An AI data partnership is a paid license: a business lets an AI company use a de-identified copy of its records, such as chat, email, documents and code, to train and test AI models. In October 2026 the main programs are run by AI data companies such as micro1, Handshake AI, Mercor, Scale AI, Labelbox Prism, SuperAnnotate and G2i. Most published ranges start near $100,000 and top out between $1 million and $4 million per company (program pages, October 2026).
On this page
The programs differ most on who they accept, whether the deal is one-time or recurring, exclusivity, and who removes personal information.
Which companies run data partnership programs?
The table lists every program we found with a public page for businesses, sorted by name. Data Licensors, the company that publishes this page, also licenses company records, so it is included. All figures are each company's own published numbers, checked October 2026.
| Program | What they publish about pay | Who they want | Structure |
|---|---|---|---|
| AfterQuery | Not published | Established companies with several systems and a few years of history; operating, winding down or closed | NDA, sample review, then terms |
| Data Licensors | Own price not published; we cite $100K to $1.2M as the market range for 20 to 500 staff, from other companies' deals | 20 to 500 staff | Written offer after a de-identified sample, paid within 30 days of signing |
| G2i | $100K to $2M | One app is enough, a year or more of history | Evaluation terms first, you approve the final dataset |
| Halluminate | Not published | Not stated; it builds AI benchmarks and training environments, starting with financial services | Contact form only |
| Handshake AI | $100K to $4M+ | Typically 20+ full-time staff and 3+ years | Scope, de-identification and terms agreed before data moves |
| Labelbox Prism | About $50K to $1M+ | Not stated | Upfront fee plus revenue share, exclusive by default |
| Mercor | $20K to $1.5M across 24 listings by company type | A size band per listing, for example 5 to 300 staff for law firms | Connects to your tools, paid when processing is complete |
| micro1 | $100K to $500K, $500K to $2M, $2M+ tiers | 30+ full-time staff, mostly English records, US first | 30-day review, paid within 30 days of invoice |
| Polyshares | $100K to $2M and above | Any company with real operating history | Buys directly, offer in 1 to 2 weeks, net 30 to 60 |
| Scale AI | $10K to $1M+, "illustrative" | 40+ employees preferred | Ongoing partnership, per-package approval |
| SuperAnnotate | $100K to $1M+, "indicative" | Any size | Licensed for a set period, you approve each buyer |
Mercor's per-listing ranges are useful for small firms: its accounting firm listing shows $25K to $865K (Mercor referral portal, October 2026).
Brokers such as Troveo ("six figures", revenue share on every sale) and Endless ("up to $4M", a fixed payment agreed up front) run similar programs but sell on to several labs. OpenAI runs an interest form for data partnerships that asks about format, estimated size and your rights to share, with no published price (OpenAI, checked October 2026).
How big are these programs?
Several programs publish their own totals. Labelbox Prism says it has distributed $100M+ and has 1,000+ companies licensing. Mercor says it has paid $50M+ to companies for their data, and its CEO said it has partnered with more than 500 companies (Mercor; RuntimeWire, October 2026). Handshake's referral page says it has paid $250M to partners, without saying how much of that went to data partners. Troveo says it has paid $20M+ to licensors across all its data categories, not only company records (program pages, October 2026). micro1 told the Spirit Airlines bankruptcy court it had completed more than 50 data transactions in the previous 45 days (Fortune, September 2026).
These are each company's own figures. None is audited.
How do the programs differ?
One-time or recurring. Scale AI says it is looking for "partnerships, not transactions" and designs for recurring revenue. Labelbox Prism pays an upfront fee at signing plus a revenue share on later use. Troveo pays on every sale of your dataset. micro1 also licenses agent trajectories, the records produced when AI agents work inside your systems, for $1M to $2M a year with your approval. At the other end, Labelbox's codebase program for closed startups is a one-time transaction (Labelbox, January 2026).
Exclusive or not. Labelbox Prism says its licenses are exclusive by default. Polyshares' published case studies are perpetual exclusive licenses. Endless says its license is non-exclusive, so you can license the same data to others. Exclusive deals should pay more, so ask for both prices.
Who sees your data. SuperAnnotate says you approve each buyer and can block competitors permanently. micro1 says it licenses only to labs and companies in the US and allied countries. Polyshares says it does not disclose which labs take which data. Ask each program to name the class of recipient in the contract.
Who removes personal information. Scale AI says it anonymizes inside your environment before anything is extracted. micro1 says you do not need to anonymize before transfer because it handles de-identification. G2i holds selected material on its own servers and you approve the prepared dataset. See how de-identification works.
How much access they need. Labelbox Prism and Scale AI offer read-only connectors, and Mercor connects to more than 100 tools. G2i pulls one snapshot with no standing connection. Troveo says it uses no OAuth connection to your systems and guides your team through an export. We work from an export your team makes, or a limited read-only connection you can revoke.
What do they all ask for?
The core asks are the same: records of real work over several years, the right to license them, and, for most US-focused programs, records mostly in English. They value records that connect across tools. G2i says records from several apps are worth more when they refer to the same customers and projects, and Labelbox Prism lists cross-system Salesforce, Jira and Slack workflows among its urgent requests (program pages, October 2026).
What they exclude varies, so read each list. AfterQuery says payroll, privileged legal material and credentials are never used. Labelbox Prism excludes HR records and compensation by default. Endless says it never licenses customer, employee or client personal data. Some programs, including Mercor and Scale AI, list healthcare as a target industry, so "no health data" is not a universal rule. We do not accept health records or payment card data.
How should I choose a data partnership?
- Check fit first. If you have fewer than 20 staff, start with programs that set no minimum, such as G2i, SuperAnnotate, Polyshares or Troveo, or a Mercor listing whose size band includes you.
- Ask at least two programs to quote the same scope. Headline ranges overlap, so only written offers can be compared.
- Compare the whole deal. Price, term, exclusivity, permitted uses, deletion, payment timing and what happens if data is rejected. Our licensing agreement guide covers each term.
- Decide between one-time and recurring. A revenue share depends on later sales you cannot see. A fixed payment is certain.
- Check the recipient chain. Ask who will receive the records, whether they can sublicense, and whether they are barred from trying to re-identify anyone.
- Look at speed last. Fast is good, but a 48-hour deadline is a reason to slow down.
For price evidence beyond these ranges, see how much company data is worth and our deal ledger. For a quick read on where you fall, try the estimate.
