Licensing or selling: what is the difference?
When you license data, you keep ownership and give a buyer permission to use it on set terms. When you sell (assign) it, ownership passes to the buyer for good, and you generally lose any say over what happens next.
What does licensing mean?
A license is permission. The owner keeps legal title and lets someone else use the material within limits: for a purpose, for a period, in a region, or exclusively or not. Licenses can end on the terms the contract sets.
For company records, a license usually says what data is covered, what the buyer may use it for (for example, training AI models), whether it is exclusive, how long it lasts and what the buyer must do with the data when it ends.
What does selling mean?
Selling, called an assignment in legal terms, transfers ownership. The new owner can generally do what it likes with the material, and the former owner can no longer use it unless the new owner allows it. Assignments are normally permanent and are usually paid as a lump sum.
News coverage in 2026 often uses "sell" loosely for any deal where a company receives money for its archives. The contract is what matters, not the headline word. Some deals described as sales are actually licenses, and the reverse.
How do they compare?
The general differences, drawn from standard intellectual property practice:
- Ownership: licensing keeps it with you; selling moves it to the buyer.
- Control: a license can limit uses; after a sale, limits exist only if the contract keeps them.
- Term: a license can expire; a sale is generally permanent.
- Payment: licenses may pay once or over time; sales usually pay once.
- Reuse: under a non-exclusive license you may license the same data again; after a sale you usually cannot.
Is there a catch with AI training specifically?
Yes. Once data has been used to train a model, it is difficult to "take back." A license that ends after two years may still let the buyer keep models already trained on your records. Read what the contract says about deletion of the raw data, about models trained before the end date, and about passing data to others.
Because of this, even a license can feel close to a sale in practice. Make sure the price reflects what you are really giving up.
What terms should I look for in either kind of deal?
Whether the contract is a license or a sale, a few terms decide how much you really give up. Look for clear answers to these:
- Exactly which records are covered, and which are excluded.
- Permitted uses, and whether the buyer may pass the data to anyone else.
- How long the rights last, and what happens to raw data at the end.
- Whether models trained on your data may be kept and used after the end date.
- What you promise about the data (for example, that you hold the rights to it) and what happens if that turns out to be wrong.
- Who is responsible for removing personal information, and to what standard.
- How and when you are paid, and what happens if payment is late.
Which is better for my company?
It depends on your situation. A closing company may prefer a clean, final deal so that nothing is left to manage, and buyers may prefer that too. An operating company often prefers a license so it keeps ownership of its own history and can set limits on use.
Either way, you can only license or sell what you actually hold the rights to. Records that include customer information, partner materials or employee data may be limited by contracts and privacy law. Ask your lawyer to review the deal before you sign, and your accountant about tax treatment.
One-time payment (7 buyers)
Licenses a defined copy of your records and pays an agreed amount once, usually on delivery or acceptance.
Target: Owners who want one known amount.
Revenue share (3 buyers)
Prepares your records and pays you a share of each sale to AI companies, with no guaranteed total.
Target: Owners comfortable trading certainty for repeat payments.
Per repository (1 buyers)
Pays a published rate per code repository or codebase.
Target: Companies that built their own software.
Priced after scoping (6 buyers)
Publishes no figure. You get a number only after it has reviewed your records, then terms in writing.
Target: Owners ready to compare written offers.
Common questions
If I license my data, can I still use it myself?
Usually yes, unless the license says otherwise. Check exclusive licenses carefully.
Can a license be cancelled?
A license ends or can be ended according to its terms. Models already trained may not be affected, so read those clauses.
Does selling mean I lose my own copy?
Not necessarily, but you may lose the right to use or share it. The contract decides.
Which do buyers prefer?
Buyers vary. Each buyer's published terms are summarized on its page on this site.
Sources
- Key Differences Between IP Licensing and Assignment Explained (Sprintlaw), accessed 2026-10-15
- Exclusive vs. Non-Exclusive Licenses (Copyright Alliance), accessed 2026-10-15
- AI labs are buying data from dead startups to train their models (MIT Technology Review, Oct 13, 2026), accessed 2026-10-15
- Failed Companies Are Selling Old Slack Chats and Email Archives to Train AI (Gizmodo, Apr 17, 2026), accessed 2026-10-15
General information, not legal, tax or financial advice.