Data monetization examples: real AI data deals, with dates and figures

01

What counts as a data monetization example?

Every business already holds data someone will pay for

Every company produces records as a side effect of doing its work. A newspaper produces an archive of articles. A card network such as Mastercard, which passes card payments between shoppers' banks and stores, records billions of purchases. A heating and air conditioning contractor keeps a service ticket for every repair, a dispatch log of which technician went to which house, and years of email with its customers.

For most of history, only a few kinds of company could turn those records into money. They needed data that covered a whole market, such as every card purchase in a country, and a team to turn it into reports other companies would buy. AI brought a new kind of buyer. AI labs and data vendors now pay for the everyday working records of a single company, because those records show an AI model how a job gets done.

Five ways companies have turned data into money

Every example on this page uses one of the five methods below.

License

The company keeps its data and grants a buyer the right to use a copy for a set purpose, such as training AI models, for a fee. Most AI data deals are licenses.

Sale

The buyer takes ownership of the data outright. This is common when a company shuts down and its records are sold with its other assets.

Insight product

The company sells reports or dashboards built from its data, never the raw records. Mastercard's retail spending reports work this way.

Access fee

Buyers pay to pull data through an API, a connection that lets one company's software read another's data, priced per call or per month.

Paid tasks

The company's staff do tasks for a data vendor, such as doing their real work while the vendor records each step, or scoring how well an AI model handles that work. The vendor pays the company for each finished task.

Who is buying company data now

The buyers in almost every recent example are AI labs and data vendors. Publishers and big platforms usually sell straight to an AI lab. Ordinary companies usually sell to a data vendor.

AI labs

Companies that build and train AI models, such as OpenAI, Anthropic, Google and Meta.

Data vendors

Companies that buy records from businesses, anonymize them (strip out names and other details that identify people) and turn them into training data, the examples an AI model studies to learn. micro1, the data vendor we typically refer companies to, is one; Mercor, Polyshares, HUMXN and SourceX are others.

The diagram below maps who sells what to whom across every example in this guide.

How to read the examples

Each example gives the date it was announced or reported, the parties, what was licensed and the figure. Most companies don't disclose how much they were paid. Where a number comes from press reporting rather than from the companies, it is marked "reported", and every figure links to its source.

02

AI content licensing deals: publisher and platform examples, 2023–2026

Why publishers and platforms went first

AI models were first trained on text copied from the public internet, such as websites, books, news articles and forum posts. Publishers owned large, well-organized archives of exactly that material, with clear rights to it. Many publishers also discovered that AI labs had already trained models on their articles without paying for them. Signing a license, a contract that lets the AI lab use a copy of the content for agreed purposes while the publisher keeps ownership, gave the publisher income and gave the lab written permission.

Most of these deals pay the publisher for two permissions at once. The AI lab may train its models on the publisher's archive, and the lab's chatbot may show current articles, with links back to the publisher, when users ask about the news.

News publishers

Platforms, forums and stock media

Platforms such as Reddit hold a different kind of archive, made of the conversations of millions of people and the images and videos those people upload. Stock-media libraries, which sell photos and video clips to advertisers and publishers, hold something AI labs need to train models that make images and video. Each picture in a library comes with a written description of what it shows, so a model can learn which words match which pictures.

Books, research and entertainment

Books and academic papers contain long, carefully reasoned arguments, which short web pages rarely do. Book and journal publishers became the next sellers. Entertainment companies followed in 2025, licensing music and characters to AI tools that make songs and video.

How the buyers moved from publishers to ordinary companies

Publishers and platforms made almost all the AI data deals in 2023 and 2024. By 2026, the sellers included ordinary operating companies and companies that had shut down. The timeline below puts the dated examples in order, with ordinary and closed companies in violet.

03

Companies that sell data to AI: ordinary business examples from 2026

What changed: AI now needs to see work being done

AI labs now build AI agents, software that carries out a whole job step by step, such as answering a customer's complaint and booking the repair. An AI model learns a job by studying many records of that job being done from start to finish. Those records exist only inside companies, in their help desks (where staff answer customer questions), team chat, email and CRMs (customer relationship management systems, where sales teams track every customer).

AI labs and data vendors use those records in two ways. First, they turn complete records of a task done well into worked examples that an AI model studies. Second, they build practice copies of a company's software, called reinforcement learning (RL) environments, where an AI agent attempts a task over and over and is scored each time until it does the task reliably. We explain both in why AI companies want your data.

Published examples from ordinary companies

micro1 appears in several rows because it publishes its terms. Polyshares works differently: it buys an exclusive license directly from the company and supplies the data to AI labs itself.

What these companies actually licensed

None of these companies licensed anything they had published. A roofing company's operational record holds the estimates its sales staff wrote, the scheduling of each crew, the supplier orders and the emails with each homeowner from first call to final invoice. That chain of records shows, step by step, how the company won a roofing job and then delivered it. Those are the steps an AI agent has to learn before it can do the same work.

The HVAC example, from a business that installs and repairs heating and air conditioning, works another way. The company's CEO does paid tasks for the data vendor: he checks how well AI models handle the paperwork his firm handles every week and scores their answers. That is the "paid tasks" method. At about $1,000 a task, the CEO was on pace to earn more than $150,000 a quarter.

The numbers behind ordinary-company deals

What a given company is offered depends on three things: how many years of records it has with nothing missing, whether its records link each step of a job to the next, and how hard similar records are to find elsewhere. Our guide to what company data is worth explains each price driver, and the estimate tool gives a range for your company in a minute.

