Companies that buy business data for AI, compared for UK sellers

Which AI data buyers accept UK companies?

Five buyers say on their own pages that they take UK sellers. Most of the others have not stated a position, and silence is not a refusal. We ask each buyer to confirm in writing that it accepts UK companies before we introduce you, and record the answer here.

All rows were read on 12 November 2026. "Not stated" means the buyer has not stated a position on UK companies on its own pages.

The confirmation column changes only when a buyer has answered us in writing. How many UK businesses fall inside these size floors is set out in our figures on the UK company data licensing market.

Pay, timing, rights and our fee

Pay figures are each buyer's marketing ranges, in US dollars, not offers. Pewter Data, which publishes a calculator, warns that "some receive no offer once their records are reviewed"³¹. How our fee works, and why we disclose it before you sign, is on our fees page. What your own records could fetch is the subject of our valuation guide.

What does each buyer want and pay?

Each buyer wants years of ordinary operating records that show how work gets done, but they differ on structure, size and rights. Each summary below uses the buyer's own words where possible.

micro1

micro1 wants SOPs, internal communications, project histories, CRM records and support tickets from established companies, "primarily in English"¹. It committed USD 1 billion over 12 months to company data in October 2026³. It licenses only to "labs and enterprises based in the U.S. and allied countries"¹. Companies must apply through the referrer's link; manual introductions "do not replace the official application process"².

Scale AI

Scale AI pitches a continuing partnership rather than a one-off sale, with recurring revenue possible⁵. It takes records from Slack, Teams, email, Jira, Salesforce, HubSpot, Zendesk and data warehouses, and says it anonymises "inside your environment, before any data is extracted"⁵. It pays one referral per deal, only if a partnership is signed⁵.

Mercor

Mercor connects to over 100 tools and says "the more tools you share, the higher your payout will be"⁶. Its platform extracts and de-identifies the data, it covers all costs, and it pays by wire when the anonymised dataset reaches the buyer⁶. Its target industries include legal, healthcare, finance and insurance, real estate and logistics⁶.

Mode

Mode buys "an agreed copy" of email, documents, chat, CRM and billing records and resells it after de-identification⁹. Its site sets thresholds of 30 staff, 10 for accounting, finance and medical firms, and 6 for law firms⁹, while its referral page sets 10, 15 and 50 for referred companies¹⁰. Expect about three months to payment⁹.

Polyshares

Polyshares is a principal buyer: "We are the buyer. No marketplace, no listing, no auction"¹². It makes an offer within one to two weeks and charges sellers nothing¹². Its licence "does not cover what you hold on behalf of your own clients"¹². Its four published deals, from USD 110,000 to 450,000, are company claims¹³.

Ventra

Ventra is an intermediary rather than an end buyer: it says it brings "your data to multiple buyers" and returns the best offer, typically in 3 to 7 days¹⁵. It also seeks codebases and CAD libraries¹⁵.

Labelbox Prism

Prism buys work documents and operational data from Salesforce, Jira, Slack, SAP and CAD systems¹⁶. Licences are "exclusive by default", every licence has a 30-day deletion clause, and models already trained "cannot be unlearned"¹⁷. It says distressed and wound-down companies are "some of our most active sellers"¹⁷.

Handshake AI

Handshake AI advertises "Earn $100K to $4M+" and lists finance, legal, healthcare, logistics, manufacturing, real estate and software among its industries¹⁸. Its referral page looks for 20 or more full-time staff and three years' trading¹⁹.

Turing

Turing's Project Lazarus takes active, winding-down and wound-down companies²⁰. It licenses records to labs on a revenue share, paying "each time" a record sells, with a 90-day period in which only Turing sells²⁰. A third-party listing still describes "a single payment, up to $1M"²¹; Turing's own page no longer does.

SourceX

SourceX licenses directly and usually pays a one-time fee per dataset²². Its terms grant it a "perpetual, irrevocable, transferable and sublicensable" licence to the de-identified dataset, with no obligation to "untrain" models²³. Its agreement records that re-identification "may still be possible"²².

