Contracts and money

Data licensing terms, explained

A data license sets what a buyer may do with your data, for how long, who else can buy it and what it pays. Here is what each term means for a seller.

·6 min read

A data license gives an AI company the right to use a copy of your data for named purposes, for a fixed period, while you keep ownership. Five terms decide most of its value: permitted uses, length, exclusivity, price, and what the buyer must never do with the data.

This page is general information, not legal advice. Have a lawyer read any license before you sign it.

You license the data, you don't sell it

Most data deals with AI labs are licenses. The buyer gets a copy and a set of rights, and you keep the original along with every right you didn't grant. That matters for two reasons. You can license the same dataset to more than one buyer unless you agree otherwise, and you can stop new deliveries when the term ends.

Public deals follow the same shape. When Reddit filed to go public, it disclosed $203 million in data licensing contracts signed in January 2024, with terms of two to three years. Reddit still owns its data, and the buyers hold time-limited rights to use it.

Permitted uses

The license lists what the buyer may do. For AI training the usual uses are:

Use What it means
Pretraining The data goes into the large first training run of a model
Post-training or fine-tuning The data shapes a model that already exists, often for one domain or task
Evaluation The data is held out to test models and never trained on

A narrow grant, such as evaluation only, usually pays less than a broad one. Write down every use you allow. Anything the license leaves out is something the buyer may not do.

Our buyer license names its permitted uses from these three. It also bans the buyer from re-identifying anyone, reselling the data, or combining it with other data to learn more about the people or companies in it.

Length

A license runs for a set term, then ends unless both sides renew. Two questions come up at the end:

  1. Raw data. The buyer should delete its copy of your raw records when the license ends. Our license requires that.
  2. Trained models. A model trained during the term keeps what it learned. Data already used in training can't be pulled back out of a model, so assume anything delivered stays in the models built during the license.

Shorter terms let you reprice sooner. Longer terms give a buyer more reason to pay for history up front.

Exclusivity

A non-exclusive license lets you sell the same data to other buyers. An exclusive license promises you won't. Exclusivity can be narrow, such as one field, one region or one year of records, or it can cover the whole dataset.

Exclusivity costs the buyer more because you give up other sales. For reinforcement learning environments, two founders interviewed by Epoch AI estimated that exclusive deals run roughly four to five times the price of non-exclusive ones. No comparable public figure exists for business records, so treat that ratio as a signal from one market.

Our licenses are non-exclusive by default. A buyer can negotiate exclusivity for a field or a time period, and you decide whether the price is worth the sales you give up.

Price: history and refresh

A license can carry two prices:

  • The history price pays once for every past record the source holds, up to the first delivery. See the history license definition.
  • The refresh price pays for each new monthly batch the buyer accepts while the refresh term is active.

The refresh price turns a one-time sale into monthly income. After the first license, our SDK pulls only the new records since the last run, scrubs them on your machine and uploads a new batch. Each license with an active refresh term pays its refresh price for every accepted batch. Recurring data revenue covers the full model, and our recurring revenue guide works through it with numbers.

Restrictions on the buyer

The restrictions protect the people and companies in your data, and they often decide whether the deal is legal at all.

  • No re-identification. The buyer may not try to work out who a record is about. Under California law, data only counts as deidentified if the business contractually obligates recipients not to re-identify it, so this clause is a legal requirement for many sellers.
  • No resale. The buyer can't pass your data on as a product.
  • No combining. The buyer can't merge your data with other sources to learn more about the people in it.
  • Flow-down. Anyone the buyer shares the data with, such as a contractor, is bound by the same terms.
  • Deletion. Raw data is deleted at the end of the term.
  • Audit rights. You, or someone acting for you, can check that the buyer follows the license.

If a buyer pushes back on any of these, ask why. Each one is cheap for an honest buyer to accept.

Rights you promise as the seller

The license also binds you. You promise that you have the right to license the data, and the buyer relies on that promise. If your customer contracts say their data may be used "only to provide the service", a scrubbed copy may still be off limits. We check contracts in a rights review before any data moves, and can I sell data I hold for my customers explains the common blockers.

Your seller agreement with a broker sits alongside the buyer license. Ours confirms you hold the rights, sets the commission, gives you approval of every buyer, and says what happens if you withdraw. You can rule out any company, including competitors. Withdrawal stops future sales, but data already delivered and trained on stays in those models.

Where personal data is involved

If any personal data remains after scrubbing, privacy law adds its own terms. Under the GDPR, a buyer that decides how to use the data for its own model is a controller, and Article 28(10) treats a processor that sets its own purposes as a controller too. Data leaving the UK or EEA for a US lab needs a transfer mechanism such as standard contractual clauses. Our guides on GDPR and the CCPA cover both.

A checklist before you sign

  • Uses are listed, and anything unlisted is barred.
  • The term has an end date and a deletion duty for raw data.
  • Exclusivity, if any, is limited to named fields or periods and priced accordingly.
  • The history price and refresh price are separate lines.
  • Re-identification, resale and combining are banned, and those bans flow down.
  • You have audit rights.
  • You approved this buyer by name.

See what your data could earn

Our calculator gives a range from your industry, systems, years of records and rights, and we confirm it on a 30-minute call. If you want us to handle buyers and terms, start at sell data to AI companies.

Frequently asked questions

Is a data license the same as selling my data?

No. A license grants named rights to use a copy for a set period while you keep ownership. You can usually license the same data to several buyers unless you agree to exclusivity, and the buyer's rights end when the term ends.

Should I agree to an exclusive license?

Only if the price covers the sales you give up. Exclusivity can be limited to one field or time period, which keeps the rest of the dataset open to other buyers. Our licenses are non-exclusive unless you and the buyer agree otherwise.

What happens to my data when the license ends?

The buyer must delete its copy of the raw data. Models already trained on it keep what they learned, because data can't be removed from a trained model, so decide what to license with that in mind.

Can I stop a buyer from using my data against my business?

You approve every buyer before a license is signed and can rule out any company, including competitors. The license also bans resale and combining your data with other sources.

Find out what your records are worth.

Value my data