Short answer
A data license grants a buyer defined rights to use data while the company keeps the data and its own right to use it. A data sale transfers the data and the rights in it outright. Almost all AI training data deals are licenses. The difference decides whether you can license the data again and what the buyer may do with it.
People talk about "selling data to AI companies." Almost none of those deals are sales in the legal sense. They are licenses, and the distinction is not a technicality. It decides what you keep, what the buyer gets and what you can do next.
Key takeaways
- A license grants permission to use data on stated terms. The company keeps the data.
- A sale (an assignment) transfers the data and the rights in it to the buyer.
- AI training data deals with operating companies are almost always licenses.
- An exclusive license can limit you nearly as much as a sale during its term, so read the exclusivity clause closely.
- Privacy law uses the word "sale" differently. A license paid in cash can still be a "sale" of personal information under statutes such as the CCPA.
The two structures side by side
| Data license | Data sale | |
|---|---|---|
| What the buyer gets | Defined rights to use the data | The data and the rights in it |
| What you keep | The data, and your right to use it | Usually nothing, unless rights are licensed back |
| Can you keep using it? | Yes | Only if the contract says so |
| Can you license it again? | Yes, unless the license is exclusive | No |
| Duration | A stated term, or perpetual if agreed | Permanent |
| Limits on the buyer | Permitted uses, restrictions, sometimes audits | Few or none |
| Where it is common | AI training data deals with operating companies | Bankruptcies, wind-downs, acquisitions |
What a data license is
A data license is a contract in which the company that holds data gives another party permission to use it for stated purposes, for a stated period, on stated terms. The company keeps its records and keeps running its business with them.
A license is defined by its limits:
- Scope. Which records, from which systems and dates.
- Permitted use. For example, training and evaluating AI systems.
- Exclusivity. Whether anyone else can get the same rights.
- Term. How long the rights last.
- Sublicensing. Whether the buyer can pass rights on to others, such as the AI labs it supplies.
See key terms in an AI data licensing agreement.
What a data sale is
In a sale, the company transfers the data and its rights in the data to the buyer. Afterward the seller has no right to use or share it unless the contract grants something back.
True sales happen mostly when a business is ending. In bankruptcies and wind-downs, data is sold as an asset along with everything else. The proposed $10 million Spirit Airlines data transaction described by Frankfurt Kurnit arose from a bankruptcy auction of this kind.
An operating company almost never wants this. It needs its own records to run.
A note on "owning" data
Data is not owned the way a truck is owned. In practice, "owning" business data means three things: you control access to it, you hold whatever intellectual property and trade secret rights attach to it, and you are not barred by contract or law from sharing it.
That is why the first question in any deal is not ownership in the abstract. It is whether your company has the right to grant the specific uses the buyer wants. See whether it is legal to sell company data to AI companies.
Why AI data deals are licenses
The company needs its data. A license lets it be paid without giving anything up operationally.
The buyer needs rights, not custody of the original. AI developers need permission to copy and use the records for training. A license grants exactly that.
A license can be limited. The company can restrict use, require de-identification and set a term. A sale gives up that control.
A license can sometimes be repeated. Under a non-exclusive license, the same dataset can be licensed to more than one buyer.
DataRelai's deals are structured this way. Deals are licenses: the buyer gets defined use of the data, and your company keeps it and keeps using it.
Where a license can feel like a sale
Two terms can narrow the gap.
Exclusivity. Where a license is exclusive, the buyer has the exclusive right to license the data to AI labs during the agreed term. You keep using the data for your own operations, but you cannot license it to anyone else until the term ends, and any right to license it again afterward comes from the agreement. See exclusive vs. non-exclusive data licenses.
Perpetual, irrevocable rights. A license with no end date and no way to terminate gives the buyer something close to permanent rights. There is also a practical point: data that has been used to train a model cannot realistically be taken back out, whatever the contract says about the term.
Neither term is wrong in itself. Both should be priced and understood.
The privacy law trap: "sale" has a second meaning
In commercial terms, your deal is a license. Privacy statutes ask a different question.
Under the California Consumer Privacy Act, a "sale" of personal information includes making it available to a third party for money or other valuable consideration. A paid license of records that contain personal information can meet that definition even though ownership never changes. The consequence is a set of notice and opt-out duties.
The usual answer is de-identification that meets the statutory standard, so that what is licensed is no longer personal information under the law. See de-identification for AI training data.
So when a contract says "license," do not assume privacy law agrees. Have your attorney check both.
Tax and accounting
A license payment and a sale of an asset can be treated differently for tax and accounting purposes, and an exclusive, perpetual license can blur the line. Ask your accountant or tax adviser before signing.
Frequently asked questions
If we license our data, do we still own it?
You keep the data and your right to use it. The buyer gets only the rights the license grants.
Can we license the same data to more than one buyer?
Yes, if the license is non-exclusive. An exclusive license prevents it for the agreed term.
Can the buyer resell or pass on our data?
Only if the license allows sublicensing. Data buyers that supply AI labs need that right, so expect the clause and read its limits.
Why does everyone say "sell data" if it is a license?
Because it is the everyday way to describe being paid for data. The contract, not the phrase, determines what actually changes hands.
Sources
This guide is general information, not legal, tax or financial advice. Whether a particular dataset can be licensed depends on what it contains and on obligations your company already has.