Selling vs. licensing data: what the difference means for your company

Most business owners hear "sell your data" and picture handing over a hard drive and never seeing it again. That is a sale, and it is almost never the right structure. Nearly every AI data deal worth doing is a license, and the difference decides what you keep, what you can do next, and what you get paid.

Sale versus license: the definitions

A sale transfers ownership. Once the deal closes, the buyer owns the records, can use them however the contract allows (often without limit), can resell them, and owes you nothing further. You cannot license the same records to anyone else.

A license grants permission. You remain the owner. The buyer receives a defined set of rights: to use a defined dataset, for defined purposes, for a defined time, under defined restrictions. Anything not granted stays with you. When the term ends, the rights end.

The distinction sounds academic until something changes: a second buyer appears, a regulator asks questions, or the buyer wants to use the data in a way you never imagined. Under a license you have an answer. Under a sale you have a memory.

Why Knowledge Fuel structures deals as licenses

We work for the seller, and a license protects the seller in four ways.

First, you keep the asset. Operational data gains value as it grows, and an outright sale gives up every future deal. Second, you keep control. A license spells out who can use the data and for what, and you approve every buyer and every permitted use before anything moves. Third, you limit liability. Restrictions on re-identification, resale and retention are only enforceable while you still have a contract with the party using the data. Fourth, you keep options. A non-exclusive license today does not prevent an exclusive one tomorrow.

Exclusive versus non-exclusive licenses

An exclusive license promises the buyer that nobody else will be licensed to use the same data, for the term and often within a defined field of use. A non-exclusive license makes no such promise, so you can license the same dataset to several buyers.

Pricing follows the promise. Exclusivity commands a premium, often a large one, because the buyer is paying for scarcity: a model trained on data its competitors cannot obtain. The cost is that exclusivity closes the door. One payment, one buyer, and the data is off the market for the term.

Non-exclusive deals price lower per buyer but can be repeated. Public deals show the pattern. Taylor & Francis gave Microsoft non-exclusive access to archive content for an initial data access fee of about $10 million, with recurring annual fees over the following three years, according to Informa's 2024 half-year report. For a mid-market company, non-exclusive is usually the right starting point. Grant exclusivity deliberately, at a price that reflects what you give up.

Permitted uses and restrictions

The permitted-use clause is the heart of a data license. It should answer three questions in plain terms.

What can the buyer do? A typical grant is training-only: the buyer may use the data to train, evaluate and improve machine learning models. It may not publish the data, display it, or build a product that reproduces it.

What can the buyer not do? Standard restrictions include no resale or sublicensing, no sharing with affiliates unless named, no attempts to re-identify any individual, and no combining the data with other sources to defeat de-identification. Our contracts also require encrypted transfer, access logging, and deletion of the raw data at the end of the term, with audit rights for the seller.

What happens if the buyer breaks the rules? The contract should state remedies: termination, destruction of the data, and damages. We work alongside your counsel on every clause and never promise legal outcomes.

Term, renewal and revenue models

Term and renewal

A term is how long the buyer's rights last. Common structures run one to five years. Shorter terms let you reprice as the market moves. Longer terms give the buyer certainty and usually carry a higher total price. Renewal can be automatic, optional or renegotiated.

One wrinkle specific to AI: a model trained during the term does not forget the data when the term ends. Contracts handle this by separating the right to keep using a trained model (often granted) from the right to keep the raw data or train new models (often not).

Revenue models

A one-time fee is the simplest: a single payment for the license, often for an exclusive or long-term grant. Wiley recognized a $23 million content rights project in a single quarter of fiscal 2024, a limited-duration license of previously published content, according to its fiscal 2024 results.

A recurring fee pays annually or quarterly for continued access, usually with fresh data added each period. Reddit's arrangement with Google was reported by Reuters at about $60 million per year.

Per-record or per-hour pricing ties payment to volume delivered and suits datasets that grow steadily, such as call recordings or tickets.

Those are large companies with negotiating power. For a typical mid-market dataset, our illustrative range is $25K to $1M+, and the actual value depends on uniqueness, quality, rights and buyer demand.

Questions to raise with your tax and accounting advisors

We are not tax or accounting advisors, and answers vary by entity type and state. Bring these questions to yours before a deal closes.

• Is license income treated as ordinary income, and would an outright sale be treated differently?
• How is a one-time fee recognized compared with recurring payments over a multi-year term?
• Does licensing create or change an intangible asset on the balance sheet?
• Do sales tax, use tax or digital goods rules apply to licensing data in your state or the buyer's?
• If the buyer is outside the United States, does withholding tax apply?
• Would holding the data in a separate entity change any of the answers above?

Which structure fits: a decision table

Our default is a non-exclusive, training-only license with a defined term, and we move to exclusivity only when a buyer pays for it and you agree. To see how the process runs from first conversation to payment, read how it works. To get an illustrative range for your own records, start with a free estimate.