A company can sell data to AI labs in three ways: list it on a data marketplace, hire a broker to prepare it and find buyers, or negotiate a license directly with a lab. Marketplaces suit data that is already packaged and needs distribution, brokers suit operational records that need cleaning, rights review and an introduction, and direct licensing suits large data owners with their own legal and data teams.
The three routes at a glance
| Data marketplace | Broker | Direct license | |
|---|---|---|---|
| Who prepares the data | You | The broker, with your review | You |
| Who finds the buyer | Buyers browse listings | The broker | You |
| Rights and privacy review | You | Shared, led by the broker | You and your lawyers |
| Typical cost | Listing or transaction fees | Commission on each license | Your team's time and legal fees |
| Best fit | Packaged, repeatable datasets | Operational records from mid-size companies | Very large or famous datasets |
Data marketplaces
A data marketplace is a catalog where sellers list datasets and buyers subscribe or request access. Examples include Datarade, Nomad Data, AWS Data Exchange and Snowflake Marketplace. Most marketplaces handle discovery and billing and leave the rest to the seller: you prepare the dataset, write the documentation, set the license and answer buyer questions.
Fees are usually published. AWS, for example, charges a 3% listing fee on AWS Data Exchange public offers. Other platforms charge listing fees, a share of each sale or both.
Marketplaces work well for data that is already a product: a feed of prices, a set of company records, a weather series. They work less well for operational records like a firm's ledgers or a lab's notebooks. Those need cleaning, de-identification, a rights review and a sample before any lab will look, and a listing does none of that.
Brokers
A broker takes your exports, prepares them and sells licenses to buyers it already talks to. Brokers in this market range from companies focused on licensed media catalogs to companies that handle existing business records. Some expert-data firms such as Scale, Surge and Mercor produce new data with paid experts, which is a different business.
A good broker does the work that a marketplace leaves to you:
- Reads your customer contracts to check you can license the data
- Removes personal information and secrets, then tests whether records can be re-identified
- Packages records into the formats labs buy, including episodes built from audit trails
- Shows a redacted sample to buyers under NDA and negotiates terms
- Collects payment and passes on your share
Brokers usually charge a commission on each license. We work this way: we take a commission on each license, and if nothing sells, you pay nothing. Your share is set in your seller agreement.
The risk with any broker is opacity. You can't see the buyer list or the negotiation directly, so you depend on the broker's terms. Our guide to red flags when choosing a data broker lists what to check before you sign.
Direct licensing
The large public AI data deals were direct. Reddit disclosed $203 million in data licensing contracts signed in January 2024, and News Corp's agreement with OpenAI was reported at more than $250 million over five years. Those sellers had famous datasets, in-house lawyers and the leverage to get a meeting with a lab's partnerships team.
Most companies don't have that position. A 60-person accounting firm or a contract chemistry lab will struggle to reach the right person at a frontier lab, and once there, it faces the lab's procurement, security and data review. Direct deals also put every task on your team: rights review, scrubbing, sample preparation, negotiation and delivery.
Direct licensing makes sense when your dataset is large enough that a lab will seek you out, and when you have legal and data staff who can run the process.
How the routes compare on the things that matter
Rights and privacy
On a marketplace or in a direct deal, you carry the full legal review. A broker should share it, starting with your contracts, because data you hold for customers often can't be sold without their written permission. See can I sell data I hold for my customers.
Control over buyers
Marketplaces sell to anyone who accepts the listing terms, unless you restrict access. A broker should let you approve every buyer. We do: no license is signed until you approve the buyer, and you can rule out any company.
Recurring income
Marketplaces handle subscriptions well for packaged feeds. For operational records, recurring income needs a pipeline that pulls new records, cleans them and delivers them on a schedule. Our SDK does this monthly on your machine, and each license with a refresh term pays for every accepted batch. The recurring data revenue page shows how that adds up.
Price discovery
Marketplaces show list prices. Broker and direct deals are private, and public price points are scarce outside the big media deals. Our guide on what AI labs pay for data in 2026 collects the sourced numbers.
Which route to pick
Pick a marketplace if your data is already a clean, documented product with a clear license and you want broad distribution. Pick direct licensing if a lab has approached you or your dataset is large and well known. Pick a broker if your data sits in business systems, needs cleaning and a rights check, and you'd rather not build a data sales team.
See what your data could earn
The valuation calculator gives a range for your systems and history in a few clicks. If you want the broker route, the seller hub explains how we work from first call to first payout.
Frequently asked questions
Is a data marketplace the same as a data broker?
No. A marketplace lists datasets and lets buyers find them, while the seller prepares the data and handles terms. A broker prepares the data, checks rights, finds buyers and negotiates on the seller's behalf, usually for a commission.
What do data marketplaces charge?
It varies by platform. AWS charges a 3% listing fee on AWS Data Exchange public offers, according to its published listing fees. Other marketplaces charge listing fees, a percentage of sales or both.
Can I use a broker and a marketplace at the same time?
Often yes, as long as your license terms don't conflict. If a buyer pays for exclusivity on a field or period, you can't sell the same data elsewhere during that time, so keep track of every license you sign.
Do AI labs buy from marketplaces?
Labs buy from many sources, including marketplaces, brokers and direct deals. For de-identified operational records, they usually want a rights statement, a sample and documentation before buying, which a listing alone rarely provides.