Exclusive or non-exclusive: how a data licence's scope changes its price

A non-exclusive data licence lets the holder license the same data to other buyers; an exclusive licence promises that no one else may use it, in the licensed field and for the licensed time, and buyers pay more for that promise. How much more depends on what the exclusivity covers. Where "exclusive" only means that data is distributed through one platform, platforms pay their suppliers about 1.25 to 2 times their non-exclusive rate. In the one disclosed case of the same buyer paying for the same data both ways, GSK paid 23andMe $50 million for a fifth exclusive year and later $20 million for a one-year non-exclusive licence: about 2.5 times, although the two years did not cover exactly the same things. And founders who sell training environments told Epoch AI that sole use by one buyer costs roughly 4 to 5 times a non-exclusive licence.

The range is the point. "Exclusive" is not one term but a family of them, and its price follows its scope: which uses it covers, for how long, against whom, and over which data. A holder that agrees to "exclusivity" without defining those four things has given away the most valuable concession in a data licence without knowing its size. Taking it back later has a price too: Tempus AI, as licensee, gave up its exclusivity over licensed clinical data in return for $7.9 million lower future payments.

This guide is for the founder, CFO or counsel of a company (a holder) that a buyer has asked for exclusivity, or that is deciding whether to offer it. It explains what exclusivity means in AI data licences, how buyers' published terms handle it, what the evidence says it is worth for each scope, why a non-exclusive licence can earn more over time by selling more than once, and how to price and write exclusivity when it is worth granting. It draws on the 186 clauses we have read verbatim in buyers' published terms, the 15 terms in our licence terms register and the 72 levers on our negotiation page. Of the 111 licensing deals on our deal ledger, only 19 state a dollar amount, and no AI training deal among them prices the same data both ways, so every multiple below comes with its caveat.

This guide is research from public sources, not legal advice.

Contents

1. Exclusive and non-exclusive in plain terms
2. Four things exclusivity can mean
3. The four dimensions that size an exclusivity
4. What buyers' own terms say
5. What exclusivity is worth, by scope
6. How to read the evidence
7. Selling more than once: what non-exclusive is worth over time
8. Worked example: one exclusive licence or several non-exclusive ones
9. When exclusivity is worth granting
10. How to price exclusivity
11. What an exclusivity clause should say
12. Exclusivity in a sale we run
13. Checklist before you agree to exclusivity
14. Questions people ask
15. What this means for you

1. Exclusive and non-exclusive in plain terms

A licence is permission. The holder (the licensor) lets the buyer (the licensee) do specified things with specified data, and keeps everything it did not grant. Whether the licence is exclusive decides one thing above all: whether the holder may grant the same permission to anyone else.

Under a non-exclusive licence the holder keeps that right. It may license the same records to a second buyer next month, keep using them in its own business, and build its own products on them. Most disclosed AI content deals take this form: the Financial Times licence to OpenAI, Axel Springer's, Stack Overflow's licence to Google and Informa's to Microsoft are all non-exclusive. Google's content-offer pilot puts it in one line to publishers: This is a non-exclusive opportunity, with the right to monetise the same content elsewhere.

Under an exclusive licence the holder promises not to grant the same rights to anyone else within the licence's scope, and in the strongest form it gives up the right to exercise them itself. US copyright law treats that as a change of ownership for the parts of a dataset copyright covers: a "transfer of copyright ownership" is an assignment, mortgage, exclusive license, or any other conveyance ... but not including a nonexclusive license. Two consequences follow. An exclusive licence of copyrighted material (documents, code, written messages) is not valid unless it is in writing and signed, so an exclusivity agreed on a call or in an email thread may not deliver what either side thinks. And for the large parts of company records that copyright does not protect (transaction records, logs, factual fields), exclusivity exists only as a contract promise: it binds the two parties, and nobody else.

Practitioners also separate a sole licence, in which the holder licenses no one else but keeps its own use, from an exclusive licence in the strict sense, in which even the holder may not use the data in the licensed field. For an operating company that difference is decisive. The company keeps running its business on its own records, and may want to train its own models or build AI features on them. Whatever word a draft uses, the licence should say in terms that the holder's own internal use is reserved.

