An exclusive data license promises one AI buyer that nobody else gets the same rights to the data, while a non-exclusive license leaves the seller free to license the same data again.
Published terms run both ways: Labelbox states its Prism licenses are exclusive by default, and micro1 takes a nonexclusive grant. Exclusivity is one of the choices that shape AI training data licensing, no public figure exists for what it adds to a company data price as of November 10, 2026, and granting none is valid.
Examples carry their check or filing date; the DataPacts Editorial Team explains how we source, date and check every figure.
Exclusive, Sole and Non-Exclusive Data Licenses Defined
The three grants differ in who else may use the data during the license: nobody, only the seller, or anyone the seller chooses.
- Exclusive: only the licensee may use the licensed rights within the scope, the seller included unless it reserves its own use.
- Sole: only one outside licensee exists, and the seller keeps its own use.
- Non-exclusive: any number of licensees may hold the same rights.
Endless states "You can license the same data to anyone else", the non-exclusive case, and Labelbox's seller "continued use" works like a sole grant.
Five Grant Models in Published AI Data Terms
Published AI data terms fall into five grant models, from exclusive by default to non-exclusive with no restriction at all. The table compares them as of 2026-11-09.
| Grant model | What the seller keeps | Published example | Source and check date |
|---|---|---|---|
| Exclusive by default | ownership and its own continued use | Labelbox Prism; buyers pay for stricter exclusivity | prism.labelbox.com/faq, 2026-11-09 |
| Exclusive for one purpose | uses outside that purpose | Kled: exclusive for developing and training third-party AI models, perpetual | kled.ai/terms-of-service, 2026-11-09 |
| Time-limited sole seller or restriction | free to license after the period | Turing Project Lazarus, 90 days; Tempus side letter, 18 months | lazarus.turing.com, 2026-11-09; Tempus 10-Q, quarter to 2026-09-30 |
| Non-exclusive with a restricting clause | other licenses, but not on similar terms or around the buyer | micro1 s.1.2 with s.5.1; Defined.ai Supplier Program s.3.5 with 12-month non-circumvention | micro1.ai/legal/data-license-master-terms, version of 2026-11-08; defined.ai/supplier-program, 2026-11-09 |
| Non-exclusive | any number of other licenses | Endless; Pangea; Protege s.1.1 | endless-data.com; pangea.ai; withprotege.ai, all 2026-11-09 |
| Set per listing or deal | depends on the listing | Grepped; HUMXN; FileYield | grepped.ai/terms, 2026-11-08; humxn.io; fileyield.com, 2026-11-09 |
Each term is the stating party's own published wording; the order form can change it.
Handshake AI, Scale AI and Mercor's data page publish no exclusivity term (checked November 9, 2026), so the order form decides. The license clause matrix compares the published contracts buyer by buyer.
What Exclusivity Does to the Price of a Data License
Exclusivity is named as a price driver by several buyers and intermediaries, but none publishes how much it adds for company data.
What Buyers and Intermediaries State About Paying More
Five parties name exclusivity as a price factor, and none attaches a number to what it adds.
- Labelbox states, checked November 9, 2026, that buyers pay for stricter exclusivity.
- HUMXN states, checked November 9, 2026, that exclusive licences pay more up front, with typical single-company deals starting at USD 250K+.
- SuperAnnotate states, checked November 9, 2026, an indicative valuation of USD 100K to USD 1M+ priced on rarity, depth and exclusivity.
- Origin Lab names exclusivity as a pricing factor in its FAQ, checked November 9, 2026.
- Mercor's software-company listing reads, checked November 9, 2026, USD 50,000 to 500,000, "More for large codebases or exclusive terms", and its trading-firm listing reads "Depends on history and exclusivity".
Exclusive and Non-Exclusive Deals Side by Side in Filings
SEC filings give the only dated amounts tied to an exclusivity term, and they come from health and genetic data, not company records.
23andMe disclosed a USD 50 million one-time payment from GSK on 2022-01-18 to extend the exclusive target discovery period by a year to July 2023, then USD 20 million upfront on 2023-10-30 for a one-year non-exclusive license. Both concern drug-target discovery, not AI training, and differ in scope and length, so the gap is not a premium.
