Choosing a Buyer for Your Data: Labs, Vendors, Marketplaces and Brokers

Choosing a buyer for company data means choosing who signs the license, who ends up holding the data and who is paid out of the deal, and those three are often different companies. The routes are a frontier lab, a company data program, a dataset vendor, a marketplace or an intermediary, and no route is endorsed here.

Each choice is one step of AI training data licensing. micro1 states a perpetual license to itself and its affiliates, while Mercor states that no AI lab buyer or third-party vendor ever sees the raw data (checked 2026-11-09).

DataPacts Editorial Team: every term on this page is dated and sourced, see how we source, date and check every figure.

What Choosing a Buyer Decides Before Any Price Is Offered

The choice of buyer fixes the license holder, the final recipients of the data and the fee structure before any offer arrives, so it comes before price comparison.

The micro1 Data License Master Terms (updated 2026-11-08) grant a worldwide, royalty-free, fully paid-up, nonexclusive, irrevocable, perpetual license to micro1 and affiliates, sublicensable through multiple tiers. micro1's program page (checked 2026-11-09) says the data is licensed "exclusively to labs and enterprises based in the U.S. and allied countries", which limits the recipients and does not make the grant exclusive.

Seven Kinds of Counterparty for Company Data

Company data reaches AI developers through seven kinds of counterparty, and only some of them train models themselves. The directory of who buys AI training data lists every party with its check date.

Table: the seven kinds of counterparty, as of 2026-11-09. Intermediaries handle the data and do not acquire it.

Frontier AI Labs: Few Direct Doors for Operating Companies

Frontier labs train the models, but their disclosed licenses are mostly with platforms and publishers, and few publish a door for an operating company's records. Reddit's S-1 of 2024-02-22 discloses USD 203.0 million in aggregate for licenses with counterparties it does not name, so the figure does not show what any lab paid.

Anthropic and Apple publish crawler opt-outs only (ClaudeBot and Applebot-Extended, checked 2026-11-09), and neither lists a paid seller program. On 2026-08-14 (Dkt. 1463), Google LLC is the successful bidder at USD 10,000,000 for Spirit Airlines' de-identified data in Bankr. S.D.N.Y. 25-11897, a bid not approved as of 2026-11-09 (hearing 2026-11-17, Dkt. 1779). See how frontier AI labs acquire data lab by lab.

Company Data Programs and RL-Environment Builders

Company data programs run by human-data companies are the most common published route for operating companies, and they license the records to themselves before passing them on. Each program's own page states the following (checked 2026-11-09 unless dated otherwise).

• Handshake AI: license terms are not published.
• Mercor: license terms are not stated on its page.
• micro1: a perpetual, sublicensable grant to micro1 and affiliates (terms of 2026-11-08), for companies with 30 or more full-time employees.
• Turing Project Lazarus: a non-exclusive license to sell, with ownership not changing hands.

RL-environment builders such as Collinear AI, pre.dev and Patronus AI take private codebases, git histories, tickets and business-app records. The overview of company data partnership programs compares every program on record.

Dataset Vendors and Marketplaces: Who Stays the Licensor

Vendors and marketplaces differ on one point that matters to a seller: whether the seller stays the licensor or licenses the data to the platform. Protege takes a worldwide, royalty-free, non-exclusive license and pays a revenue share "with each use", and its licensees keep access for 12 months after termination (checked 2026-11-09).

Defined.ai states no exclusivity and a 12-month non-circumvention period after the term. Troveo states that it licenses on the seller's behalf and the seller keeps ownership; its own pages show two claims, "$20M+ Paid to Licensors" and "$50M+ paid to creators" (checked 2026-11-09).

AWS Data Exchange charges a 3 % listing fee on public offers from 2024-01-05, and Opendatabay a commission sliding from 30 % to 5 % (fees page of 2026-10-06). Read AI training data marketplaces and their fees for the comparison.

