AI Data Companies That Pay for Company Data (2026)

In short (13 November 2026)

As of 13 November 2026, this directory lists 36 companies in 6 groups; 13 buy an operating company's records, and 9 of those publish a price, from $10K to $5M. Each row was read on the company's own page on 10 November 2026.

Which AI data companies pay companies for data?

An "AI data company" can mean three different things: a frontier lab that trains models, a vendor that supplies those labs with experts and training data, or a program that buys records from companies. Only the last group matters to an owner who wants to sell records, and most of it is run by the vendors in the middle.

The tables below list the programs that publicly invite companies to license or sell data and say what they pay. Each row was read on the company's own page on 10 November 2026. Where a company publishes no number, the table says so instead of guessing. We left out sites that collect sellers' details for other programs without buying anything themselves (more on how to spot them below), and the individual-contributor apps that pay people rather than companies.

If you only need the five best-known company-record programs side by side, see micro1 vs Mercor vs Handshake vs Turing. For a head-to-head on the two most searched, see Mercor vs micro1.

Company records: the direct buyers (companies like Mercor)

These programs buy the working records of an operating company: chat, email, documents, tickets, CRM and finance history. They are the closest "companies like Mercor" for a seller. Mercor publishes no range on its data page; it says compensation "is based on data volume, the tools you connect, and the depth of your data" and that it "covers all costs associated with data extraction, anonymization, and transfer". micro1 and Handshake AI publish the highest ranges among the large vendors.

Source: micro1: company data partnerships

Code, agent and RL-environment data

Code buyers pay for private repositories with their full history: commits, pull requests, tickets and tests. A newer group builds reinforcement-learning environments, simulated workplaces where AI agents practise real tasks, and needs company records to make them realistic. Ooak Data says "We turn real company data into RL environments where AI agents learn to work in the real world." Turing also buys company records and code; per our Project Lazarus guide, it pays a share of each sale under non-exclusive licenses. Turing's program page renders only in a browser, so confirm the current terms with Turing before relying on them.

Source: Pangea: code licensing for companies

Wind-down buyers: data from closing or closed companies

A company that is shutting down can still sell its records, and several programs say so. AfterQuery says "Operating, winding-down and closed companies are all welcome," and Appen asks winding-down companies to flag it for a faster reply. Three programs are built for this case.

Prices are lower than for operating companies. MIT Technology Review reported that SimpleClosure's Asset Hub "has handled over 150 deals and expects to return over $10 million to startups by the end of 2026", and that one founder "netted roughly $25,000 for his code base". The full routes are in selling data after closing.

Source: MIT Technology Review: AI labs buying data from dead startups (Oct 2026)

Health data programs

Health buyers mostly want hospitals, health systems and imaging centers, and they pay a share of revenue each time de-identified data is used rather than a lump sum. Gradient Health gives a worked example in which one partner "receives $240k from this one deal": the deal was worth $3M, that partner supplied 40% of the studies, and the "20% (typical revenue share for data partners)" applied to its share. The health data licensing comparison covers eligibility and de-identification rules in depth.

Source: Gradient Health: become a data partner

Video, audio and physical-AI programs

Video and audio buyers license archives that a company owns outright. Physical-AI programs are different: they pay a business to have staff record the hands-on work they already do, so robots can learn from it. Rates are per month or per hour, and none promises a fixed volume. Shift says plainly: "This is not a promise that a fixed number of recording hours will be available." See business video programs compared for the full set.

Source: Troveo: for content owners

Data marketplaces

A marketplace does not buy your data; it lists it and takes a cut when a buyer does. That suits a company with a ready, documented dataset more than one with raw records. Data marketplaces compared covers listing rules and fees.

Source: Datarade: apply as a data provider

Well-known AI data companies that do not publish a company program

Some of the most searched AI data companies hire experts or build datasets but do not publish a program that pays companies for records. That does not mean they never buy; it means there is no public door, so any approach in their name deserves a check of who is really making it.

• Surge AI: according to MIT Technology Review, which reports that it buys data to build "reinforcement-learning gyms". We found no public seller page. See our Surge AI review.
• Datacurve: describes itself as "pioneering the research and data collection infrastructure" for frontier AI (Datacurve); its homepage offers no company program. See our Datacurve review.
• Mechanize: says "We build environments and evals for frontier coding agents" (Mechanize); no seller program is published. See our Mechanize review.
• Toloka: its site links to "Become a data partner" (Toloka) but publishes no terms for companies. See our Toloka review.
• Kled: pays individuals for data "sourced from verified contributors" (Kled), not companies for their records. See our Kled review.
• Sepal AI: see our Sepal AI review for what it does and whether it buys from companies.

How the market is layered: labs buy through vendors

Frontier labs rarely buy from a mid-sized company directly. They buy from vendors that recruit experts, collect records and package them. TechCrunch describes AfterQuery as one of "the new crop of startups following in the footsteps of Mercor and Scale" (TechCrunch). The labs buy direct mainly when one seller holds a very large archive: in August 2026 "Google won a bankruptcy auction for the data, agreeing to pay $10 million" for Spirit Airlines' records, and "Mercor was the next-highest bidder, at $7.5 million" (Axios).

Below the vendors sit companies that buy records and resell them, such as Polyshares, Mode and Replay; marketplaces that list data for a commission; and introducers. Introducers exist because buyers pay them for introductions. Each layer between you and the lab can take a share, so ask who the end buyer is and who is being paid.

Source: Axios: Google wins Spirit Airlines data auction (Aug 2026)

How to pick a buyer for your data type

Start from what you hold, not from the biggest headline. The table matches common record types to the programs that published demand for them on 10 November 2026.

How to tell a buyer from an intermediary

Not every site that invites you to sell data is a buyer. Some pass your details to one of the programs above, which may pay them an introducer reward. That is not wrong in itself, and a buyer may pay Origin too, but you should know which you are dealing with before you share anything. The test is substance, not the headline number or the form: does the site name the buyer that will sign, publish its terms and disclose how it is paid?

• A real buyer names itself as the counterparty and signs the license. Polyshares, for example, states "Counterparty Polyshares. We are the buyer." A site that cannot say who will sign is not the buyer.
• A real buyer can describe deals it has completed. SimpleClosure's checklist starts with "Completed deals: whether it has bought a dataset from another startup and paid for it" (SimpleClosure).
• A real buyer publishes how it handles data, who de-identifies it and when payment is due. An intermediary should say which buyers it works with and how it is paid. A site that does neither is asking you to trust it blind.
• A headline is not an offer. In SimpleClosure's words, "An ad offering up to $5 million doesn't tell you what this buyer will pay for your assets".
• Check the company behind the site: its registration, founders, funding and how long it has existed. Our buyer vetting checklist sets out each step.

Healthcare claims and prior authorization data

Physician groups and billing companies: see the guide to healthcare claims, denial and prior authorization data for who signs and what HIPAA requires.

Next step

The right buyer depends on your records, your size and the terms you will accept, and published ranges differ widely for the same company. Compare at least two programs before you sign anything.

Not sure whether a buyer or an offer is sound? Origin checks who the buyer is, what the license allows and how the price compares with published deals before anything is signed. Get a free estimate and a second opinion.

Origin Data Partners works with companies that own data and helps get it bought. We are independent of the services on this page, and a buyer may pay us when a deal closes (how Origin is paid). See what your data is worth.