Should our data license be exclusive or non-exclusive?

Exclusivity is the term that most changes what a data license is worth, and it is the one buyers publish least. An exclusive license should pay more because you give up the right to license the same records to anyone else. A non-exclusive license pays less per deal but leaves you free to license again. This guide walks through the choice, the payment structures that go with it and the clauses to read. To see where your records might land first, get a confidential estimate.

Is a data license a sale?

Usually not. Most programs license a copy and leave ownership with you. micro1 says companies keep ownership of their underlying data (micro1). Turing acquires defined rights to a prepared record while the seller keeps ownership (Who Sells Data, Oct 14, 2026). SuperAnnotate licenses rights for a set period (SuperAnnotate). One caution: once data is used in training it is practically impossible to remove (FKKS, Oct 1, 2026). A time-limited license still has effects that last.

What do buyers publish on exclusivity?

Few publish anything. Here is what we found on Oct 16, 2026.

micro1, Mercor, Scale AI, Handshake AI, Polyshares, Turing and Nyne AI did not state an exclusivity position on the pages we read.

When does exclusivity make sense?

When the premium is real

Ask the buyer for both prices, exclusive and non-exclusive, in writing. If they differ by little, keep your freedom.

When few buyers would want the records anyway

A niche archive may only interest one buyer. Exclusivity then costs you little.

When the company is closing

A closed company has no use for a second license and may want one clean payment.

When should you keep it non-exclusive?

When several buyers want the same kind of record

Troveo says it licenses each dataset to more than one buyer and reports an average of 4 sales per dataset (Troveo, its own figure). SuperAnnotate pays on each relicense.

When the exclusivity is broad

“Exclusive for all AI uses, worldwide, forever” is a sale in practice. Narrow it. Broad exclusivity also stays with the business if you sell the company, and an acquirer’s lawyers will read it closely; see how a data license affects M&A due diligence.

Upfront, revenue share or both?

Three structures appear in published terms.

• One payment. Turing pays a single payment of up to $1M with no recurring commitment (Turing). Polyshares pays Net 30–60 after delivery (Polyshares).
• Upfront plus revenue share. Labelbox Prism pays an upfront licensing fee at signing plus a revenue share on downstream use (Prism).
• Revenue share only. Troveo and Magma pay a share as data licenses. Troveo publishes no percentage (Who Sells Data, Oct 14, 2026).

Our comparison of AI data partnership programs sorts each published program by these structures. A revenue share is worth only what you can audit. Ask for statements per sale, the right to check them, and a minimum payment if you are giving exclusivity.

One practitioner’s rule of thumb: Ryan Lock, a data broker, said on the NGMI podcast in October 2026 that perpetual use of delivered data is standard because files cannot be tracked down the chain, but exclusivity should run for a fixed term, usually two years, so you are not locked out of later deals. That is his practice, not a published buyer term.

What happens when the license ends?

Buyers handle this differently, so ask. micro1 says data is retained for agreed periods and deleted on request or at the end of the engagement (micro1). SuperAnnotate retains shared data for an agreed period and then deletes it (SuperAnnotate). Ooak Data deletes raw archives within 30 days of delivery and keeps only the anonymized dataset for the license term (Ooak). Deletion of the dataset does not undo training. FKKS notes that once data is used to train a model it is practically impossible to remove (Oct 1, 2026). Price an exclusive license as if its effects will outlast its term.

The clauses to read before you sign

Our guide to what’s in an AI data licensing agreement reads two published agreements clause by clause. The short list:

1. Scope. Which systems, which date ranges, which fields. Excluded systems should be annexed, as Ooak Data does (Ooak).
2. Field of use. Training, evaluation, or both. A competitor carve-out if you still operate. FKKS notes a buyer could train a model that competes with you.
3. Exclusivity. Who it covers, for what use, for how long, and in which territory.
4. Term. A fixed period, with what happens to the data at the end.
5. Sublicensing and resale. Can the buyer pass your records to a lab or other customers? Troveo, SuperAnnotate and Appen all license on to labs.
6. Approval of end buyers. SuperAnnotate lets you approve each buyer. Ask for the same elsewhere.
7. De-identification standard. Who does it, where, and to what standard. Ask to see a sample first.
8. No re-identification. A contractual ban, with a remedy.
9. Deletion. Raw data deleted on a date, with written confirmation. Ooak commits to deleting raw archives within 30 days of delivery.
10. Payment. Amount, milestones, acceptance criteria and what happens if the buyer rejects the data.
11. Warranties and indemnity. You will be asked to warrant you own the data. Limit the warranty to what you know, and cap the indemnity.

How an advisor helps here

An advisor gets the same scope priced exclusive and non-exclusive by more than one buyer, so you can see the premium. Marin Data costs you nothing.

Get a confidential estimate.

Sources

• NGMI podcast, interview with data broker Ryan Lock, Oct 16, 2026 (his figures are his own and unverified)
• Labelbox Prism, Pewter, SuperAnnotate, Mode, Turing Project Lazarus, Polyshares, Troveo, Ooak Data, micro1, all observed Oct 16, 2026
• Opendatabay, fees, last updated Oct 6, 2026
• Who Sells Data, buying programs, observed Oct 14, 2026
• Frankfurt Kurnit Klein & Selz, selling company data for AI training, Oct 1, 2026