The rest of this guide is about the middle two rows, because that is where buyers ask and where the price is decided. Our guide to what a data licensing agreement contains covers the other clauses of a licence.

2. Four things exclusivity can mean

When a buyer asks for "exclusivity", it can mean four quite different promises, and each has a different value to the buyer and a different cost to the holder.

Sole use by one buyer

This is the strongest form: one buyer, and only that buyer, may train on the data within the licensed scope. It is what founders selling reinforcement-learning environments described to Epoch AI when they put exclusive deals at roughly 4 to 5 times the non-exclusive price. It is also what Polyshares lists on its track-record page, where every licence is a perpetual exclusive license to the operational record. A perpetual, exclusive licence of a company's records is close to a sale: the holder can never license those records to anyone else. It should be priced as one.

Sole distribution through one channel

Here the holder promises to offer its data only through one intermediary, which may then license it to many buyers. Nobody gets sole use; what is exclusive is the channel. Troveo, which licenses video for AI training, reports that 95 percent of its 7,000+ licensors signed exclusively, and says its exclusivity is time-limited and it has to be earned. Protege's own case study explains why an intermediary wants it: by holding a broadcaster's exclusive AI training rights, it could maximize the price because AI buyers were not able to negotiate with multiple parties. That can serve the holder well, or it can tie the holder to one channel for as long as the exclusivity lasts.

Stock libraries have priced this form of exclusivity for years, which is why their rates are useful evidence (section 5): they show what a platform pays for exclusive supply, not what a buyer pays for sole use.

A time-limited window

A window gives one buyer exclusivity for a period, after which the data opens to others. Disney's three-year licence to OpenAI carried about one year of exclusivity, according to Disney's chief executive. GSK's exclusive period over 23andMe's database, begun in 2018, was extended a year at a time, with the fifth year bought for a one-time payment of $50 million, and when it ended GSK moved to a non-exclusive licence. In genomics the window has also been bought by paying for the data to be made: five drug companies funded sequencing for UK Biobank in return for a limited period of exclusive access before the data opened to approved researchers. A window is often the best trade for a holder, because it sells a head start and keeps the long tail.

First look, and the promise that nobody had it before

The fourth family is not exclusivity in name, but it works like it. A first look lets one buyer see the data, or make its offer, before anyone else; a right of first refusal lets it match any later offer; a last look lets it see the others' offers and improve its own after the deadline. Each one reduces the competition the holder's price depends on. In the one public auction of a company's records, competition moved the price to 2.5x the opening bid ( lever), which is what a first look quietly gives up.

A related promise looks backward. micro1's published terms grant it a licence that is itself nonexclusive, irrevocable, perpetual, yet ask the holder to warrant that it has not previously entered into any transaction granting a third party rights to license, access, use or otherwise exploit the same Data on the same or substantially similar terms. That does not stop the holder licensing the data later, but it disqualifies data that was licensed first to someone else. A holder that plans to license the same records to several buyers should ask for that warranty to cover only earlier exclusive grants, so that a non-exclusive licence to another buyer is not a breach.

3. The four dimensions that size an exclusivity

Whatever form it takes, an exclusivity has a size, and the size is what should be priced. Four dimensions decide it.

Field of use

The field is what the buyer gets sole rights to do: pretraining, fine-tuning, evaluation or retrieval; models of one kind or for one industry; or everything. Licence templates already provide for a narrow field: one published form offers a licence that is NONEXCLUSIVE / EXCLUSIVE WITHIN THE DEFINED FIELD, and our register's position is to grant exclusivity only for a defined field and a defined window, at a multiple. A narrow field is cheaper for the holder because it leaves every other use free to license. It is also cheaper for the buyer, which is why a narrow field and a lower premium usually go together: our valuation shows an evaluation-only licence at 0.2 to 0.5 times the value of a full commercial one, and the same logic runs through exclusivity. The register's general rule for purpose applies here too: name the permitted uses and exclude what is not paid for.