04

Closed companies are selling their data too

Why a shutdown now includes a data sale

When a company closed, its chat history, email and files used to be deleted along with its accounts. Today AI labs and data vendors will pay for those records. A closed company is also often an easier seller, because nobody is still working in the systems and the whole history can be transferred at once.

In a bankruptcy, the money from the sale goes into the bankruptcy estate, the pool of assets a court uses to repay the people and companies the business owes money to (its creditors). In a smaller, orderly wind-down, the money can cover final costs and go back to the founders and investors.

The Spirit Airlines archive

Spirit Airlines, a US budget airline, ended up in bankruptcy court in 2026, and its business records were auctioned on their own. Google won the auction in August 2026 with a $10M bid, ahead of the data vendor Mercor at $7.5M. micro1 then filed a late bid of $12.5M.

The archive covers Spirit's internal email, Microsoft Teams chat and software code. Passenger and loyalty-program records were excluded, and the rest is to be stripped of personal details before it is transferred.

Unions for Spirit's flight attendants objected to the sale, and a privacy ombudsman, an independent reviewer the court appoints to check how a sale handles personal information, reviewed it. At the most recent reports we could find, in October 2026, the bankruptcy court had not yet approved a buyer, so "won the auction" is the accurate description.

The buyers and services that sell closed companies' data

The legal checks matter even more after a shutdown, because the people who could confirm what the records contain may be gone. Our guide to whether it is legal to sell company data lists what a lawyer will check first.

05

How much companies have been paid for data: deal sizes compared

Why the sizes span six orders of magnitude

The published figures are counted in different units. Some are paid per task, some per dataset, some per year and some over a whole multi-year contract. The diagram below places each figure on one scale, where each step up is ten times larger than the one below it, so a $1,000 task and a $250M contract fit on the same chart.

What the comparison shows an ordinary company

A publisher's license covers decades of work from thousands of journalists, so it lands in the tens of millions a year. An ordinary company's dataset covers one business, and it still sells for $100K to $2M or more. A single company archive with decades of history, such as Spirit's, has drawn eight-figure bids.

A company can also earn recurring income on top of a one-time license. micro1's agent-trajectory licenses run $1M–$2M a year, and paid tasks run about $1,000 each. The guide to data licensing explains flat fees, yearly licenses and per-task pay side by side.

06

What happens without a license: settlements that set a price

The Anthropic authors' settlement

In Bartz v. Anthropic, authors sued Anthropic, the AI lab that makes Claude, for training on books downloaded from pirate libraries, websites that share copies of books without the publishers' permission. Anthropic agreed in September 2025 to pay $1.5B, about $3,000 for each of roughly 482,000 books, and the judge gave preliminary approval on September 25, 2025 (Publishers Weekly). The court granted final approval in July 2026 (Authors Alliance).

The settlement is not a data monetization deal; nobody chose to sell. It belongs on this list because it put a public price, about $3,000 a book, on data an AI lab used without permission.

Lawsuits that turned into licenses

Why settlements push buyers toward licensed company data

A license costs less than a lawsuit and gives the buyer proof that it may use the data. That is why AI labs and data vendors now ask sellers to show they own their records and have removed personal details. The same proof protects the seller from later claims, which is why our guide on how company data is anonymized walks through exactly what happens to names, emails and phone numbers before anyone sees a file.

07

Classic data monetization examples from before AI

Selling insight, not records

The classic model sells conclusions drawn from data rather than the data itself. A card network can report how much shoppers spent on electronics last month because it processes those payments. A retailer can tell a cereal brand which households switched to a rival, because it sees every basket at checkout.

Twitter's and Reddit's 2023 API prices led directly to the first platform AI deals. Both companies began charging outside software for bulk access to their posts in 2023. A year later, Google and OpenAI received Reddit posts for AI training through that same paid API.

The cautionary example: phone carriers' location data

Data monetization without consent can cost more than it earns. In April 2024 the FCC, the US communications regulator, fined AT&T, Verizon, T-Mobile and Sprint nearly $200M in total for selling customers' real-time location to data aggregators, companies that buy data from many sources and resell it, which passed it on without proper consent (FCC). The same lesson applies to every example on this page: the seller removes personal details before the data changes hands.

Why the classic models rarely work for an ordinary business

Every classic example needs data that covers a whole market, plus a team to build and sell a product from it. A 50-person contractor has neither. AI licensing is different because the buyer pays for the record of one company's work, and the data vendor, not the company, does the work of cleaning and packaging the records. Our guide to data monetization models and strategy compares all six models side by side.

08

What successful data monetization examples have in common

Four traits behind the deals that paid

• The rights were clearEvery publisher on the list owned its archive, and every company licensed only its own records. Buyers pay for data they can prove they may use.
• Personal details came out firstSpirit's passenger records were excluded, Polyshares anonymizes before delivery, and the FCC's fines on phone carriers show what happens when they don't.
• The records were hard to get elsewherePaywalled news, private forums and internal company records can't be copied automatically from the public web, so buyers have to pay for them.
• Connected beats bigA roofing company's chain of estimate, schedule, order and invoice is worth more than a larger pile of unrelated files, because it shows each step of one job in order, which is what an AI agent has to learn.

Timing matters

Prices are high because AI labs and data vendors need more company records than are on offer. MIT Technology Review expects data prices to stabilize over the coming year as more companies sell (MIT Technology Review, above). micro1 has committed to spend $1B on company data in the 12 months from October 2026.

What these examples mean for your company