One payment or a revenue share: how AI data buyers pay
· Company Data Advisors
The largest group of AI data buyers pays once. A few pay a share of every resale, and one now pays nothing upfront. How each model works and what to ask.
AI data buyers pay for a company's records in one of three ways: a single payment once the data is delivered and accepted, an upfront fee plus a share of later sales, or a share of sales with nothing upfront. About half of the 17 programs we track describe a single payment on delivery or acceptance. As of November 2026, Labelbox Prism, Replay and SuperAnnotate describe paying again when the data is relicensed, and Turing's Project Lazarus pays nothing upfront: it pays when the data sells or when Turing builds on it. The structure matters as much as the size of the offer, because it decides who carries the risk that the data never sells.
What are the ways AI data buyers pay?
AI data buyers use three payment models, and a fourth arrangement (paying again for new data) can sit on top of any of them. Here they are as buyers describe them on their own pages:
- A single payment. The buyer pays an agreed amount once the records are delivered and accepted. You know the number before you export anything. At the buyers that publish a timeline, the money arrives about one to two months after acceptance or invoice.
- An upfront fee plus a revenue share. The buyer pays something at signing or delivery, then pays you a share each time it licenses the data onward.
- A revenue share only. The buyer prepares the data at its own cost and pays you a share of each sale. Nothing is paid upfront.
- Paying again for new data. Some buyers offer recurring deliveries of records you create after signing, each paid under the agreement.
A revenue share, in a data license, is a percentage of what the buyer earns each time it licenses your de-identified records to an AI lab or another customer. It is paid as those sales happen, so its total is unknown when you sign.
Which buyers pay once, and which share revenue?
The largest group of buyers in our index pays once. micro1 says "We pay in full within 30 days of acceptance." Mercor says payment "is issued via wire transfer upon delivery of the anonymized dataset to the buyer." Mode says the written agreement "sets the amount and payment schedule before export." Polyshares pays "net 30 to 60 once the data is approved, anonymized, and delivered." SourceX says it "usually pays a one-time fee," that "there are no royalties or usage fees," and that some agreements set recurring deliveries, each paid as the agreement states.
Three buyers describe paying again when your data is licensed onward. Labelbox Prism's FAQ says "You receive an upfront licensing fee at signing plus a revenue share on downstream use." SuperAnnotate lists "Payment on delivery, and on each relicense." Replay says its offers "are typically structured with a perpetual revenue share," that "You're paid a revenue share of every license," and that you are paid for your historical data first. Replay's page described an upfront cash payment plus the share when we checked it on 2026-11-08; on 2026-11-11 it no longer mentions an upfront payment.
| Buyer | How it says it pays | Paid again later? | Exclusivity, where stated |
|---|---|---|---|
| micro1 | In full within 30 days of acceptance | Recurring for agent trajectories, a separate offer | Not stated |
| Mercor | By wire on delivery of the anonymized dataset | Not stated | Not stated |
| Mode | Amount and schedule set in a written agreement before export | Not stated | Not stated |
| Polyshares | Net 30 to 60 after approval, anonymizing and delivery | Not stated | Its published case studies are perpetual exclusive licenses, though its homepage describes a defined term |
| SourceX | Usually a one-time fee; no royalties or usage fees | Some agreements set recurring deliveries | Exclusive for AI training |
| Centific | On acceptance | Optional quarterly refresh | Not stated |
| Scale | Not stated on its page | Designed with "opportunity for recurring revenue" | Not stated |
| SuperAnnotate | On delivery | On each relicense | Priced partly on exclusivity |
| Labelbox Prism | Upfront fee at signing | Revenue share on downstream use | Exclusive by default |
| Replay | A revenue share of every license; historical data paid first | Perpetual revenue share; new data keeps paying | Not stated |
| Turing (Project Lazarus) | Nothing upfront | A share of every sale, and pay when Turing builds on the records | Non-exclusive licenses to labs; Turing alone sells the record for the first 90 days |
Sources: each buyer's own pages, read 2026-11-08 to 2026-11-11 and summarized on its page in our buyer index. "Not stated" means not on the pages we read; final terms are in each agreement.
What changed at Turing?
Turing's Project Lazarus moved to a revenue share with no upfront payment. When we checked its page on 2026-10-29, Turing described a single payment confirmed after sampling, and we logged the change on 2026-11-08. Its page now says it will "license it to labs, and pay you every time it sells. Nothing upfront." Turing says it prepares the record at its own cost, that "Preparation costs are covered from sales, never charged to you," and that the agreement "sets out exactly how each sale is split before you sign."
Turing pairs the revenue share with non-exclusive licenses. Its licenses "are non-exclusive, so the same record can go to more than one lab." For the first 90 days, it says, Turing is the only one selling your record, and after that you can renew or opt out. It also says it pays you when it builds training tasks, evaluations or RL environments on your records. Its case for the model is a claim about the rest of the market: "A fixed-price buyer pays you once, then keeps reselling your data." Some single-payment buyers do say they pass the data on: Mode describes reselling the copy it buys for AI research and training, and SourceX says its right to sublicense the de-identified dataset is unaffected by your exclusivity promise. Ask each buyer what onward rights it keeps.
Turing's own page states the cost of the model. Under "If nothing sells," it says: "You owe nothing. After 90 days you can opt out and take the record elsewhere." You are not charged, and you are not paid.