Term

The term is how long the exclusivity lasts, and it need not match the licence's term. Disclosed AI licences run two to three years, and the only disclosed window we hold is shorter than the licence it sits in ( Disney). Our register records that no public source states typical window lengths in months for company data ( register), so the length is open to negotiation in every deal. A perpetual exclusivity is a sale in all but name. Model survival is a separate question: buyers ask to keep models trained during the term, and our register's position is that any survival must be written in. An exclusivity can end on schedule while the buyer keeps one model family it trained, which is a reasonable trade.

Against whom

Exclusivity excludes someone, and the clause should name who. The usual answer is other buyers. The questions holders miss are whether it also excludes the holder's own use (section 1), other intermediaries already carrying the data, and software vendors that may already have trained on it. Atlassian, GitHub (for Copilot interactions) and Figma may, on some of their plans, use the content held with them to train or improve their own AI, and our rights checklist recommends that the licence states exclusivity only against third parties you control, with each vendor's training setting listed. A holder cannot promise exclusivity against a party it has no power over.

Over which data

The last dimension is the data itself: a snapshot of records up to a date, named systems, or everything the company produces during the term. An exclusivity over future records locks up the holder's future production, which is exactly what a living company should keep licensing. Derived data needs the same care. A buyer may make cleaned, labelled or synthetic versions, and micro1's terms let it sublicense through multiple tiers; our register treats a perpetual sublicensable grant as what it is, a resale right. A non-exclusive licence with broad sublicensing can therefore compete with the holder's own later licences almost as an exclusive one would.

Territory, the fifth dimension in most licences, rarely matters here: models are trained in one country and used in many, so data licences are usually worldwide, and a geographic exclusivity is worth little to a buyer.

4. What buyers' own terms say

Buyers' published terms show the whole range, from non-exclusive grants that ask for a promise of first use, to perpetual exclusive licences, to intermediaries that take exclusivity from holders and license on non-exclusively. More than half of the buyers whose terms we have read publish nothing on exclusivity at all, so for them the licence they send is the only place it is set. The table covers those that do.

Exclusive upstream, non-exclusive downstream

The most instructive row is the pairing of Kled's two sides. Its contributor terms take a licence that is fully sub-licensable (through multiple tiers), worldwide, perpetual and irrevocable and exclusive to Kled. Kled then announced a $12M non-exclusive data licensing deal over two years with a buyer it did not name. An intermediary that takes exclusivity from those who supply the data and sells non-exclusively to buyers keeps the value of selling more than once for itself. That is a legitimate business model, but it shows where the value of non-exclusivity sits, and a holder that grants exclusivity to an intermediary should be paid for it, or keep the right to sell the same data directly.

Prices that are exclusive by construction

Polyshares' listed prices are for perpetual exclusive licences, from $110,000 for a global marketing agency's records to $450,000 for a roofing company's. These are the firm's own published claims, not confirmed sales, and because the rights are perpetual and exclusive they belong beside sale prices, not shared licences. A holder comparing an offer for a non-exclusive, time-limited licence with these figures is comparing different things: the right comparison is what the same records would fetch from several non-exclusive buyers over the period, as section 7 explains.

5. What exclusivity is worth, by scope

The evidence on the premium is thin, but it has a clear shape: small when exclusivity means a single distribution channel, larger when it means one buyer's sole use. The table sets out every source we hold, and what each actually measures.

Sole distribution: what platforms pay for exclusive supply

Stock libraries pay contributors more when they supply exclusively. Pond5 pays exclusive video artists 40% against 30% for non-exclusive artists. Alamy's contributor contract raises the rate only at its top tier, where exclusive images earn 50% against 40%, and states that below that tier exclusivity does not change the rate at all. Envato charged exclusive authors a fee of 12.5% to 37.5%, falling with sales, against 55% for non-exclusive authors, so an exclusive author kept between roughly one and a half and twice what a non-exclusive one kept. Then, in September 2026, Envato replaced both with an Author Fee of 50% for everyone.

