If your company generates data while doing its ordinary work, you may be able to earn revenue from it without building anything new. A data licensing business model is one where a company earns revenue by granting others the right to use data it already generates. The company isn't selling a product built on that data; it's licensing the data itself. In 2026 the buyers driving it are AI developers, who pay for proprietary, rights-cleared data because the public web no longer contains what their models need. Publishers, forums, stock libraries, an airline in bankruptcy, and a growing number of startups winding down have all run some version of the model. Read on to find out what it is, the four ways it pays, why a business would choose it, what it takes to run, and when it's the wrong idea.
What the model is
Two things separate a data licensing model from simply selling a database. First, it's almost always a license, and our explainer on what a data license is covers the five terms every one settles. The company keeps ownership and grants a defined right to use defined data, for defined purposes (like training or evaluation), for a defined period, exclusively or not. The same data can be licensed more than once. Second, the data is a byproduct. It was produced while doing something else (running a newsroom, hosting a community, operating an airline, building software) and it already cost money to create. The licensing model earns a second return on that spend without changing what the company does. The full mechanics of the underlying agreement (permitted uses, exclusivity, warranties) are in our guide to how AI data licensing deals work. This guide helps you with the business decision above the contract. Licensing is one of four ways companies earn from their data, and our guide to data monetization examples puts it next to the other three.
The four ways it pays
| Model | How revenue works | Example |
|---|---|---|
| One-time archive license | A single fee for a finished corpus, sometimes paid in tranches | Spirit Airlines' operational history, a 10 million dollar bid from Google in bankruptcy court |
| Recurring license | Annual or multi-year fees for a living source, often with refreshes and API access | Google paying Reddit around 60 million dollars a year |
| Exclusive premium | A larger fee in exchange for denying the data to competitors for a term | Common in publisher and archive deals. Buyers price the premium against the other deals it stops you signing |
| Marketplace revenue share | You're paid each time the data licenses through a marketplace | Thousands of creators and companies licensing through platforms like Troveo |
You can combine these structures. A publisher can run a recurring license with one lab and a non-exclusive archive license with another, and a marketplace participant can license the same catalog to several buyers over time. Which structure fits depends on whether the data is finished or still being produced, how scarce it is, and how much freedom you want to keep for later deals. Our guide to data monetization models puts these licensing structures next to the other ways data pays and gives you a five-step framework for choosing. You can find the disclosed numbers behind each structure on our AI training data statistics page, and the pricing factors in our guide to what AI training data costs.
Examples across the spectrum
At the top are media companies with famous catalogs. News Corp licensed its archive to OpenAI for a reported 250 million dollars plus over five years, and Shutterstock earned 104 million dollars from AI licensing in a single year. Owners with legal and data teams negotiated these deals directly, and the prices set the public reference points for the whole market.
In the middle are platforms and communities whose value is that they keep producing. Reddit's deal with Google is the clearest example of the recurring model: Google pays annually for a source that Reddit's users refresh every day. Forums, review sites, and developer communities sit in the same category.
Then there are operating companies whose data was never meant to be a product. Spirit Airlines' archive became a 10 million dollar asset only in liquidation. It drew a bid because it held years of connected operational records across booking, operations, and finance; our breakdown of the Spirit Airlines data deal covers the details. Not every company's data is worth 10 million dollars. Operating history now has a market value you can see, and you can engage with that market on your own terms, not only in a wind-down. Our guide to enterprise operational data maps what that category includes.
Below that are startups shutting down. Closure platforms now broker code repositories, ticket histories, and workspace archives from defunct companies, with reported deals running from about 10,000 to 100,000 dollars each. Labs and research groups buy them to train agents on how real teams worked. Our guide to selling source code to AI labs covers how buyers in that market price code.
At the widest point are individual creators and small companies licensing through marketplaces. A marketplace pools video, audio, gameplay, and now business data from thousands of owners, so that people who could never negotiate with a lab directly still get paid. Troveo alone has paid more than 20 million dollars to over 7,000 rights holders that way. Our guide to data monetization companies covers the marketplaces, exchanges and brokers that help at each point on that spectrum.
Why a business would choose it
The data already exists and already cost money to produce. Licensing it is revenue against a sunk cost, with no new product, no new headcount to serve customers, and no change to the core business. A non-exclusive license can earn that revenue more than once.
Licensing also gives you control. Data that isn't licensed doesn't stay unused. Someone scrapes it, models trained on whatever was public summarize it, or, in the worst case, a trustee auctions it. Courts have concentrated liability on how data was acquired rather than on training itself, so buyers now pay for rights-cleared training data, and owners who license on their own terms set the scope, the exclusions, and the privacy standard themselves.
And the market has matured enough to be worth the effort. Three years ago a mid-sized company had no realistic buyer for its operational records. Today there are published price signals, standard deal structures, marketplaces that handle packaging and privacy review, and buyers looking for data that shows how real work gets done. Our guide to AI data licensing covers how licensing went from an afterthought to a standard procurement channel.
What it takes to run
A licensing model is an ongoing process, and the companies that do well at it treat it that way. The steps are the same whatever the data:
- Inventory what exists and where it lives.
- Establish what the company owns and has authority to license.
- Define scope: what's in, what's excluded, what safeguards apply.
- Review for privacy and remove names and personal details to fit the dataset.
- Package the data so a buyer can evaluate it.
- Negotiate terms.
Customer personal information, employee personal information, third-party confidential material, and anything under contractual restriction come out before anything ships. The owner's-side version of that process, step by step, is in our guide to licensing company data for AI.
That's real work, which is why most owners below publisher scale use a marketplace instead of building a licensing team in-house. The marketplace handles buyer matching, standard terms, packaging, and privacy review once for every owner on it. You decide what's in scope and get paid as the data licenses.
When it’s the wrong model
Licensing is the wrong call when:
- the data isn't scarce, because generic records that look like everyone else's have little training value
- the rights aren't clean and can't be made clean
- customer contracts or regulation forbid it
- the company can't separate the operating knowledge in its data from the personal information of the people in it
- an exclusive premium would stop the company signing deals it will want in two years (the arithmetic is in our guide to exclusive vs non-exclusive data licenses)
- it distracts from the core business
For most companies this is a second source of revenue alongside the core business. There's no universal price per record, and a company that can't say what makes its data different from a thousand others' should expect a modest outcome or none.
Does it apply to your company?
The quickest way to tell whether your data is worth licensing is to take Troveo's free, five-minute data value assessment (ten questions about your systems, history, and industry). It scores the AI-training value of what your company holds, you don't need a data team to fill it in, and it tells you whether a licensing conversation is worth having before you spend a day on one. Our guide to selling data to AI companies covers what sells and what it pays.
Where Troveo fits
Troveo is the marketplace version of the model, built for owners. We've spent years helping owners of proprietary real-world data license it for AI across video, audio, gaming, and robotics, have paid rights holders more than 20 million dollars, and now run the same process for business data. We identify what's valuable, work with you on rights and scope, package it for evaluation, and match it with AI buyers. You set the terms, you decide what's excluded before anything ships, your company is never named publicly, and you get paid on every sale, not just the first. Start with the data value assessment, or talk to us about whether a licensing model fits what your company holds.