How Much Is Your Company's Data Worth to AI Labs? (What Buyers Pay in 2026)

Most AI data deals are confidential, so the visible numbers are the edge of the market rather than its average. They're still the best reference points you have, and they cluster by what was sold. Our guide to data monetization companies puts the companies behind these deals side by side, with the numbers.

The closure-market figures come from reporting on platforms that broker assets from companies winding down, and our guide to selling source code to AI labs covers them in detail. The Spirit figure is a court-record bid, and our breakdown of the Spirit Airlines data deal explains why an airline's internal records drew a bid at all. We keep the publisher and platform deals, with sources, current on our AI training data statistics page. If your company is closing and those first two rows are the ones that apply, our guide to selling your code and data to AI labs when shutting down covers what to export first and what buyers check.

The range is wide because the things being priced are different products. A repository is a finished, bounded asset. A company archive is years of connected activity across many systems. A publisher's catalog is a brand-name corpus that keeps growing. Each row shows what a particular shape of data earned, and your company's data has a shape too.

The Spirit number is the one most often misread. It reflected an airline with tens of thousands of employees and decades of operations, sold in a liquidation where the buyer took the whole archive in one purchase. A 60-person software company isn't going to see that figure. What Spirit established is narrower and more useful. Operating history has a market value you can see, and you can engage with that market on your own terms while you're still running, before a trustee does it for you. Our guide to how AI data licensing deals work covers the mechanics of that market and why the structure of a deal moves the number as much as the data does. And if you want to see AI licensing next to the other ways companies earn from data, our guide to data monetization examples compares all four.

Buyers don't price data by the gigabyte, and they rarely start from a rate card. An AI lab evaluating a dataset asks one question. What does this let us train or test that we can't already do with what we have? The answer sets the ceiling. Data that fills a specific gap, such as real multi-step business workflows with outcomes, real support conversations with resolutions, or real engineering history with the reviews and the bugs, is worth what closing that gap is worth to the buyer. Data that duplicates what's already on the public web is worth close to nothing, however much of it there is.

Three things follow from that:

• A buyer will usually want to evaluate a sample before committing, so packaging and documentation are part of the value.
• The same dataset can be worth very different amounts to different buyers, which is why marketplaces that reach many buyers tend to find higher prices than a single direct negotiation.
• The pricing structure follows the data. Buyers pay flat for a finished archive, license a living source on a recurring basis, and usually price marketplace data per unit or as a revenue share.

Our guide to AI training data cost covers the buyer's side, including what the disclosed deals reveal about pricing logic.

Inside the range, six factors do most of the work. They apply whether the company has 50 employees or 5,000.

Uniqueness. Data that exists nowhere else is the whole premise. A specialized product, an unusual industry, a process most companies don’t run, or a customer base with distinctive behavior all push value up. A company whose records look like every other company's records pushes it down.

History. Five years beats five months. Buyers pay for patterns, and patterns need time (seasonal cycles, product changes, how a team's practices evolved, how problems recurred and were eventually solved). A long history is also harder to substitute, because nobody can generate it quickly.

Connected context. A support ticket on its own is a data point. The same ticket linked to the Slack thread where it was diagnosed, the Jira issue that tracked it, the pull request that fixed it, and the follow-up to the customer is a complete example of real work. Records that connect across systems are worth more than the sum of the exports, and our business data page says it plainly. The more systems licensed, and the more users and years behind them, the more valuable the data becomes. Our guide to enterprise operational data explains why buyers think in connected records rather than single sources. The same applies to a company whose history lives mostly in files: a folder of SOPs and proposals on its own sits at the lower end, and the value climbs when the email, project and CRM records around those files come with it. Here's what your company's docs are worth to AI labs.

Visible outcomes. A record of work with the result attached, resolved or escalated, won or lost, merged or rejected, is training material. The same record without the result is much less useful, because a model can't learn what good looks like if it never sees what happened next. Our glossary entry on workflow trajectories covers why the sequence with its outcome is the unit buyers care about.