Endless Data

Endless Data agrees a fixed payment before anything is shared, then licenses the data on to "AI labs such as micro1, Mercor and Handshake"²⁶. Its licence is non-exclusive and excludes client books, privileged client matter content and health information²⁶.

Grepped

Grepped values, prepares and lists company corpora on a marketplace, paying the seller "when a lab buys"; each listing shows "whether it has been sold before"²⁷.

Troveo

Troveo licenses data on the seller's behalf and pays a share of revenue "on every sale, not just the first"²⁸.

Which programmes are US-only?

Two programmes we checked are closed to UK companies, and one widely seen eligibility line is not from the buyer at all.

• Cognor Labs requires "a US-based company with 20+ W-2 employees", and its programme for closed companies is "Open to US-based companies"^29,30.
• Pewter Data works "for owners of established US firms", brings one unnamed licensing partner's offer, and charges sellers 10% of the licence price on completion³¹.
• SimpleClosure AssetHub sells closing startups' code and workspace data, within a shutdown service built around US processes such as closing state payroll tax accounts³². Its page does not say whether non-US companies can use it.
• DataValueCheck shows "US, CA & EU only" beside micro1's name, but its footer says it is "operated by an affiliate marketer. We are not micro1"⁴. micro1's own FAQ names the UK¹.

Data labs, AI labs and marketplaces: who are you actually licensing to?

You will usually sign with a data company, not with the lab that trains the model. That company may prepare your records, build them into training environments and license them on, so your licence must say who can receive them.

MIT Technology Review found that founders selling closed companies' data had "little to no visibility into who is buying the data"³³. Ask every buyer to name the classes of end recipient and to limit onward transfer. Our licence terms guide lists the clauses, and what AI companies buy explains how they use records.

Buyers also differ on where personal data is removed. Scale AI anonymises inside your environment before extraction⁵; micro1 says a company need not anonymise before transfer because micro1 de-identifies it¹; Mercor says its platform extracts and de-identifies the data⁶. We prefer personal data to be removed before records leave your systems. Where a buyer removes it after transfer, we tell you in writing first and your own solicitor confirms the transfer route; see how personal data is removed.

Which buyers ask for exclusivity or lasting rights?

Several do, and it matters more than the headline price. An exclusive licence stops you licensing the same records elsewhere; a perpetual licence means you cannot take them back.

• Exclusive by default: Labelbox Prism¹⁷; SourceX, for AI training, for a term set per deal²².
• Perpetual: SourceX's rights over the de-identified dataset survive termination²³; Polyshares' published deals are perpetual and exclusive¹³.
• Time-limited: Turing sells alone for 90 days, then you can renew or opt out²⁰; Troveo's exclusivity is "time-limited"²⁸.
• Non-exclusive: Endless Data²⁶.
• Not published: micro1, Scale AI, Mercor, Handshake AI.

Exclusivity has a price. In the related market for AI training environments, two founders told Epoch AI that exclusive deals cost roughly 4 to 5 times as much as non-exclusive ones³⁵. If a buyer wants exclusivity, ask what it pays for it, and limit the term and the field.

What changed this month?

This is our first edition, so it records the changes we found up to 12 November 2026. Future editions will log each month's changes here.

How we check these facts

We read each buyer's own pages and quote them, label anything from a third party, and date every row. Every fact in this edition comes from published pages, not from private conversations with buyers.

1. Primary sources first. Facts come from each buyer's programme page, FAQ and referral terms. Readings by others, such as Onfolio, aitrainer.work and TechRound, are labelled^21,25,34.
2. Pay is quoted, not predicted. Ranges are the buyers' marketing figures in US dollars, not offers.
3. Our interest is shown. We could earn introduction or referral fees from some buyers in the table, never from sellers. That is why the reward column exists and why our fees page explains the conflict.
4. Monthly review, next by 11 December 2026, with changes logged above.
5. Corrections. Tell us about an error using the contact details on our About page and we will correct and log it.

Names are used to describe each programme; Sleeping Asset is not endorsed by, or a partner of, any buyer listed. To compare agents rather than buyers, read how to choose a data licensing agent, and see how it works for the steps once you choose.

Not sure which buyer fits your records? Book a free, confidential 30-minute valuation call. We reply within one working day.