Brokers and Sell-Side Advisers Place Data and Do Not Acquire It

Brokers and sell-side advisers find the buyer and never become the licensee, and they differ mainly in who pays them. Regulr AI, Inc. states it is free to the data owner and that the buyer pays a fee when a license closes. DataSupply.ai states a 20 % success fee paid by the seller from an accepted offer, with the data going directly to the lab (both checked 2026-11-09). Endless differs: it takes a license from the company and licenses the data on to labs, with no commission on the seller's side (checked 2026-11-10).

Six Criteria for Choosing Between Buyers

Six criteria separate buyers before any offer: data fit, the license holder, exclusivity, payment structure, de-identification and track record.

Table: the six criteria with the question to put to each buyer, as of 2026-11-09.

Fit Between Your Data and What the Buyer Takes

A buyer is worth approaching only if it takes the data type and systems the company holds. Mercor lists 25 industry listings (checked 2026-11-09), and Turing names legacy stacks such as Lotus Notes and Novell.

Pangea takes company data on a non-exclusive license for a set use and term. For formats in demand, see the data types AI companies buy.

Who Holds the License and Who Receives the Data

Some counterparties take the license themselves and sublicense it onward, while others sell on the seller's behalf and leave ownership where it was. micro1 is the licensee under its master terms.

Turing and Troveo state that ownership does not change hands, and DataSupply.ai sends the data directly to the lab. Handshake AI promises a review of who may receive the data and how it may be used but publishes no terms (checked 2026-11-09).

Exclusivity and Freedom to License Elsewhere

Exclusivity decides whether choosing one buyer closes the door on the others, and the published terms run from exclusive by default to no exclusivity at all (checked 2026-11-09 unless dated).

• Labelbox Prism: exclusive by default, with the seller keeping ownership.
• Turing Project Lazarus: "For the first 90 days after you sign, Turing is the only one selling your record."
• micro1: a nonexclusive grant plus a no-prior-grant warranty in s.5.1 of its terms of 2026-11-08.
• Defined.ai: no exclusivity under s.3.5.

The exclusivity clause of an AI data licensing agreement is the line to read before signing.

Payment Structure: Fixed Fee, Upfront Plus Share, or Share Only

Buyers pay in three structures, and the structure decides whether the seller is paid once or depends on later sales.

1. Fixed payment: Endless states a fixed payment agreed before anything is shared.
2. Upfront fee plus revenue share: Labelbox Prism states a licensing fee at signing plus a share of downstream use.
3. Revenue share only: Turing states nothing upfront and payment every time it sells; Protege pays a share with each use and Troveo pays on every sale.

Mercor states a wire on delivery of the anonymized dataset, typically 2 to 4 weeks after signing (checked 2026-11-09). Triggers and clawbacks are covered in when and how AI data buyers pay.

Who De-identifies the Data, Where and at Whose Cost

Buyers split on whether they de-identify the data themselves, inside the company's systems, and at their own cost (all checked 2026-11-09).

• Mercor: covers extraction, anonymization and transfer.
• Scale AI: anonymizes inside the company's environment before extraction.
• micro1: states it de-identifies and transforms sensitive information itself, so companies need not anonymize before transfer.
• Turing: covers de-identification, cleanup and packaging costs.

The standards behind the term are set out under PII removal and de-identification.

Identity, Track Record and Corporate Changes

A buyer's legal entity, its incidents and its corporate changes tell a seller who will actually hold the license years later. The entities on record are Mercor.io Corporation, Stryder Corp. d/b/a Handshake, micro1 Inc. and Turing Enterprises, Inc.

The California Attorney General's report lists unauthorized access to Mercor on 2026-03-24 and 2026-03-30. The class actions are consolidated as In re Mercor.io Corporation Data Breach Litigation, N.D. Cal. Lead Case No. 3:26-cv-02821-WHO (consolidated 2026-05-19); a stipulation to stay pending mediation was filed as Dkt. 44 on 2026-07-21, and there is no ruling on the merits as of 2026-11-09.

Corporate changes also move the counterparty. EXL completed its iMerit acquisition on 2026-07-31, Cloudflare announced on 2026-01-15 that Human Native joined it (price not disclosed), and the Getty Images and Shutterstock merger agreement was terminated on 2026-07-07, so the two are not merged. Getty Images filed chapter 11 (Bankr. D. Del. 26-11643 (JKS), petition 2026-10-21, prepackaged plan, solicitation pending as of that filing). For unsolicited approaches, see the red flags and scam checks.