Three things make these figures a floor rather than a guide to an AI buyer's premium. They are what a platform pays for being the only shop, not what a buyer pays for being the only user: the same file is still licensed to many buyers. The premium is paid as a share of each sale, so it rewards volume through one channel. And Envato's decision to drop its exclusive rate suggests that, for that platform at least, exclusive supply stopped being worth what it paid for it. A holder offered a platform's exclusive rate should not mistake it for the price of sole use. Platforms that carry data into AI deals also keep a large share of the deal itself: footage and photo platforms keep about half, which is one more reason to price a direct licence first.

The one same-buyer pair: 23andMe and GSK

The only disclosed case of the same buyer paying for the same data with and without exclusivity is in 23andMe's filings. In February 2022 GSK chose to extend its exclusive target discovery period over 23andMe's genetic and health survey database for a fifth year, and 23andMe received a one-time payment of $50 million. When the exclusive period ended, GSK took a $20 million upfront payment for a one year, non-exclusive data license, announced in October 2023: about 2.5 times for the exclusive year.

It is not a clean comparison, and the differences pull in opposite directions. The exclusive years carried a joint drug discovery collaboration, so the higher price bought more than sole use of the data. The non-exclusive licence included new data and some added analyses, so the lower price bought more data. And after five years of exclusive work, GSK may simply have valued another year of the same database less. The net effect of those three is unknown. What the pair does show is that a sophisticated buyer, with years of experience of the data, paid materially more for exclusivity than for the same access shared, and kept licensing the data once exclusivity ended.

Sole use by one buyer: four to five times

The largest multiple on record comes from the market for reinforcement-learning environments, where two founders independently told Epoch AI that exclusive deals are roughly 4 to 5 times the non-exclusive price. Environments are often built to order and are valuable precisely because a rival lab has not trained on them, which is when sole use is worth most. Our valuation shows an exclusive licence beside the non-exclusive figure at 2.5 to 5 times, from the 23andMe pair at the low end to this source at the high end, and never blends it into the main range. For company records, buyers say exclusivity matters for the same reason: one buyer's executive noted lower demand for a codebase that a rival lab already has its hands on.

Giving exclusivity up: Tempus

The clearest evidence that exclusivity has a separable price comes from the licensee's side. Tempus AI's annual report says that in January 2023 it amended a data licensing agreement, which reduced the future data license payments the Company owes in exchange for waiving exclusivity rights. It booked a $7.9 million gain from the lower payments and a $7.4 million impairment of the licensed data, which had been carried at more than $40.1 million gross. The licensor is not named, and the filing does not give the total remaining payments, so the share of the licence that exclusivity represented cannot be computed. But both sides put a number on exclusivity alone, and agreed it.

Made to order: the reverse default

In one corner of the market the default runs the other way. Robot and teleoperation data collected to a buyer's specification is licensed to you exclusively by default at one collector, with non-exclusive licensing offered at a discount. When data is made for one buyer, exclusivity is the starting point and sharing is the concession. For data a company already holds, made in the course of its own business, the reverse is true, and the premium should be paid by whoever wants sole use.

6. How to read the evidence

A holder told that exclusivity is worth "four to five times" or "about one and a half times" is being given a number from one of the rows above, and each row measures something different. Five rules help read any such claim.

The shape is firm; the number is not. Every source agrees that exclusivity is worth a multiple and that the multiple grows with scope, from a single distribution channel to one buyer's sole use. No source gives a number a holder could apply to its own records without judgment, and only two (23andMe and Tempus) are disclosed transactions with amounts.

Ask what was excluded. A premium measures a scope. "Exclusive" in a stock library's contract means supplying no other library; in an RL founder's account it means no other lab trains on the environment. The same word can carry a multiple of one and a quarter or of five.

Supplier-side rates are a floor. What a platform pays its suppliers for exclusive supply is the least that exclusivity is worth to anyone in the chain, because the platform still makes money licensing non-exclusively to buyers.

Bundles inflate, and new data deflates. An exclusive period that comes with a research collaboration, an equity investment or services is worth more than the data alone; a later non-exclusive licence that adds new data is worth more than the old data alone. Read every pair for what each side contained.

Consider who is speaking. An intermediary whose contracts take exclusivity from holders has reason to describe it as valuable; a buyer seeking it has reason to describe it as standard. Troveo's own pages run from its worked example of 150, sometimes 200 to 4 to 5 times on its environments page. Neither is wrong for the case it describes, which is the point.

