If your company holds data that someone else wants to use, a data license is how you let them use it without giving it away. It's permission, in writing, for a named party to use defined data for a defined purpose, for a defined time, at a price, while you keep ownership. AI labs sign these every week, with publishers, forums, airlines in bankruptcy, and ordinary companies with years of operating records. Read on to find out what a data license covers, how it differs from selling your data, what you keep, and what the public deals show about how they're priced.
What a data license is
A data license is a contract that grants a right to use data. Three things make it a license. You keep ownership of the data, the other party gets a copy and a set of permissions, and the permissions have edges (what they can do with it, for how long, and whether anyone else can have the same deal). Software works the same way. You don't buy Excel, you license it, and Microsoft still owns it. A data license puts your company in Microsoft's seat.
The word "data" in a data license covers more than a spreadsheet. In AI deals it means the record of how a company runs (support tickets, sales histories, project boards, code, documents, messages, and the links between them). Our guide to enterprise operational data covers why that record is what buyers pay for.
The five terms every data license settles
| Term | The question it answers | What it means for you |
|---|---|---|
| Scope | Which data, from which systems, and what has to stay out | You decide what's in. Nothing is licensed by accident. |
| Permitted use | What the buyer can do with it (training, fine-tuning, evaluation, building products) | Narrow uses pay less and keep more for later deals |
| Term | How long the license runs, and what happens when it ends | A model trained during the term keeps working after it, so nearly every license says so |
| Exclusivity | Whether anyone else can license the same data | Exclusive pays a premium and closes every other door for the term |
| Payment | A flat fee, a recurring fee, a per-unit rate, or a share of each sale | Decides whether you're paid once, every year, or on every deal |
Everything else in a data license agreement, the warranties, the privacy obligations, the confidentiality clause, exists to protect one of those five. Our guide to how AI data licensing deals work goes through the rest of the terms and which ones get argued over.
How a license differs from a sale
A sale transfers the data. The buyer owns it, you don't, and there's no later deal. A license keeps the data with you, and the same dataset can be licensed to more than one buyer if the license is non-exclusive. The public market is almost entirely licenses. Spirit Airlines is the exception. Spirit is being wound down, so its operating records are being sold outright in bankruptcy court, where Google bid 10 million dollars and Micro1 then bid 12.5 million. Every publisher and platform deal on record is a license, because those companies are still operating and still producing data.
For an operating company the choice is usually easy. License, keep ownership, and keep the right to do it again. Our guide to data monetization models walks through the six ways a data deal can pay and a framework for picking one.
What the public deals show
The deals that made the news this year show the five terms at work, even though the contracts themselves are confidential.
- News Corp licensed its archive to OpenAI for a reported 250 million dollars over five years. A recurring license on a living source.
- Google pays Reddit about 60 million dollars a year for conversation data that refreshes every day, and Reddit signed a separate deal with OpenAI on the same data. Two non-exclusive licenses on one dataset.
- Amazon and The New York Times at a reported 20 to 25 million dollars a year. Recurring again.
- Thousands of creators and companies license through marketplaces like Troveo, where the license is signed once at the platform level and the owner is paid on each sale.
The pattern is that data which keeps growing gets a recurring license, and a finished archive gets a one-time license or a sale. Our guide to data monetization examples has the full list with prices.
What you keep
Under a data license you keep four things. Ownership of the data. Your own use of it, so nothing changes about how your company runs. The right to decide what's in scope and what's excluded. And, under a non-exclusive license, the right to license it again. What you give up is control over what the buyer builds from it. A model trained on your data belongs to whoever trained it, and nothing can be untrained, so every serious license says that models already trained survive the license ending. That's the one clause buyers almost never give up, and the reason the term matters more than most owners expect.
The one term that moves the price most
Exclusivity. An exclusive license can be worth several times a non-exclusive one, because it denies the data to the buyer's competitors. It also means no second deal for the term, so the premium has to beat everything you'd have earned from the other buyers. Field-of-use exclusivity, exclusive for one kind of model and open for the rest, is the common middle ground. Our guide to exclusive vs non-exclusive data licenses covers how to price that decision.
What happens before anything is delivered
A data license is signed before the data moves, and three things happen between the two. A rights review, to confirm your contracts let you license what's in scope and to pull out anything that came from third parties. Privacy work, where names, contact details, credentials and client identities are replaced with pseudonyms before any buyer sees anything. And packaging, which turns exports from your systems into something a buyer can test. If you license through a marketplace, the marketplace does all three. If you license directly, you do. Our owner's guide to licensing company data for AI covers that process step by step.
Where to start
The quickest way to tell whether your data is worth licensing is our free data value assessment (ten questions about your systems, history and industry, about five minutes). You get an estimate of what your data could license for, which tells you whether the five terms above are worth negotiating at all.
Where Troveo fits
Troveo is a data licensing marketplace. You sign one license with us, we handle the rights review, the packaging and every buyer conversation, and you're paid on every sale. You keep ownership, your company is never named publicly, and we take no fees and no deductions from your payouts. We've paid rights holders more than 20 million dollars across video, audio, gaming, robotics and business data, and we work with more than 40 active buyers. Our explainer on the Troveo business data license agreement walks through our version of the five terms. Start with the data value assessment, read our business data page for how the process works, or talk to us about what your company holds.