Comparing Written Terms Before Committing to One Buyer

Choosing a buyer works best on written terms from more than one candidate, because several programs publish no license terms at all. Handshake AI publishes none, Scale AI does not publish exclusivity, term, deletion or retraining terms, and Mercor states none on its page (checked 2026-11-09).

Four questions go to each candidate in writing:

1. Who is the licensee?
2. Who receives the data?
3. What are the exclusivity and prior-grant terms?
4. What is the payment structure and its trigger?

No seller-confirmed amounts are public. Once offers arrive, evaluating an offer for your data sets out the comparison.

Benefits, Risks, Downsides and Alternatives of Each Route, Including Not Licensing

Each route trades control against effort: dealing with a lab or program directly keeps the chain short, while an intermediary or marketplace adds reach and adds a party to the chain.

Benefits

• A program covers de-identification cost (Mercor, Turing).
• An intermediary handles the search for a buyer.
• A marketplace such as Troveo leaves the seller as owner and licensor.

Risks

• A sublicensable, perpetual grant cannot be recalled.
• A counterparty can change owner or enter chapter 11.
• Breach litigation against a counterparty is on the public record.

Downsides

• Exclusive by default or a sole-seller window closes other routes.
• A seller-paid success fee reduces the amount received.
• Revenue share pays only on later sales.

Alternatives

• Another kind of counterparty from the seven.
• Waiting until terms are published.
• Keeping the data and not licensing at all.

The decision on whether to sell or license your data at all comes first for many companies.

Where Choosing a Buyer Sits in the Process to Sell or License Company Data

Choosing a buyer follows valuation and comes before the data manifest, due diligence and negotiation. The full sequence to sell or license company data to AI companies runs in these steps:

1. Previous step: how much company data is worth.
2. This step: choosing a buyer.
3. Next steps: the data manifest, due diligence in an AI data deal and negotiating an AI data licensing deal.

Related: selling direct or through a broker, running several AI data buyers in parallel, dataset vendors and data labs, AI data licensing brokers and sell-side advisers, enterprises buying data to train their own models and data buyer incidents.

Choosing a Buyer for a Closed or Insolvent Company

A closed company chooses among buyers that accept wound-down companies, and in insolvency the trustee or administrator, often through a court-supervised auction, makes the choice. Turing and Labelbox Prism accept wound-down companies, micro1 states a defunct-company program of USD 100K to 3M+, and SimpleClosure takes a percentage only if assets are sold or licensed (all checked 2026-11-09). See selling data from a defunct company. The Spirit bid above is not approved as of 2026-11-09; see data assets in insolvency.

Buyer Pages Cited on This Page

The program terms above come from each party's own pages, checked November 9, 2026.

• Handshake AI: Handshake AI data buyer profile
• Mercor: Mercor data buyer profile
• micro1: micro1 data buyer profile
• AI data partnership payouts and program terms compared
• Also cited, without a profile page yet: Defined.ai, DataSupply.ai, Endless, Labelbox Prism, Pangea, Protege, Regulr, Scale AI, SimpleClosure, Troveo, Turing Project Lazarus

Frequently Asked Questions

Who buys company data for AI training?

Frontier labs, company data programs of human-data companies, RL-environment builders, dataset vendors, marketplaces and enterprises training their own models license company data, and brokers place it with them. The directory linked above lists each party.

Is a data broker a buyer?

No; a broker or sell-side adviser finds the buyer and is paid by the buyer or the seller, and the license runs to someone else. Regulr states the buyer pays it a fee, while DataSupply.ai states a 20 % success fee paid by the seller (checked 2026-11-09).

Should I choose the buyer with the highest advertised payout?

No; advertised payouts are each buyer's own claims, some contradict themselves, and no seller-confirmed amounts are public. micro1 states USD 100K to 500K, 500K to 3M and 3M+ tiers, and Handshake AI shows "Earn $100K to $4M+" next to an estimator saying "up to $100,000" (all checked 2026-11-09).