What is AI data licensing?

How does an AI data licence work?

A licence is a permission, not a transfer. The data owner, called the licensor, grants the buyer, the licensee, a right to use a defined copy of its records for a stated purpose. The original records stay with the owner.

A licence has to pin down what is on offer: which sources, which date ranges, what kind of records and what the buyer may do with them. When we work with a company, it decides which sources, date ranges and uses are included before it signs, and nothing is licensed until it has agreed the terms.

Most licences answer the same handful of questions. These are the clauses we read first:

• Permitted use: training, testing, fine-tuning or something wider. Vague words cause disputes. Proskauer describes a 2025 US lawsuit over whether a 2021 licence limited to "internal research purposes" allowed the data to train a commercial AI product.
• Term: how long the licence runs, whether either side can end it early, and what happens at renewal.
• Territory and sublicensing: where the copy may be used, and whether the buyer may pass its rights on to others.
• Deletion and trained models: what happens to the copy at the end, and whether models already trained on it may be kept.
• Confidentiality: whether the buyer may name you, or try to identify people in records that had personal details removed.
• Payment: how much, and when it falls due.

Why do AI companies license data instead of just taking it?

Much of the data AI companies now want is private: records of real work that were never published. OpenAI's Data Partnerships programme asks for large datasets "not already easily accessible online to the public today", and for data that "expresses human intention", such as long-form writing or conversations rather than disconnected snippets. Private records can only be used with the owner's agreement, and that agreement is a licence.

A licence also gives the buyer legal cover. In its 2024 filing to list on the stock market, Reddit complained that some companies had built large language models on its data "without entering into a license agreement with us", and said it planned to enforce its terms. The same month, Reuters reported that Google had agreed to license Reddit's content for about $60 million a year. A signed licence protects both sides.

What is the difference between licensing and selling data?

In a licence, you grant a right to use a copy and keep the original. In a sale, the buyer takes the data itself as an asset. Reuters reported in August 2026 that Google bid $10 million to buy Spirit Airlines' business data in its bankruptcy, a sale that still needs court approval.

The line can blur. A perpetual, irrevocable licence lets the buyer use the copy for ever, even though you keep the title. That is why we read the terms, not the word on the cover page.

With Pangea.ai, it's a licence, not a sale: the data is licensed for a set use and term, and the company is paid for the licence and keeps the original.

What types of AI data licence are there?

Two choices shape most licences: whether the buyer gets the records to itself, and how long its rights last. A buyer that wants exclusivity is asking you to give up every other deal while it lasts, and in our view the price should reflect that. Our guide to what company data is worth covers what moves the price.

No model is standard across the market. Each deal sets its own terms.

How are AI data licensing deals paid?

Few contracts are public, but reported deals fall into three patterns. They are examples, not prices to expect.

Which structure fits depends on the data. Reddit's filing said its data is valuable because it "constantly grows and regenerates", which suits a yearly fee. A finished archive, such as the records of a company that has stopped trading, has nothing new to deliver, so a single payment fits it better.

Recurring deals also end. In July 2026, CNBC reported, citing The Wall Street Journal, that Reddit's $60 million a year deal with Google was ending soon and that the two were in talks about renewing it.

• Flat fee: one agreed price for a defined dataset, delivered once.
• Recurring fee: a yearly fee over a fixed term, often with fresh data. Google's licence for Reddit content was reported at about $60 million a year, and Reddit's filing disclosed $203 million of licensing contracts over two to three years, with continuous access and quarterly data transfers.
• Mixed: cash combined with credits or other value. The Wall Street Journal estimated News Corp's multi-year deal with OpenAI at $250 million over five years, in cash and credits.

Who is involved in an AI data licensing deal?

There are usually two main parties. The data owner, or licensor, holds the records and grants the rights. The buyer, or licensee, is an AI lab, a data company that supplies AI labs, or a tech company building its own models. Our guide to who buys company data for AI covers each type.

Some deals also have an intermediary that finds buyers or prepares the data. Ask early who negotiates with the buyer and who holds the contract.

Pangea.ai starts on the owner's side. We look at what a company holds and tell it whether the records could qualify and what affects their value.

Lawyers matter on both sides, because the owner must have the right to license the records in the first place. Customer contracts and personal data rules are the usual limits.

What are the ways to license your company data?

There are four main routes. They differ most in who you deal with, who pays for what, and how much of the work you do yourself.

If you already have a buyer in mind, going direct is the shortest path. If you want to compare buyers and are happy to prepare the data yourself, a broker fits. Gameplay, video and robot data have their own specialist buyers. If you want someone to tell you what your records are worth and run the process, including removing personal details, start with a free check.

Whichever route you take, ask the same three questions: is the licence exclusive, who removes personal data, and who holds the contract with the final buyer.

What should a company check before licensing its data?

When we review a company's data, size alone decides little. Depth does. These are the things we look at first, as set out on our page about how to sell your company data to AI companies:

• Years of history: a long-running record shows how decisions played out over time, not just a single moment.
• Outcomes attached: closed deals, resolved tickets and paid invoices let a model connect a decision to its result.
• Reasoning written down: threads where people explain why they chose something, not only what they did.
• Specialist work: decisions from finance, legal and engineering teams are among the records buyers name most often.
• Ownership: only records your company has the rights to can be licensed.

What stays out of a data licence?

Data you hold on behalf of clients is out of scope, as is anything the company doesn't have the rights to. Buyers set limits of their own: OpenAI says it is not seeking datasets with sensitive or personal information, or information that belongs to a third party. Our guide to the company records AI companies buy goes through each record type and what usually has to come out.

How do you start licensing your company data?

Our process has four steps, and nothing is licensed until you've agreed the terms. The first is the free check, with no commitment, and someone reads every answer, even if your data isn't a fit.

• Check: tell us which tools your team uses and how many years of history they hold.
• Review: we look at what you have and tell you whether it could qualify and what affects its value.
• Agree terms: you decide which sources, date ranges and uses are included before you sign.
• Get paid: the data is licensed for a set use and term. You're paid for the licence and keep the original.