The Tempus and Pathos neurology order form of 2026-10-22 covers up to USD 55.0 million: USD 50.0 million at USD 100 per delivered record plus USD 5.0 million for updates. A side letter bars Tempus from licensing the records to anyone but BMS, Pathos and certain subcontractors for 18 months after the later of delivery and 2027-10-22.
Why No Premium Figure Exists for Company Data
No buyer, seller or filing has published an exclusivity premium for company data as of November 10, 2026. The only stated multiple, roughly 4 to 5 times the non-exclusive price, is the statement of two founders interviewed in an Epoch AI article of 2026-01-12 about RL environments, not market data. The premium is set in negotiation; see methods for estimating how much company data is worth to AI companies.
Lock-in: What an Exclusive Grant Takes Away and How It Ends
Exclusivity costs the seller its other buyers for as long as it runs, and the contract decides how long that is and how it ends.
Other Buyers Turned Away While Exclusivity Runs
An existing exclusive grant can disqualify the same data with the next buyer, and buyers say so in their intake criteria. Mercor's medical-practice listing names "Data already licensed exclusively through a broker" as not a fit, and its imaging-center listing "Studies under an exclusive contract with another buyer" (checked November 9, 2026).
Exclusive and Perpetual in the Same Grant
An exclusive grant without an end date removes the data from every other buyer permanently. Kled's terms, on the contributor side, take a perpetual, irrevocable license exclusive to Kled for developing and training third-party AI models, with deletion at Kled's option (checked November 9, 2026). Polyshares' home page says "Defined material for a defined term", while its track-record entries read "Perpetual exclusive license to the operational record". Duration is covered under license term and perpetual licenses.
Exit Routes: Expiry, Renewal and Kill Clauses
Published terms show three ways an exclusive arrangement ends: it expires, the seller declines to renew, or a kill clause revokes it. Turing states it is the only seller for the first 90 days, after which the company "can renew or opt out" (checked November 9, 2026); Tempus's restriction ends 18 months after the later of delivery and 2027-10-22.
Labelbox states a kill clause forcing deletion within 30 days on material breach, regulatory change or a seller-defined change-of-control trigger, with no further training, transfer or sublicensing after revocation. It adds that "Models already trained on the data cannot be unlearned", and what survives revocation is governed by retraining and model retention rights.
Carve-outs That Narrow an Exclusive Grant
Carve-outs cut an exclusive grant down to what the buyer actually needs, by user, purpose, recipient or time.
Keeping the Seller's Own Use
The first carve-out reserves the seller's right to keep using its own data, which turns an exclusive grant into a sole one. Labelbox states the seller "retains full ownership and continued use" under its exclusive default (checked November 9, 2026).
Exclusive Only for One Purpose
Exclusivity can be limited to one purpose, leaving every other use of the data open to other licensees. Kled's exclusivity covers developing and training third-party AI models, and the CCC AI Systems Training License is the non-exclusive counterpart: a voluntary collective license for training uses external to the licensee (checked November 9, 2026). Purpose limits are covered under field of use and scope limits.
Named Recipients and Blocked Competitors
A recipient carve-out names who may still receive the data, or who may never receive it, instead of closing the market to everyone. The Tempus side letter keeps BMS, Pathos and certain subcontractors inside its restriction, and SuperAnnotate states the seller approves each buyer and "can block competitors permanently" (checked November 9, 2026).
A Fixed End Date for the Exclusivity
Exclusivity can end before the license does, so the buyer keeps a non-exclusive license after a head start. Turing's 90 days and Tempus's 18 months are the only recorded periods; no company-data norm is published.
Non-Exclusive Grants That Still Restrict Other Sales
A grant labeled non-exclusive can still limit other sales, through a warranty, an order form or a non-circumvention clause.
micro1's Data License Master Terms (last updated November 8, 2026) grant a nonexclusive license in s.1.2, yet the licensor warrants in s.5.1 that it has not granted a third party rights to the same Data on the same or substantially similar terms, and s.4.2 lets the Order Form add exclusivity. micro1's program pages say the data is licensed "exclusively" to labs and enterprises in the U.S. and allied countries, which limits recipients and does not make the grant exclusive.
Defined.ai's Supplier Program states no exclusivity and no minimums in s.3.5, but sets a 12-month non-circumvention period (checked November 9, 2026). Selling to several buyers at once is its own process, described under running several AI data buyers in parallel.
