How much is company data worth to AI companies?

By Cody Houser, Founder · Last updated October 24, 2026

The short answer: buyers’ published figures put most business data deals between about $50,000 and $2 million. Small startups that closed have sold archives for $10,000 to $100,000, while established companies with years of complete records sit in the six or seven figures. The largest public deal for a single company’s internal records is Google’s $10 million winning bid for Spirit Airlines’ data.

What buyers publish

A few AI data companies publish their price ranges. These are their own figures, not offers, and they change:

Source	Published range	What it covers
micro1	$100K–$500K; $500K–$2M for larger or ongoing deals; $2M+ for highly unique data	Operational data from companies; payments typically $150K to over $2M
Labelbox Prism	From about $50K to $1M+	Smaller targeted datasets up to large enterprise partnerships
Nyne	Commonly $100K+	Historical business records such as accounting and CRM
Startup wind-down sales (reported by Forbes)	$10K–$100K per company	Nearly 100 closed startups’ email, Slack, and code in one year
Spirit Airlines (bankruptcy auction)	$10M (runner-up bid $7.5M)	Years of an airline’s internal communications and records

Demand is rising: micro1 alone committed $1 billion in October 2026 to buying and licensing company operational data over 12 months.

Rough bands from the published figures

A closed startup’s archive: $10,000 to $100,000, based on the wind-down deals Forbes reported
A smaller, targeted dataset: from around $50,000
A qualified partnership with an established company: $100,000 to $500,000
Large-scale, specialized, or ongoing data: $500,000 to $2 million or more
Exceptional archives: very large records, like Spirit’s, can go higher

These bands summarize what buyers and reporters have published. They aren’t a quote, and your company’s value depends on an actual evaluation. For how the sale itself works, see how to sell data to AI companies.

Every publicly reported AI data licensing deal with a price

Most AI data deals never get a published price. These are the ones where a company disclosed a figure or a major outlet reported one:

Deal	Reported value	What was licensed
Reddit – Google (Feb. 2024)	About $60M per year, per one Reuters source	Reddit content for training Google’s AI models
News Corp – OpenAI (May 2024)	More than $250M over five years, per the Wall Street Journal, including credits for OpenAI tech	Current and archived content from the Journal, Barron’s, MarketWatch, the New York Post, and more
Dotdash Meredith – OpenAI (reported Nov. 2024)	At least $16M per year fixed, plus a variable component	Content from brands such as Investopedia and Better Homes & Gardens
Shutterstock (2023 revenue)	$104M from AI licensing in one year	Images and video licensed to AI developers
Wiley (2024)	$44M expected from AI partnerships ($23M already earned)	Academic and professional publishing content
Taylor & Francis – Microsoft (2024)	More than $10M in the first year; runs until 2027	Academic publications
Spirit Airlines – Google (2026)	$10M winning bid, pending bankruptcy court approval	Part of a closed airline’s business records

Why platform deals don’t price your company’s records

It’s tempting to divide a $250 million deal by something and get a number for your business. That doesn’t work, for four reasons:

Different asset. Reddit, News Corp, Shutterstock, and Wiley licensed huge public or published libraries. A single company licenses the internal record of how its own work gets done: email, files, chat, CRM, and accounting.
Different rights. The News Corp deal lets OpenAI display content in answers to user questions, not only train on it. Display rights are priced differently from a training license.
Ongoing supply. Platform deals are paid per year because new content keeps arriving. A historical archive is often delivered once.
Non-cash terms. The reported News Corp value includes credits for using OpenAI technology, and Dotdash Meredith’s fixed fee is only a floor. Headline numbers aren’t all cash.

The closest comparison for an operating business is the Spirit sale, the one public price for a single company’s internal records. Read the Spirit Airlines data sale breakdown for what was sold and how.

How you get paid: common payment models

One-time fee. You deliver a defined historical dataset and are paid once. micro1 says it pays in full within 30 days of accepting the data.
Payment on milestones. Part of the money is tied to steps such as signing, delivery, and the buyer’s acceptance review. Check what “acceptance” means, because it decides when you get paid and whether you can be paid less.
Upfront fee plus revenue share. Labelbox Prism says sellers receive an upfront licensing fee at signing plus a revenue share on downstream use.
Ongoing or recurring data. If you keep supplying new data, you can be paid each year. micro1 lists recurring purchases of AI agent work records at $1 million to $2 million a year.
Exclusive or non-exclusive. An exclusive license stops you from licensing the same data to anyone else, so it should pay more. Labelbox Prism says its licenses are exclusive by default. A non-exclusive license lets the same records earn more than once.