What is missing matters as much. There is no disclosed AI training licence that prices the same company records both ways, and no published length for a typical exclusivity window. Our research plan lists both as open gaps, and the negotiation page shows each lever's sources and strength, 39 of them backed by a primary record or several sources.

7. Selling more than once: what non-exclusive is worth over time

The case against exclusivity is not that buyers pay too little for it. It is that a non-exclusive licence can be sold again, and the total of several licences can exceed one exclusive price.

The evidence that data sells more than once

The largest licensors show it plainly. Reddit licensed its users' posts to Google for a reported $60 million a year and, three months later, to OpenAI for a reported $70 million a year; its prospectus put the aggregate contract value of its January 2024 licences at $203.0 million over two to three years. News Corp licensed its journalism to OpenAI, for a reported $250 million over five years, and later to Meta, for up to US$50 million a year. Neither company could have signed its second licence had the first been exclusive in the same field.

Smaller data businesses work the same way. A Chinese speech data company's prospectus shows it sold usage licences on 395 of its own off-the-shelf datasets in one year, 576 licences in all, and that those owned datasets earned 45.48 percent of its revenue; the data it made to order passed to the buyer and could not be resold. Troveo advertises 4× AVG Sales Per Dataset, labelled as illustrative, and SuperAnnotate pays holders on delivery, and on each relicense. For code, Pangea's partner licenses private repositories non-exclusively at $1,000 to $1,500 per licence for a typical repository, so the same repository can earn that more than once.

Why repeat sales are not automatic

Three forces cut the other way, and an honest comparison includes them.

Later buyers pay less. Data that a rival already holds is worth less to the next buyer, as buyers say of a codebase that a rival lab already has its hands on. The second licence of the same records rarely fetches what the first did.

Some archives sell once. Getty attributed its forecast revenue decline to one-time revenue recognition from licensing deals the year before. A static archive, once licensed to the buyers that want it, has no one left to sell to.

The number of buyers is small. Demand for training data is concentrated in a few large buyers: one listed data vendor drew approximately 58% of its revenue from a single buyer. A holder should count the buyers that could realistically license its records, not the buyers that exist.

What changes the picture is new data. A company that keeps operating produces new records every month, and buyers pay for continuing access: our register's lever is to sell access to refreshing data, not a one-off file. A living company that licenses non-exclusively can sell the same kind of records to several buyers and sell each of them the next quarter's records too. An exclusivity that covers future records gives up both.

8. Worked example: one exclusive licence or several non-exclusive ones

This is an illustrative example, not a real company, and its figures are relative units, not prices from any buyer. A legal-services software company with ten years of matter-management records, still operating, is approached by a buyer that wants exclusive rights to train on them. The company's research suggests that four buyers might license records like these, but it has written interest from only one.

Call the price the first buyer would pay for a non-exclusive licence "one unit". The buyer offers one and a half units for exclusivity: Troveo's worked example puts the exclusive price at 150, sometimes 200, where the non-exclusive price is 100. Is it a good offer?

Exclusive path. One and a half units now, and nothing more from these records for the term, since the company cannot license them elsewhere. If the exclusivity covers future records, the company also forgoes refresh licences to everyone else.

Non-exclusive path. One unit from the first buyer. If a second buyer licenses the same records, even at less, the company passes the exclusive offer. If four buyers license them, which is what Troveo labels its illustrative average, the total is several times the exclusive offer, less whatever the later buyers knock off because others already hold the data. If no second buyer appears, the company has given up half a unit.

Window path. Exclusivity for one field and a short window, after which the records open to others, as in Disney's licence. The buyer pays a premium for the head start; the company keeps the long tail of non-exclusive licences after the window. This usually dominates the other two when more than one buyer is plausible.

The arithmetic gives a simple test. An exclusivity premium of M times the non-exclusive price is worth taking only if the company expects fewer than M buyers, at full price, over the exclusive term. At one and a half times, two realistic buyers already beat it. At the 4 to 5 times the RL founders reported, a holder would need four or five buyers paying full price, or more at falling prices, to do better without exclusivity. The test is rough (it ignores timing, the risk that later buyers never arrive and the value of money now), but it puts the question where it belongs: on the number of buyers, not on the size of one offer.