Exclusive Rights Given to an Intermediary Instead of an AI Developer
Some exclusivity goes to the party that resells or places the data, not to the AI developer that trains on it (all checked November 9, 2026).
- Turing states it is the sole seller for 90 days while lab licenses stay non-exclusive, so "the same record can go to more than one lab".
- Newsflare requires footage to be exclusive to Newsflare for AI licensing, with a 90-day minimum after upload and deletion on request within 30 days after that.
- Troveo states that 95 percent of its licensors signed exclusively (resource article of 2026-07-25), while its home page phrases the 95 percent as content "licensed direct from the source".
An exclusive mandate to a reseller differs from an exclusive license to a developer, and the terms of AI training data marketplaces differ on who holds the exclusivity.
Granting Exclusivity or Not: Benefits, Risks, Downsides and Alternatives
Exclusivity trades the seller's other buyers for a possibly higher price from one, and the carve-outs decide how much is traded.
Benefits:
- Buyers state they pay more for it.
- A single counterparty and one set of terms.
- Carve-outs can keep the seller's own use.
Risks:
- Later buyers decline data already exclusively licensed.
- Exclusive plus perpetual removes the data from the market.
- Trained models stay after revocation.
Downsides:
- No published premium anchors the price.
- Restrictions survive in non-exclusive grants anyway.
Alternatives:
- Non-exclusive licenses that leave other buyers open, as Endless, Claru, pre.dev (by default) and Protege state.
- Time-limited or purpose-limited exclusivity.
- A collective non-exclusive license, such as CCC's.
- Not licensing at all.
The wider question sits under alternatives to selling data to AI labs.
How the Exclusivity Clause Interacts With the Other Clauses of a Data Licensing Agreement
Exclusivity is one clause of the grant, and its effect depends on the duration, purpose and sublicensing clauses beside it. The parent page explains every clause of an AI data licensing agreement, and the wider question of should you sell or license your data sits beside it.
Related clause pages:
- License term and perpetual licenses
- Sublicensing and onward transfer
- Retraining and model retention after a license ends
Process steps:
- Choosing a buyer for your data
- Evaluating an offer for your data
- Term sheets, order forms and LOIs
- Negotiating an AI data licensing deal
Not Legal Advice, and the EU Rule on Imposed Data Terms
This page explains exclusivity clauses for information and is not legal advice; a lawyer in the seller's jurisdiction should review any grant before signing. Regulation (EU) 2023/2854 Art. 13(1) makes an unfair term unilaterally imposed between enterprises not binding, and Art. 13(5)(c) presumes unfair a term that stops a party using data it provided or generated during the contract. Art. 13(8) excludes price and main subject matter, and Chapter IV applies to contracts concluded after 2025-09-12 and from 2027-09-12 to certain earlier ones (Art. 50); see Data Act Article 13 unfair terms.
Terms, Filings and Deals Cited on This Page
The examples above come from these documents, checked on November 9, 2026, or dated by filing.
- 23andMe and GSK: 8-K exhibits, 2022-01-18 and 2023-10-30
- Epoch AI: article, 2026-01-12
- Labelbox Prism buyer profile: FAQ
- micro1 data buyer profile: Data License Master Terms, updated 2026-11-08
- Tempus AI clinical data licensing deals: 10-Q, quarter to 2026-09-30
- Turing Project Lazarus profile: program page
- Company data partnership programs: program terms
- AI data licensing deals tracker: filed deals
- RL environments and who buys them: the Epoch AI statement
Frequently Asked Questions
What is the difference between an exclusive and a non-exclusive data license?
An exclusive license bars the seller from licensing the same rights to anyone else within its scope; a non-exclusive license lets the seller license the same data again.
What is an example of an exclusive data license for AI training?
Labelbox states that its Prism licenses are exclusive by default, with the seller keeping ownership and continued use and a kill clause forcing deletion within 30 days (checked November 9, 2026).
Do exclusive data licenses pay more?
Labelbox, HUMXN and Mercor state that exclusivity raises the price, but no party has published a premium for company data as of November 10, 2026.
Can I license the same data to several AI companies?
Yes, under non-exclusive terms such as Endless's or Protege's, but check for a no-prior-grant warranty like micro1 s.5.1 (checked November 9, 2026). Doing so is its own process. See running several AI data buyers in parallel.