AI data brokers compared

October 23, 2026 · 9 min read

AI labs have mostly used up the public web, so a new set of companies now pays for the data that only exists inside businesses and inside experts’ heads. They don’t all buy the same thing. Some license a company’s records; others pay people by the hour to produce training data. This guide sets out who does what, using each company’s own public statements, so you can tell which one fits what you have.

Two different markets

“Selling data to AI” covers two separate businesses. In the first, a company licenses records it already has: email, tickets, CRM history, procedures and project files. The company is paid once per licence, and the work that created the records is already done. In the second, a platform hires experts such as doctors, lawyers and analysts to write answers, grade model output or build test tasks. Those experts are paid for their time.

If you own a business, the first market is the one that pays you. If you’re a professional looking for flexible work, the second one is.

At a glance

Figures are each company’s own claims or published funding reports, not audited numbers. Sources are listed at the end.

micro1

micro1 started as a platform that recruits experts for AI labs and has since moved into buying company data directly. On October 9, 2026 it said it would spend $1 billion over 12 months acquiring and licensing enterprise operational data, financed by Citi and Hercules Capital. It uses de-identified company records to build environments where AI agents practise real business work, with its messy information and judgment calls. Its partnership page advertises payments from $100K to $2M+ per company, in tiers that start at $100K+ for a qualified partnership and pass $1M for highly unique data. It looks for operationally mature companies with 30+ employees and English-language documentation, with the US first.

Mercor

Mercor connects AI labs with professionals who produce training data and feedback: doctors, lawyers, financial analysts and other specialists, paid by the hour. It raised a $350 million Series C at a $10 billion valuation in October 2025, and was reported in 2026 to be in talks at roughly twice that. It is the clearest example of the expert-labor side of the market. A company looking to license its archive won’t find a buyer there, but its employees might find paid work.

Grepped

Grepped buys data from companies through a short online estimate followed by a review. It publishes ranges of up to $5 million for one company’s data, says payouts can arrive within 7 days of a sale, and pays people who refer companies 10% of the deal, up to $100K. It also runs a public listing of datasets available to buyers.

Maryah

Maryah runs a marketplace for business data. It says it has paid more than $20 million to companies, works with over 100 active buyers, and licenses the average dataset several times over, which matters for sellers: a non-exclusive licence can earn more than once.

The expert-labor platforms

Alongside Mercor, companies such as Scale AI and Surge AI built large businesses supplying human-labelled and expert-written data to AI labs. They pay people for tasks rather than companies for archives, so they belong in the second market above.

Where 6/8 Labs fits

6/8 Labs works on the company-data side. We help companies license the records they already have, such as email, documents, CRM history, support tickets and procedures, to AI labs, with identifying details replaced before any lab sees them. Companies with 25 to 50 employees take a direct route; companies with 50 to 100+ go through our own review. Anyone who refers a company earns 6% to 10% of the deal, up to $100K, paid on top of what the company receives. See how we de-identify data or estimate what yours is worth.

Questions to ask any buyer

• What exactly are they licensing? Which systems, date ranges and record types, and what stays out.
• Is it exclusive? A non-exclusive licence lets the same data earn again. Exclusivity should pay more and last a set time.
• How is it de-identified, and by whom? Ask whether identities are replaced consistently or simply blacked out, and when raw files are deleted.
• When are you paid? Get the acceptance window and the payout date in writing.
• Can it be resold? Check whether sub-licensing is allowed, and to whom.
• Who referred you, and what are they paid? A referral fee should come on top of your price, not out of it.

Sources

• micro1: $1B commitment to enterprise data acquisition and data partnerships page
• Sacra: Mercor company profile and TechCrunch coverage of Mercor’s funding
• Grepped
• Maryah: business data

6/8 Labs is not affiliated with the companies above, and they have not reviewed this page. Details change quickly; check each company’s own site before relying on them.