For this company, with one written expression of interest and four plausible buyers, the best response is neither yes nor no. It is to ask the buyer to price a window in one field, and to put the same records before the other plausible buyers at the same time, so that the non-exclusive offers exist before the exclusive one is weighed against them. Our guide to how much company data is worth explains how the unit itself is estimated.

9. When exclusivity is worth granting

Exclusivity is not a mistake in itself. In several situations it is the right trade, provided it is scoped and paid for.

• The data has one realistic buyer. If only one company in the world will train on a narrow kind of record, exclusivity costs the holder little, and any premium is gain.
• The data was made for the buyer. Data collected or built to order (robot episodes, custom environments) is exclusive by default in some markets, and the buyer has paid for its making.
• The company is closing. A closed company's records will not refresh, and an operating company's records are worth more than a closed one's ( lever). A clean, perpetual, exclusive licence or a sale can be the right exit, priced as a sale.
• The buyer pays for a short window up front. A window sold for cash on signature, with the records open afterwards, takes the premium without giving up the tail.
• The data is what differentiates the buyer's model. Buyers pay most for exclusivity when sole use is the point, and buyers' own executives name exclusivity as a price driver ( lever); our register's advice to holders there is to price exclusivity to one buyer as a separate term.

Exclusivity is usually a poor trade when the company is operating and its records grow every month; when several buyers could plausibly license the records; when the request is for perpetual, sublicensable and exclusive rights together, which is a sale under another name; and when the request comes from an intermediary that will license the data on to many buyers non-exclusively, keeping the value of repeat sales for itself.

10. How to price exclusivity

The sequence matters more than any single number.

Get the non-exclusive price first

Exclusivity is priced as a premium over something, so the something must exist. The best evidence of a non-exclusive price is an actual offer, and the best way to get several is to put the same description before every buyer that fits at the same time, with one deadline: in the one public auction of company records, competition took the price to 2.5x the opening bid ( lever). A written offer also has a direct use: one buyer commits to beat a qualifying written offer for your company's operational data on the same rights and payment conditions.

State exclusivity as a separate, optional price

Ask each buyer for its non-exclusive price, and for exclusivity as an option on top: "this licence, non-exclusive, at this price; exclusive within this field for this window, at that price." Two separate numbers let the holder see what the buyer will pay for exclusivity alone, and compare it with what non-exclusive licences to others would bring. A buyer that will only make an offer for exclusive rights is telling the holder something about how much it values sole use.

Size the premium to the scope

Use the evidence in section 5 as a scale, not a formula. Exclusivity of one channel, or of a narrow field for a short window, sits near the bottom of the range, closer to what platforms pay for exclusive supply. Sole use of the whole of a dataset that no rival holds sits at the top, where the RL founders put it. The valuation shows the exclusive figure beside the non-exclusive one so that both sides of the trade are on the page.

Time-limit it, and decide what happens after

Every exclusivity should end. Set a window with a start and an end, and say in the licence what the buyer has afterwards: a non-exclusive licence for the rest of the term, at the same fee or a lower renewal fee, is a common shape. A renewal of exclusivity should be priced again, as GSK's fifth year was.

Get paid for it in cash or a minimum

An exclusivity paid only by a revenue share is the worst combination for a holder: it cannot license elsewhere, and its income depends on one buyer's use. Ask for the exclusivity premium up front, or for a guaranteed minimum; the market's model for usage-linked licences is a floor plus a usage-linked variable. Where a buyer insists on a share, make the exclusivity lapse if the minimum is not paid.

Price the exit before it is needed

Tempus shows that exclusivity can be unwound for money years into a licence. Writing the price in at the start is easier than negotiating it later: a buyer may convert to non-exclusive at any time for a stated lower fee, or the holder may end exclusivity by refunding a stated part of the premium. Both sides then know what exclusivity costs at every point in the term.

