Selling data transfers ownership, so the buyer controls it and you cannot sell it again. Licensing grants permission to use a copy for defined purposes while you keep ownership and can license it to others. Most AI data deals are licenses, because data can be copied and suppliers want to keep control and reuse.
The short version
People say they want to sell their data, but in most AI deals nothing is sold in the legal sense. A sale, usually called an assignment, transfers ownership of the dataset and the rights in it. A license keeps ownership with the supplier and grants the other side permission to use a copy on stated terms.
The difference decides who controls the data afterward, whether you can earn from it again, and how the price is set. It is worth getting the vocabulary right before the first conversation with a buyer. A term sheet that says purchase may describe a license, and one that says license may be drafted so broadly that it works like a sale.
Ownership
Data ownership is less tidy than ownership of a building. Facts on their own are generally not owned. What a company holds is a mix of rights: copyright in documents and code, database rights in some jurisdictions, trade secret protection, and contractual control over access. Selling data means assigning those rights, and usually handing over the only authoritative copy or agreeing to stop using it.
Under a license none of those rights move. The licensee receives a copy and a permission. If the licensee goes beyond the permission, it is in breach of contract and may be infringing your rights. Outrich Data's commitment to suppliers follows this model: a license covers a copy of the data for defined uses, and the data stays with the supplier's company.
Control
A buyer who owns data can use it for anything, pass it to anyone, and keep it forever, subject only to law. The seller has no say. That matters when the data came from your operations and still reflects your customers, staff and methods, even after de-identification.
A license lets you set conditions. You can limit the purpose, for example model training but not resale of the raw data. You can limit who receives it, how long they may keep it, and whether they may name you. You can require deletion of the dataset at the end of the term. These controls are only as good as the drafting, so the scope clause deserves the most attention.
One limit applies to both structures. Once a model has been trained on data, that training cannot practically be reversed. A license can require deletion of the dataset, but it cannot pull the data's influence out of a finished model.
Reuse
Data is not used up when someone copies it. The same archive can be licensed to several buyers, and each license is a separate source of income. A sale ends that. So does an exclusive license for as long as it runs, which is why exclusivity is a commercial term and not a formality.
Reuse also covers your own operations. A licensor keeps using its data in the business as before. A seller may need a license back from the buyer just to keep running its own systems on its own records, and that license back has to be negotiated.
Price
A sale usually means a single payment for everything. A non-exclusive license often pays less per deal, because the buyer knows others can obtain the same data, but it can be repeated. An exclusive license sits between the two. The licensee pays a premium for being the only one with access during the term.
Reported figures show both ends of the market. MIT Technology Review reported on October 13, 2026 that closed startups were selling a code base for about $25,000 and workspace data for about $15,000. Deals like these are closer to outright sales, since a company that has shut down has no further use for its data. Among operating companies, a Neudata market analysis dated June 10, 2025 put the Google-Reddit licensing agreement at about $203M in total contract value. Reddit kept operating and kept its data.
License vs sale vs exclusive license
The table compares the three structures on the points that matter most to a supplier. Real agreements vary, and a contract's label does not always match its effect, so read the terms and not the title.
| Point | Non-exclusive license | Exclusive license | Sale (assignment) |
|---|---|---|---|
| Ownership | Stays with you | Stays with you | Transfers to the buyer |
| Your own use | Continues | Continues unless restricted | Ends unless licensed back |
| Licensing to others | Allowed | Blocked for the field and term | Not possible |
| Control over use | Set by the scope clause | Set by the scope clause | None after closing |
| Price pattern | Lower per deal, repeatable | Premium for exclusivity | One payment for everything |
| At the end of the term | Dataset deleted or returned | Dataset deleted, rights free again | No end. The buyer keeps it |
| Best fit | Operating companies with reusable data | A buyer that needs a competitive edge | A business that is closing or exiting a line |
Why most AI deals are licenses
AI labs need the right to copy and process data for training. They rarely need to own it. Ownership would bring obligations they do not want, including responsibility for the underlying records and for the rights of the people and customers in them. A license gives them what training requires and leaves the rest with the supplier.
Suppliers prefer licenses for the mirror-image reasons. An operating company cannot hand over its customer records and stop using them. It also wants to keep the option of licensing to other buyers as demand develops. Privacy law pushes the same way. Several regimes attach specific duties to selling personal information, and a narrow, documented license of de-identified data is easier to defend.
For these reasons Outrich Data licenses data from suppliers and does not buy it outright. The supplier approves each dataset and scope in writing, keeps using its data, and keeps the right to license it to others.
When a sale makes sense
A sale can be the right structure when the business is closing, when a product line has been shut down and the data has no further internal use, or when the data is being transferred as part of selling the business itself. In those cases there is little to protect by keeping ownership, and a clean transfer may be simpler for both sides.
Even then, the seller's obligations to customers and employees do not disappear. Contracts, privacy notices and privacy law still limit what can be transferred and to whom. Check them with counsel before agreeing to assign anything.