You're just not getting paid.

Companies hold the records AI labs say they need: support tickets, SOPs, CRM history, code, chat, and email. This guide sorts the buyers into types, using public sources only, and shows how to reach them.

What this page is

A plain map of who licenses company data for AI, and how a license usually reaches them. It describes each buyer type from its own public pages or published reporting. It isn't legal advice, and it doesn't rank or endorse any buyer.

Highford is a directory and double opt-in intro tool. The license is between you and the buyer.

Who it's for

Company owners and operators wondering whether their records are licensable.
Companies winding down with archives (Slack, email, code) to consider.
Business development and partnerships teams at data vendors and labs mapping the market.

The buyer types

1. AI labs (direct)

Large model developers license data directly. Most publicly reported licenses so far have been with large content holders, such as publishers and platforms, rather than ordinary operating companies. Reuters and other outlets have documented these licenses. 1 2 Some labs run their own data partnership intake. One review of lab pages notes that OpenAI's page says what it will not take, including personal and third-party information. 3

Fit: large, distinctive libraries. Smaller companies more often start with a vendor or program. 2 4

2. Data vendors and data labs

Firms that assemble training data for labs now also license operating records from companies. Examples named on their own pages or in comparison guides include:

micro1 runs enterprise data partnerships. Its page states it looks for operationally mature companies with 30 or more employees, prioritizing the US. 5
Mercor describes licensing, packaging, and de-identifying workflow data from connected business tools. 6
Scale AI, Appen, Handshake AI, Turing, Labelbox (Prism), Centific, Nyne, and Mode are each listed as licensing company workflow or operating records, with eligibility stated on their own pages. 4 7

Fit: companies with years of linked records (tickets plus resolutions, project history, code with reviews). Most state minimum headcount or history requirements. 4 7

3. Licensing programs and marketplaces

These intermediaries sign data owners and license onward, often for a revenue share. Examples include Troveo (video, audio, gaming, robotics, and enterprise workflow data), Protege (with depth in healthcare), and Defined.ai (speech and multimodal). 8 9 Seller-side services have multiplied in 2026. One comparison site notes that many were registered this year and advises checking each one before sharing anything. 4

Fit: owners who want someone else to package and clear the data, and who accept a share and often exclusivity. 8

4. Wind-down and shutdown services

For closing companies, shutdown services can include licensing archives as part of a wind-down. 4 Reporting in 2026 covered AI labs licensing records from shut-down startups. 10

What every buyer checks

Across public guides, buyers ask the same things first:

Do you own the records, and do customer contracts, privacy notices, and software terms allow licensing?
What's excluded (HR, legal, health, and third-party confidential material)?
How will it be de-identified?
What's the permitted use, term, exclusivity, and deletion? 11 12

Don't send samples before that review. A sample can itself be a disclosure. 12

How Highford helps

Highford is a searchable database and matching marketplace for finding these counterparties.

See who's asking. Browse data requests and wanted posts.
List what you hold, for free. The card states data type, size, rights, and license terms. The files stay with you. For a company archive, start at List a company archive.
Connect through a double opt-in intro. Unlock a contact, preview the intro, and confirm the send. The other side accepts or declines. Nothing goes out automatically.
Highford doesn't require exclusivity, so you can talk to several buyer types before choosing.

There's a free tier. See pricing.

FAQ

Who buys company data for AI training?

Three main groups: AI labs licensing directly, mostly from large content holders so far; data vendors and data labs such as Scale AI, Appen, micro1, and Mercor, which license operating records from companies; and licensing programs and marketplaces such as Troveo and Protege, which license onward for a share. Wind-down services also handle archives from closing companies.

Can a small company license data directly to an AI lab?

Publicly reported direct licenses have mostly involved large content holders. Smaller companies more often work with a data vendor or licensing program, or find a buyer's posted request.

What do buyers want?

Linked records that show how work gets done, such as tickets with resolutions, SOPs, project history, and code with reviews. They also want clear rights, exclusions, and a de-identification plan.

Does Highford buy or broker company data?

Highford does not sell or broker data, and takes no cut. A license is between the data owner and the buyer.

How do I find buyers looking for my kind of data?

Browse data requests and wanted posts, or list your dataset for free so buyers can find you and request an intro.

Sources

Machine view of /who-buys-company-data

This page, for an agent

The same page as markdown, plus the API and MCP calls that belong to it.

API

Search listings

GET /api/v1/listings?q=support+ticket+archive

Skill file

GET /skill.md

Create a free account

POST /api/v1/signup
Content-Type: application/json

{"email":"ada@example.com"}

api_key and token come back immediately. Search works with that key. Unlocks, intros, and credits wait until the sign-in link is opened. POST /api/v1/verify with {"email":"ada@example.com","code":"<6 digits>"} still confirms the address.
A new account starts with 10 credits. Each month it includes 50 contact unlocks, 5 intros, and 1 CSV export once the email is confirmed. Summary search stays free.

MCP

MCP server

{
  "mcpServers": {
    "highford": {
      "url": "/mcp"
    }
  }
}

Start here

search_deals {"q":"support ticket archive"}

JSON

{
  "site": "Highford",
  "page": "/who-buys-company-data",
  "markdown": "/who-buys-company-data?format=md",
  "api": [
    {
      "label": "Search listings",
      "request": "GET /api/v1/listings?q=support+ticket+archive"
    },
    {
      "label": "Skill file",
      "request": "GET /skill.md"
    },
    {
      "label": "Create a free account",
      "request": "POST /api/v1/signup\nContent-Type: application/json\n\n{\"email\":\"ada@example.com\"}\n\napi_key and token come back immediately. Search works with that key. Unlocks, intros, and credits wait until the sign-in link is opened. POST /api/v1/verify with {\"email\":\"ada@example.com\",\"code\":\"<6 digits>\"} still confirms the address.\nA new account starts with 10 credits. Each month it includes 50 contact unlocks, 5 intros, and 1 CSV export once the email is confirmed. Summary search stays free."
    }
  ],
  "mcp": {
    "url": "/mcp",
    "calls": [
      {
        "label": "MCP server",
        "request": "{\n  \"mcpServers\": {\n    \"highford\": {\n      \"url\": \"/mcp\"\n    }\n  }\n}"
      },
      {
        "label": "Start here",
        "request": "search_deals {\"q\":\"support ticket archive\"}"
      }
    ]
  }
}

Markdown

Reading this page…