Rights cleanliness. Buyers pay for licensed data because it comes with a documented, lawful record of where it came from. Data you clearly own, with personal information replaced by pseudonyms and third-party restrictions respected, is worth more than data with an unclear chain of custody, and data you can't prove you have the right to license is worth nothing at all to a serious buyer. Our guide to AI data provenance covers what that documentation looks like.

Buyer demand. The last factor is outside your control. Demand moves by category and by year. Agent training pushed up the value of workflow and code data, and voice products did the same for real conversation. Knowing which categories are being bought right now is part of any valuation, and our guide to selling data to AI companies tracks what buyers are paying for.

Founders tend to start with size: terabytes of logs, millions of rows, years of email. Volume matters far less than assumed. A thousand resolved support tickets with full context and outcomes are worth more than a million log lines, and a well-documented repository with five years of commits, reviews, and bug fixes is worth more than ten abandoned ones. Buyers who are training models to do real work want depth and connection, and they can already get bulk from the public web.

This cuts the other way too. A company that assumes it's too small to matter is usually wrong. Value comes from uniqueness, history, and context, and a 40-person firm with a decade of specialized operational records can hold more training value than a 4,000-person company with generic data. Our guide to how tech startups and SaaS companies license their data to AI covers what a typical startup stack holds, system by system.

Everything above describes what a dataset is worth in one transaction. Most companies don't have to settle for one transaction, and that changes the total more than any single factor.

Selling data is a one-time payment for something that could have earned more than once. Licensing keeps ownership with you and grants a right to use a copy (with names and personal details replaced by pseudonyms), which means you can license the same dataset to more than one buyer and turn it into more than one product (a historical export, an enriched dataset with added structure, a continuing feed, and evaluation data built on top of it). Our business data page describes the model as one dataset, many products, with you sharing in every sale, and gives four times the average sales per dataset as an illustrative example of how the total compounds. Exclusivity adds a further lever. An exclusive license commands a premium because the buyer gets something competitors can't also have, and you can sell a non-exclusive one repeatedly. Our guide to exclusive vs non-exclusive data licenses works through which is worth more in a given case, and our guide to data licensing as a business model covers the economics of running this as recurring revenue.

So "what is my data worth" has two answers. What a first deal pays, and what the data can earn over time. The first is the floor. The second is where a licensing program is judged.

The other direction matters too, because it saves you from valuing the wrong thing. The number goes down when:

• the data has no outcomes attached
• the data is generic and looks like everyone else's
• you can't cleanly show you own it, or it sits under a customer contract restricting its use (that prices at zero until it's resolved)

And the data a company is often proudest of, the product usage data and customer records it collects, is frequently the least licensable, because customer contracts and privacy law restrict it even after names and personal details are replaced with pseudonyms. A company's most valuable licensable data is usually the record of its own work (engineering, support, sales process, operations, and decision records). That reframing alone changes most valuations, and our owner's guide to licensing company data for AI walks through the inventory and rights review that establish what's in scope.

A serious valuation starts with an inventory:

1. List the systems, the years of history in each, the number of users, and the categories of records with outcomes.
2. Establish what the company owns and what customer contracts or privacy obligations exclude.
3. Get the data in front of buyers, which for almost every company means going through a marketplace.

The quickest first step is to take Troveo's free, five-minute data value assessment (ten questions about your company's systems, history, and industry). It gives you an estimate of what the data could license for, and a scoping call turns it into a real number. There are no fees, no setup costs, and no deductions from payouts, so finding out costs nothing but the time.

Troveo helps companies understand what proprietary data they hold, protect what matters, and selectively license what's valuable. We've paid more than 20 million dollars to rights holders across video, audio, gaming, robotics, and business data, work with more than 40 active buyers, and run the same process for operating companies (inventory, rights review, scoping, pseudonyms for personal details to a documented standard, packaging, and buyer matching). You decide what's in and what's out at every step, your company is never named publicly, and you get paid on every sale. Start with the data value assessment to see what your company holds, or talk to us about what it might be worth.