What is your company's data worth to AI developers?

"What is our data worth?" is the first question every owner asks, and the honest answer is that it depends. There is no public price list. But enough deals have now been reported to show the range, and the factors that move a company up or down it.

This guide sets out what companies have actually been paid in 2026, what drives the price, what lowers it, and how buyers arrive at a number. If you are new to the subject, start with our UK guide to selling company data to AI developers.

What companies have been paid

These are the prices reported publicly so far. All figures are in US dollars, and most involve US companies. UK deals are newer and few have been made public.

Competition can push prices up. After Google's $10m bid won the Spirit auction, micro1 made a late offer, reported at $12.5m.

What drives the price

Years of history

More years means more examples of how work is done, how decisions change and how problems get solved. Brendan Mahony, the chief executive of Sunset, a US startup wind-down firm with a growing data resale business, told MIT Technology Review that he values assets by headcount, years in operation, data volume and how the team used its tools.

Team size

More people means more conversations, more decisions and more variety. micro1's main programme looks for companies with 30 or more employees.

The range of systems, and how they were used

Email on its own is useful. Email alongside team chat, shared documents, project tickets and a CRM is worth more, because together they show a piece of work from request to result. How the team used each tool matters too: an active, well-used Jira is worth far more than an empty one.

Depth and variety

Buyers want real disagreement, judgement calls and messy problem-solving, not just routine messages. Nick Heiner of Surge AI said one company's Slack history can be worth more than another's Slack, email and calendars combined if its conversations are varied and deep.

Scarcity

Rare records are worth most. According to SimpleClosure's data, reported by Upstarts Media, code in rarer programming languages such as Ruby or C++ "can fetch top dollar". Context from tools such as Jira, Linear or Figma, alongside related emails, customer records and support logs, adds value. Datasets with rare origins are among the most prized.

For most UK companies, scarcity means specialist work. An engineering consultancy's design reviews, an accountancy firm's working papers or a software team's code reviews show expertise that is hard to find anywhere else.

Exclusivity and ongoing supply

An exclusive licence usually earns more than a non-exclusive one. Ongoing supply can earn more again: micro1's $500,000+ tier covers large datasets or ongoing participation across several teams and business functions.

What lowers the price

• Heavy personal data. Records that are mostly about individuals need more removal work, and some categories have to come out completely.
• Client-owned or confidential material. If client contracts restrict how material can be used, it has to be left out.
• Unclear ownership. Buyers discount records when they cannot be sure the company has the right to license them.
• Missing history. Deleted or archived accounts leave gaps. Act before subscriptions lapse.
• Small teams and short histories. Very small or very young companies generate fewer usable examples.
• Records not in English. Many buyers currently want English-language material.

Our UK guide explains how to handle personal data, client material and ownership so they cost you as little as possible.

How buyers arrive at a number

Volume is only one factor in the price. According to MIT Technology Review, valuations rest on "research theses": estimates of how many useful training tasks a dataset can produce. That is why two buyers can quote very different prices for the same records, and why one offer is not the market price.

The same reporting expects prices to stabilise over the coming year, unless an AI bubble bursts first. For now, the best protection is competition: put your records in front of several buyers, not one.

Deal terms that change the value

The headline price is only part of an offer. These terms can make a lower offer worth more than a higher one:

• Exclusivity. Will you be free to license the same records to other buyers? Exclusivity should cost the buyer more.
• Scope. Which systems and which years are included? A buyer who wants everything should pay for everything.
• Permitted use. Training and testing models is standard. Rights to resell or pass the records on are not, and should be limited or priced.
• Duration. How long can the buyer keep and use the copy, and what happens at the end?
• Payment schedule. A price paid on delivery is worth more than the same price spread over milestones the buyer controls.
• Deletion. The buyer should delete the copy on request, or when the licence ends, and confirm it in writing.

How to protect your value before you sell

1. Keep accounts active. Do not cancel subscriptions that hold records you might license.
2. Do not tidy up. Bulk-deleting old mailboxes, channels or projects removes the history buyers pay for.
3. Write down what you have. Which systems hold which years, and roughly how many people used each one.
4. Spot exclusions early. HR, health and client personal data will come out. Knowing how much is affected helps set expectations.
5. Check client contracts. Find any terms that restrict how client material can be used.
6. Get more than one offer. Competition is the best protection while prices are still settling.

Where does your company sit?

Without a sample, nobody can give you a firm number. But the factors above sort companies into a stronger or weaker starting position.

How to get a valuation

A first valuation should not need any files. These facts are enough to judge whether your company is a fit: your sector, headcount, years of records, the systems you use, the company's status and whether the accounts are still accessible.

Get a free valuation. It takes about two minutes, and you get an instant fit check. If you are an office-holder, our guide for insolvency practitioners explains how records can be realised for creditors.