Nyne’s advice on comparing offers is worth following: separate contracted cash from advertised ranges, estimates, and possible future revenue. The data licensing agreement checklist covers the clauses that set these terms.

What moves the price

Buyers describe the same handful of factors. Here is how each one pushes the number:

Factor	Pushes the price up	Pushes it down
Years of history	Several years of records across multiple systems	Short history, gaps, or purged mailboxes
Completeness	Email, files, chat, CRM, and accounting that tie together	One system in isolation
Kind of work	Expert decisions, workflows, reviews, and edge cases	Routine or templated records
Uniqueness	Work few other companies could supply	Data a buyer can get elsewhere
Quality and consistency	Records kept the same way for years	Messy exports needing heavy cleanup
Exclusivity	Exclusive license	Non-exclusive license (but you can license again)
Rights	Clear authority to license every record included	Customer contracts or privacy promises that block use

micro1 names size, workflow complexity, quality, domain expertise, and uniqueness. Labelbox Prism names data type, scale, exclusivity, and buyer demand, and says value is driven “less by volume and more by how authentically it captures complex, real-world work.” Nyne adds that value depends on more than employee count and that more records don’t automatically mean a higher offer. Run through the data value checklist to score your company against these factors.

Why most offers are private

Public prices are rare, and that’s by design:

Terms are confidential. News Corp and OpenAI didn’t disclose financial terms; the figure came from people close to the company. Reddit and Google declined to comment on theirs.
Prices leak through filings. Dotdash Meredith’s figure surfaced on its parent company’s earnings call, not in an announcement.
Every dataset is priced on its own. micro1 says each partnership is evaluated individually.
Court sales are the exception. Spirit’s price is public because bankruptcy sales run through a court. The National CIO Review called the bids “a rare public example” of business data getting a standalone valuation.

So if someone quotes you a firm number before seeing anything about your records, treat it as marketing.

Taxes: license income or sale proceeds?

How a payment is taxed can depend on whether you sold an asset or licensed it and kept rights. The IRS explains that when you sell a capital asset, the gain is the amount you realize minus your adjusted basis. In a related context, IRS Publication 544 says that if you keep a significant power, right, or continuing interest in a transferred franchise, trademark, or trade name, the payment is ordinary royalty income rather than sale proceeds.

Whether either rule applies to a data license is a question for your tax adviser. Bring the draft agreement, since its terms decide the answer.

Questions people ask

How much is my company’s data worth to AI companies?

There is no list price. Buyers’ own published ranges start around $50,000 for smaller targeted datasets and run from $100,000 to $2 million or more for established companies with deep operational records. Closed startups have sold archives for $10,000 to $100,000. Your figure depends on an actual evaluation of your records and rights.

How much do AI companies pay for data?

Publicly reported platform deals include about $60 million a year from Google to Reddit, more than $250 million over five years from OpenAI to News Corp, and at least $16 million a year from OpenAI to Dotdash Meredith. Those license huge content libraries, so they don’t set a price for one company’s internal records.

What is my data worth to AI if I’m an individual?

This guide covers company records, not personal data. Businesses license their own operational records; an employee can’t sell an employer’s data, and buyers check who holds the rights before anything moves.

Do I get paid once or over time?

Both exist. micro1 says it pays in full within 30 days of accepting the data. Labelbox Prism says it pays an upfront fee at signing plus a revenue share on downstream use. Ongoing data feeds can be paid annually.

Is an online data value estimate an offer?

No. Nyne, which runs an estimator, says an estimate is an initial indication, not a written commitment to pay. A real offer names the dataset, permitted uses, acceptance criteria, and payment conditions.

Is data licensing income taxed like a sale?

Not necessarily. Whether a payment is treated as license income or proceeds from selling an asset depends on what you transfer and what rights you keep. Ask a tax adviser before you sign.

Sources

This guide is general information, not legal, tax, or financial advice. Review any proposed agreement with your own advisers.

Related guides

Check whether your company is a fit. Start with a description of your records; no files needed.