A data program, a direct licence to an AI lab, or a data marketplace: which route fits your data?
Quick answer, Nov 9, 2026As of 9 November 2026, three routes take company data. A buyer program scopes and de-identifies the records for you and pays on delivery. A direct licence to a lab suits large datasets you fully own that are not already online. A marketplace suits a product you can keep current and support yourself.
Owners usually meet one route, read its number, and treat that as the decision. The route comes first. Each one wants a different kind of data, asks a different question about your rights, and leaves a different amount of work with you after the money moves.
What each route asks of you
| What you are deciding | A buyer program | A direct licence to a lab | A data marketplace |
|---|---|---|---|
| What it is | a company that licenses your operating records and sells them on, connecting to "100+ supported tools" you already run (Mercor) | a partnership with the lab itself, in two published forms: an "Open-Source Archive" dataset anyone can train on, or "Private Datasets" for its own models (OpenAI Data Partnerships, 9 November 2023) | you become a seller, listing a data product subscribers pay for (AWS Data Exchange, providing data products) |
| What data it is built for | the records of how your company works: "Contracts, case files, discovery docs", "Work orders, QA logs, spec sheets" and six more industries (Mercor); "established companies with 30+ full-time employees, mature business operations, and substantial historical operational data" (micro1) | "large-scale datasets that reflect human society and that are not already easily accessible online to the public today", in any modality, "particularly … data that expresses human intention" (OpenAI) | anything that clears the publishing rules, with API, Amazon Redshift and Amazon S3 delivery named as options (AWS Data Exchange) |
| What it asks about rights | you approve each package; Scale AI lists what never leaves as "Employee and customer identities", "Raw systems, credentials, or source access" and "Anything outside your signed agreement" (Scale AI) | everything, before you start: "We are not seeking datasets with sensitive or personal information, or information that belongs to a third party" (OpenAI) | no information "that can be used to identify any person, unless that information is Publicly Available Information", absent a qualified place in the Extended Provider Program (AWS publishing guidelines) |
| Who prepares the data | the buyer: micro1 answers "No, micro1 has processes for de-identifying and transforming sensitive information" (micro1) | the lab offers help, with OCR and speech recognition named, and says "we can work with you to remove" sensitive or third-party information (OpenAI) | you: sensitive categories "must be aggregated or anonymized so that no person in your data product can be identified" (AWS publishing guidelines) |
| What it says about pay | published ranges: $100K to $4M+ (Handshake AI), $100K to over $3M (micro1), $10K–$1M+ as an "illustrative value per data partnership" (Scale AI), and $10k to $1.5M across Mercor's 25 open listings (Mercor listings, read 9 November 2026) | UNKNOWN: the page states no figure, no range and no size screen (OpenAI) | whatever your offer sets, less the marketplace's terms; you may also add private and Bring Your Own Subscription offers (AWS Data Exchange) |
| What you run afterwards | little: Mercor pays "via wire transfer upon delivery of the anonymized dataset to the buyer", "within 2-4 weeks of the signed agreement" (Mercor); micro1 reviews for 30 days, then pays within 30 days of invoicing (micro1) | UNKNOWN as published | a business: "a defined customer support process and support organization", "a means to keep data regularly updated and free of vulnerabilities", and tax and bank details, a W-9 and a US bank account for US entities (AWS, getting started as a provider) |
Every quotation was read on the page it links to on 9 November 2026.
Does a program already describe your company?
Programs publish screens, and the screens decide whether the route is open at all. micro1 states 30 or more full-time employees, documentation in English, and US companies first, then other Western markets. Handshake AI states "20+ full time employees & 3+ years in operation". Scale AI states "40+ employees is our preferred starting point — though we assess every partnership on a case-by-case basis".
Mercor goes further and publishes the situation itself. On 9 November 2026 its listings page carried 25 open listings, each naming an industry, the records wanted and what the company earns, and 23 of them a size band: a law firm of "5–300 attorneys and staff" licensing "complete files from closed matters and completed deals" at $50k to $1.2M, a manufacturer of 20 to 2,000 employees licensing "production, maintenance and inspection records from the shop floor" at $100k to $1.5M, a cybersecurity company at $20k to $750k. Across all 25, the stated company ranges run from $10k to $1.5M (Mercor listings).
Those are advertisements, not offers. What they tell you cheaply is whether anyone is already asking for records like yours. The side-by-side of the four programs is in AI-data buyer programs compared, and what moved this month is in the data-buyer tracker.
Where the three routes really differ
They differ on who carries the rights work. A program takes de-identification on itself and gives you a per-package approval gate. The lab's page says the opposite in one sentence: it is not seeking information "that belongs to a third party", so you settle every customer contract, employee privacy question and inbound licence before the conversation is useful. The marketplace sits between the two and writes its rule as a list, including what you must aggregate or anonymize so "no person in your data product can be identified".
They also differ on what happens in month six. A program ends at delivery and payment. A marketplace listing is a product with subscribers, support obligations within one business day for API products, and a Data Subscription Agreement you write.
Whether any route is open is a rights question first. The checks, in the order a buyer's lawyers read them, are in can you license data your customers created; where data arrived under someone else's terms, start with can you relicense data you licensed.
Where Frontier Desk fits
In the Cybernetic Age a company's operating record is an asset that can be sold, borrowed against, or left in place to raise the price of the company itself, and those three answers compete. Frontier Desk is the company's advisor on which of them fits: which route, which counterparty, which terms, and how a licence reads later in diligence on a raise, a loan or a sale. The data advisory works through whether the records are ready and what to check in an offer. Start with a free first look.
Questions
Which route pays the most? UNKNOWN, and no page answers it. Programs publish ranges for themselves, a marketplace pays whatever subscribers pay, and the lab partnership page names no figure at all. What is comparable is the ask: a program wants records it can de-identify, a lab wants scale and clean title, a marketplace wants a product.
Can I use more than one? The pages do not forbid it, but an exclusive licence to one changes what the others can take, and what a later buyer of the company is buying. Decide exclusivity before you sign the first one.
My records are mostly about customers. Which route is that? Probably a program, if any, because the buyer runs de-identification. Both of the others ask you to arrive with the question already settled: the lab excludes third-party information outright, and the marketplace excludes anything that can identify a person unless it is publicly available. Work through the rights checklist before choosing.
No program lists my industry. Is that a no? No, and the sector lists do not even agree with each other. Mercor names eight target industries on its data page and sorts the 25 open listings into seven categories that only partly overlap them on its listings page (both read 9 November 2026); micro1's stated criteria are about size, history and language rather than sector. A marketplace listing has no sector screen at all, only the publishing rules and the seller requirements above.