Home Guide
AI data licensing: how companies license data to AI labs
What a data license is, who is in the market, the kinds of licenses and how they are priced, explained for the company on the licensing side.
Last updated October 2026 · 6-minute read
What is AI data licensing? AI data licensing is a contract in which a data owner grants an AI lab, or a company that supplies labs with data, the right to use a defined dataset for training, evaluating or fine-tuning AI models. The owner keeps the data. The license sets the sources and volumes, the permitted uses, how personal data is removed, whether the license is exclusive, how long it lasts and how the owner is paid.
Who is in the market
- AI labs train frontier models and buy data directly or through suppliers.
- AI data companies source, prepare and deliver datasets to the labs, including work data from ordinary businesses.
- Data marketplaces list datasets from many sellers and match them with buyers.
- Advisers work on the seller’s side: they check the data and rights, find the buyer that fits and help on terms.
- Licensors are the companies that hold the data: publishers, platforms, and increasingly operating companies with years of internal work history.
Kinds of data licenses
Licenses differ by what the buyer may do with the data:
- Training: the data is used to train or pre-train models.
- Evaluation: the data is used to test models, often without being trained on.
- Fine-tuning and agent training: the data teaches models to do specific kinds of work, such as handling a ticket or reviewing code.
Each use should be named in the license. Anything not named is not permitted.
Exclusive or not, one-time or recurring
| Choice | What it means for the licensor |
|---|---|
| Non-exclusive | The same data can be licensed to more than one buyer. |
| Exclusive | One buyer only, usually for a higher price. It is safest when limited in time and to the data actually delivered. |
| One-time | A single delivery of historical data. |
| Recurring | Fresh data delivered on a schedule, which some buyers value more than one archive. |
| Fixed term | The buyer may use or keep raw data for a set period, then deletes it. |
How data licenses are priced
There is no public price list. Most licenses combine a few common structures:
- a fixed fee for a defined dataset;
- pricing by volume, such as messages, tickets, documents or hours of recordings;
- staged payments, with the main payment after the buyer accepts the delivered data;
- recurring payments for data delivered on a schedule.
What drives the price is the data itself: years of history, how much of the work is visible, how much is in English, how clean the rights are and how much must be excluded. A first offer is rarely the best one, which is where knowing what similar datasets go for matters.
What the licensor keeps
- Ownership of the data, and the right to keep using it.
- The final say on scope: each source is approved in writing.
- Deletion and return terms: how long the buyer may keep raw data.
- Confidentiality of the company’s identity, if it is agreed in the license.
Common questions
Do we need our own lawyer?
Yes. The license is between your company and the buyer, so your counsel should review it. The terms that decide most of the value are the liability cap and indemnities, the warranties you give, exclusivity, acceptance, partial rejection and payment timing.
Who does the export and de-identification?
Usually the buyer’s delivery team, under the license, with access you grant and can see. The license sets what personal data must be removed or masked before delivery, who does it and how it is checked.
Can a company outside the US license its data?
Often, yes. Many buyers prefer US-based companies, and some also accept companies registered in the UK, the EU or Canada. Which entity signs, and which entity owns the data, should be settled before the license.
This guide is general information, not legal or tax advice.