11. What an exclusivity clause should say

An exclusivity clause that says only "the licence is exclusive" leaves every question in this guide open. A clause that is fit to sign answers each of them. Our register records the market position on exclusivity and exclusivity windows; the elements below turn that position into drafting points for the holder's counsel.

1. The data covered. Name the records, systems and date range. Say whether deliveries after signing are covered, and whether derived data (cleaned, labelled, synthetic or embedded versions) falls inside the exclusivity or outside it.
2. The field. Define the uses that are exclusive (for example, training models of one kind, or for one industry) and say that all other uses remain free to license.
3. The window. State the start date, the end date, and what the buyer holds after the window ends.
4. Who is excluded. Third parties the holder controls, by licence or contract. Reserve the holder's own internal use, including training its own models, in terms.
5. What is already outside it. List licences granted before signing, software vendors that trained on the content under their own terms, and disclosures the law requires.
6. The price and its conditions. State the exclusivity premium separately from the licence fee, when it is paid, and that exclusivity lapses if a minimum is not paid.
7. Conversion and exit. Give the buyer the right to convert to non-exclusive for a stated lower fee, or the holder the right to end exclusivity for a stated refund.
8. The prior-licence warranty. If the buyer asks the holder to warrant that the data was not licensed before, limit it to earlier exclusive grants, so that a non-exclusive licence to another buyer does not breach it ( the clause buyers use).
9. Writing and signature. Put the exclusivity in the signed licence, since an exclusive licence of copyrighted material must be in writing and signed to be valid.

Two further points belong in the same negotiation. The licence should still bar resale of the raw data as a standalone product, as micro1's own terms do ( clause), whatever the exclusivity; and the term and model-survival clauses should be read together with the window, because a buyer that keeps a perpetual right to models trained during the window has, in effect, a permanent head start ( lever).

12. Exclusivity in a sale we run

AIData.Broker acts as the holder's agent: the holder itself licenses or sells its data to the buyer, and we do not take ownership of it. Under our mandate the holder alone decides whether to sell or license at all, to which buyer and on what terms, and the mandate names exclusivity, field of use and term among those terms. The holder may reject any offer, including the highest, without giving reasons. When the holder asks, we advise on the terms and help negotiate them, including the exclusivity clause; the licence remains the holder's agreement, and the holder takes its own legal advice on it. We run sales only for holders based in the United States, the United Kingdom and Canada.

Two parts of the mandate bear directly on exclusivity. No buyer gets a first look or a last look, whatever its relationship with us. If a buyer asks for either, or its standard terms require that the data has not been offered to anyone else (the kind of promise micro1's prior-licence warranty asks for), we tell the holder in writing what granting it would cost, and the holder decides. And every buyer that fits receives the same datasheet with one deadline for offers, and we pass every offer to the holder with its full terms as the buyer stated them. That is what makes the test in section 8 usable: an exclusive offer arrives beside the non-exclusive offers for the same records, by the same deadline, so the holder can see what exclusivity is actually being priced at against what sharing would bring.

Our mandate has an exclusivity of its own, and it is a different thing. For 2 months we are the holder's only agent for the data: the holder does not appoint another broker or list the data on an exchange in that time, though it may still deal itself with buyers it was already talking to. That exclusivity is about who markets the data, not who may use it, and after it ends the mandate continues without it. Whether any buyer gets exclusive use of the data is decided only in the licence, by the holder.

Our fee is a share of what the holder is actually paid, so an exclusivity premium the holder chooses to accept is part of the price the fee is counted on, band by band: 15% of the first $250,000, 10% from $250,000 to $1 million and 6% above $1 million. No buyer pays us in a sale we run, so our interest is the holder's: the highest total the records can earn, whether from one exclusive licence or several shared ones. The full terms are on the pricing page and in how it works.