How does the payment model connect to exclusivity and resale?
Our view: the payment model tends to follow what the buyer plans to do with the records. A buyer that resells the same records to several labs can share each sale with you, and a buyer that takes exclusivity, or keeps the right to sublicense, prices that right into a single payment. The published terms bear this out only loosely. Turing licenses non-exclusively and shares every sale. SourceX takes exclusivity for AI training and pays once, with no royalties. Labelbox Prism is the mixed case: exclusive by default, with an upfront fee and a revenue share on downstream use.
So read the payment clause and the exclusivity clause together. A revenue share is worth something only if the buyer is free to sell your records again, and a single payment should be judged against the rights it buys. Our guide to what you're agreeing to when you license company data covers exclusivity, term, resale and ownership of the de-identified copy in detail.
What should you ask about a revenue share?
A revenue share is only as good as its definition, and none of the buyers above publish their split. Before you sign one, get written answers to these questions:
- What is the share a percentage of? Gross receipts from each sale, or what is left after the buyer's costs. Turing says its preparation costs are "covered from sales," so ask how much and in what order.
- Who counts as a sale? A license to a lab, a renewal of that license, and a product the buyer builds on your records (Turing says it pays for the last of these) may be treated differently.
- How will you know? Ask how often you get a statement, what it shows, and whether you can check the buyer's figures.
- How long does it last? Replay calls its share perpetual. Ask what happens if the buyer is acquired, stops selling, or shuts down.
- What happens if nothing sells? Ask whether there is a minimum, a date by which unsold records come back to you, and whether you can then license them elsewhere. Turing's answer is an opt-out after 90 days.
- Who approves the buyers? SuperAnnotate says you approve each buyer and can block competitors permanently. A revenue share means more buyers will see your records, so this clause matters more.
Should you sign up for recurring deliveries?
Several buyers describe recurring payments, usually for records you create after signing. Centific says it pays "on acceptance rather than submission," and that "Most partners then move to a quarterly refresh, which is where a one-time sale becomes a recurring line." Scale says it designs its partnerships "so there's opportunity for recurring revenue," with value "scaling with cadence." Replay says new data you produce "continues to pay you," and SourceX says some agreements set recurring deliveries.
A recurring feed changes more than the payment schedule. Frankfurt Kurnit's guide for companies asked to sell data (October 2026) notes that for data already collected, a company is bound by its prior privacy commitments unless it obtains consent, while for data collected going forward, such as a recurring feed, it can update notices prospectively and, where appropriate, obtain opt-in consent. That is a question for your counsel before you agree to deliver current records. Our post on records that include employee and customer data covers the personal-information side.
How do you compare a single payment with a revenue share?
Our view: compare them by who carries the risk, not by the headline number. A single payment is fixed once the data is accepted, and the buyer takes on the risk that it doesn't resell. A revenue share moves that risk to you: you may earn more if your records sell to several labs over years, or nothing if they don't sell. Nobody can forecast those sales for you. Turing, for its part, says it prices from a real sample because "A form can't see what's actually in your Slack or Jira." A mixed structure, such as Labelbox Prism's upfront fee plus a share, splits the risk.
Three things in an owner's own situation tip the choice:
- Need for certainty. A company that is closing, or that needs the money to settle debts or distribute to shareholders, is usually better served by a fixed amount paid within a month or two. Our guide to licensing when you're closing the business covers who signs and who is paid first.
- How long you'll be around to collect. A perpetual share needs someone to receive statements and payments for years. Decide who that is before you sign, especially if the company may be sold or wound down.
- What else you could do with the records. An exclusive single-payment deal closes off other buyers for the same use. A non-exclusive revenue share leaves room for other licenses, but check whether the agreement allows that.
Where we come in
Owners often see two offers for the same records that can't be compared directly: one a fixed amount, the other a share of sales nobody can forecast. Comparing them on the whole deal is what we do. We track the published terms of every major program, have direct contacts at the major data buyers and at AI labs, and read the payment, exclusivity and resale clauses with you before you choose. Nothing about your company is shared until you approve a specific, named buyer. Participating buyers may pay us referral fees, and we tell you which before any introduction. Owners pay us nothing, and you can apply to any buyer directly.
What to do next
Before you take any offer, write down three things: the date by which you need the money, who will receive revenue-share statements in five years, and which records you would never license exclusively. Then ask each buyer the same questions: Is it one payment, a share, or both? What is the share a percentage of? What happens if nothing sells? Can you license the same records elsewhere? For the full sequence from first form to payment, see how to sell or license your company's data to AI companies, or check where you fit.
Programs in this post
More from the blog
- Is my company too small to license its data to AI buyers?
Most AI data buyers with a size floor set it at 20 to 50 employees, but law, accounting and medical firms qualify smaller, and nine programs state no minimum.
- What company data AI buyers want: November 2026
Across the 17 AI data-buyer programs we track, accounting, healthcare and software lead the industries, and CRM, ticket, email and finance records lead.
- How to check that an AI data buyer is legitimate
Six checks an owner can run in an afternoon before sharing records. Who the buyer is, what you pay, what it asks first, personal data, payment and license terms.
See which buyers fit your company
Eight quick questions against every program in our buyer index. Participating buyers may pay us referral fees.