13. Checklist before you agree to exclusivity

• Name the form. Is the request for sole use, sole distribution through one channel, a window, or a first look? Write down which.
• Reserve your own use. Make sure the licence keeps the company's right to use its records, including to train its own models.
• Narrow the field. Exclusive for one use or one kind of model, with every other use free to license.
• Set the window. A start date, an end date and what the buyer has afterwards.
• Fix the data. Records to a date and named systems; decide whether future deliveries and derived data are inside or outside.
• List what is already out. Earlier licences, vendors that trained under their own terms, legal disclosures.
• Get two prices. The non-exclusive price, and exclusivity as a separate option on top.
• Count your buyers. Compare the premium with the number of buyers that could realistically license the same records over the window (section 8).
• Take cash or a minimum. No exclusivity paid only by a revenue share; a lapse if the minimum is not paid.
• Price the exit. A conversion fee or refund that ends exclusivity early.
• Limit the prior-licence warranty to earlier exclusive grants.
• Check every channel's terms. If an intermediary asks for exclusivity, ask how many months, in which field, with what minimum, and what happens if nothing sells.
• Put it in the signed licence. Not in an email, a call or a term sheet alone.

Questions people ask

What is the difference between an exclusive and a non-exclusive data licence? Under a non-exclusive licence the holder may license the same data to other buyers and keep using it; under an exclusive licence it promises not to license the same rights to anyone else within the licence's field and term, and may give up its own use too. US copyright law treats an exclusive licence as a transfer of the rights it covers, but not a nonexclusive license.

How much more is an exclusive data licence worth? It depends on what the exclusivity covers. Platforms pay suppliers about 1.25 to 2 times for exclusive distribution; the one same-buyer pair, 23andMe and GSK, shows about 2.5 times with a different scope; founders selling training environments put sole use at roughly 4 to 5 times.

Should I license my data exclusively or non-exclusively? Usually non-exclusively, with exclusivity offered as a separately priced, time-limited option for a defined field. A non-exclusive licence can be sold again, as Reddit did to Google and OpenAI; exclusivity is worth taking when its premium exceeds what the realistic number of other buyers would pay over the window.

What is an exclusivity window? A period during which one buyer has sole rights, after which the data opens to others. Disney's three-year licence to OpenAI carried about one year of exclusivity. A window sells a head start and keeps later licences.

Can I keep using my own data under an exclusive licence? Only if the licence reserves it. A strict exclusive licence can exclude the holder too; a sole licence excludes other licensees but not the holder. Write the company's internal use, including training its own models, into the licence.

What does it mean when a buyer asks me to warrant the data was never licensed before? It is a backward-looking form of exclusivity: micro1's terms ask the holder to warrant it has not previously entered into any transaction granting a third party rights to the same data on similar terms. Ask for it to cover only earlier exclusive grants, so that a non-exclusive licence elsewhere is not a breach.

Can a buyer pay to end exclusivity later? Yes, and the price can be large: Tempus AI cut its future payments by $7.9 million in exchange for waiving exclusivity on a licence. Write the conversion price into the licence at the start.

Does an intermediary's exclusivity mean only one buyer can use my data? No. An intermediary's exclusivity usually means only that intermediary may sell the data, often to many buyers: Kled takes a licence exclusive to us and announced a non-exclusive licence to a buyer. Set the length, the field and a minimum.

What this means for you

Treat exclusivity as a priced option, never as a default. The word covers four different promises (sole use, sole distribution, a window and a first look), each sized by four dimensions (field, term, who is excluded and which data), and the evidence puts its value anywhere from about 1.25 times to 4 to 5 times a non-exclusive licence depending on that scope. Before you agree to any of it, get a non-exclusive price, ask for exclusivity as a separate number, count the buyers that could license the same records over the window, and keep your own use and your future records outside the grant.

If you hold an offer that asks for exclusivity, run it through the offer check, which places its amount against the market and lists that buyer's published clauses, and read each clause against the licence terms register. The negotiation page carries every lever cited here with its sources, and the valuation shows the exclusive figure beside the non-exclusive one, binds no one and needs no upload. The rest of the licence is in what a data licensing agreement contains, payment timing in when AI data buyers actually pay, the buyers' terms side by side in buyer programs compared, and the whole process in how to sell company data to AI labs. The rights hub holds the legal background, and how it works shows how a sale we run puts exclusive and non-exclusive offers side by side before the holder decides.

This guide is research from public sources, not legal advice.