Data licensing is when the owner of a dataset gives another party written permission to use it for a set purpose, period and price, while keeping ownership. The license, not the file transfer, decides what the licensee may do: which data, which uses (for example AI training, retrieval or analytics), whether the right is exclusive, for how long, and how payment works. Reddit, News Corp and Shutterstock have all licensed data to AI companies, and stock exchanges and real estate listing services have licensed data for decades.
- A data license grants permission to use data; it does not transfer ownership. The owner can usually license the same data again unless it grants exclusivity.
- Every data license answers the same questions: what data, which uses, who may use it, where, for how long, exclusive or not, and how payment works.
- Revenue comes in four common shapes: a flat fee, a subscription or feed, a price per record or use, and a revenue share.
- Open licenses such as CC BY and ODbL give data away on standard terms; commercial licenses are negotiated and paid.
- Before licensing, check that you hold the rights, what personal data is inside and what your contracts allow.
What is data licensing?
Data licensing is an agreement in which the owner of data (the licensor) allows another party (the licensee) to use that data under stated conditions, usually in return for payment. In law, a license is permission to do something that would otherwise be prohibited. Applied to data, it means the licensee may copy, analyze, display or train on the data only in the ways the license allows.
The key word is permission. In a typical data license the owner keeps the data and the right to decide who else may use it. The licensee gets a defined right of use, which can be narrow (one team, one purpose, one year) or broad (worldwide, any internal purpose, perpetual). That is why the same dataset can earn money several times, from several licensees, for different uses.
Licensed data is simply data used under such a license. When an AI company says a model was trained on licensed data, it means the company paid for, or otherwise obtained, written permission from the people who hold the rights. For a one-line definition see the glossary entry for data licensing.
Source: Cornell Law School LII: license
How does a data license work?
A data license is a contract, and most of its value sits in a handful of terms. The table lists the terms almost every license sets, with the plain question each one answers.
- The contract itself is covered clause by clause in the data licensing agreement checklist.
- Exclusivity usually moves the price more than any other term. See exclusive vs non-exclusive data licenses.
| Term | The question it answers | Typical choices |
|---|---|---|
| Licensor | Who grants the right? | The company, publisher or person that holds the rights to the data |
| Licensee | Who may use the data, and can they pass it on? | One named company; sometimes its affiliates or contractors; sublicensing only if stated |
| Dataset | Exactly which data is covered? | A fixed snapshot, a defined set of fields and years, or a live feed with updates |
| Permitted use | What may the licensee do with it? | Internal analytics, display to customers, AI training, AI retrieval, AI evaluation, resale |
| Prohibited use | What is ruled out? | Re-identifying people, reselling the raw data, competing products, uses outside the field |
| Term | For how long? | Months, a fixed number of years, or perpetual for a delivered snapshot |
| Territory | Where may it be used? | One country, a region or worldwide |
| Exclusivity | Can the owner license the same data to others? | Non-exclusive (yes), exclusive (no), or exclusive only for one use, field or period |
| Payment | How and when is the licensor paid? | Flat fee, subscription, per record or per use, revenue share, or a mix |
| End of term | What happens when the license ends? | Delete or return the data; whether models already trained on it may stay in use |
What are the main types of data licenses?
Data licenses can be grouped four ways. A single deal sits somewhere on each of them: for example, a commercial, non-exclusive, training-only license paid as a one-time fee.
- Glossary: exclusive data license, non-exclusive data license, retrieval-augmented generation, evaluation dataset, fine-tuning.
- The difference between training and retrieval rights matters most in AI deals. See AI training vs retrieval data licenses and licensing data for AI evaluation.
| Split | Option | What it means |
|---|---|---|
| Who can use it | Commercial license | Negotiated with one licensee, usually paid, with confidentiality and use limits |
| Open license | Standard public terms anyone can accept, such as CC BY or ODbL; usually free | |
| How many licensees | Non-exclusive | The owner may license the same data to others |
| Exclusive | One licensee only, for the licensed use, field, territory or period | |
| What the AI may do with it | Training or fine-tuning | The data changes the model's weights; hard to undo after the license ends |
| Retrieval (RAG) | The model looks the data up at answer time; access can be switched off | |
| Evaluation | The data is used to test models, not to train them; often kept private | |
| How it is delivered | One-time snapshot | A fixed dataset delivered once, usually for a one-time fee |
| Subscription or feed | Ongoing access to new data, paid monthly or yearly |
What about open data licenses like CC BY and ODbL?
Much of the material that ranks for "data license" comes from universities and open-data projects, which use standard public licenses. Under Creative Commons Attribution 4.0 (CC BY), anyone may copy and redistribute the material for any purpose, even commercially, as long as they give appropriate credit and follow the other terms. The Open Database License (ODbL), used by OpenStreetMap, adds a share-alike rule: if you publicly use an adapted version of the database, you must offer that adapted database under the ODbL too (ODbL summary; OpenStreetMap).
Open licenses are about sharing, not revenue. A company that wants to be paid for its data uses a commercial license instead: one licensee, negotiated terms, confidentiality, and limits on use and onward sharing. The rest of this guide is about commercial licensing.
Source: Open Data Commons: ODbL summary
How does data licensing make money? Business models and revenue
Data licensing revenue comes in four common shapes, and large deals often combine two. The examples below are published figures from public filings, company pages and press reports, not quotes for any particular dataset.
- Flat fees are simple and paid early, but cap the upside. Subscriptions and feeds pay for as long as the data stays fresh. Per-record pricing rewards volume. Revenue share pays only when the pooled data sells.
- Most company-records deals published so far are licenses priced as a flat fee for a delivered dataset or as a revenue share. See data licensing pricing models for the trade-offs, and data licensing as recurring revenue for the subscription route.
| Model | How the licensor is paid | Published example |
|---|---|---|
| Flat fee or fixed contract | A set amount for a dataset or a multi-year right, often paid in installments | Reddit reported data licensing arrangements with an aggregate contract value of $203.0 million and terms of two to three years, signed in January 2024 (TechCrunch). Bristol-Myers Squibb will pay Tempus AI $137.5 million in fixed payments for oncology data (8-K text). |
| Subscription or feed | A monthly or yearly fee for ongoing access, often per user or per site | A 2022 Nasdaq filing describes TotalView enterprise licenses at $100,000 a month plus per-subscriber fees, or $500,000 a month for broad distribution, and an $80 monthly fee per professional subscriber for one display product (Federal Register). |
| Per record or per use | A price for each record delivered, query run or unit used | Tempus AI disclosed in October 2026 that Pathos pays $100 per delivered de-identified neurology record, up to $50 million in license fees (8-K text). |
| Revenue share | A percentage of what a platform earns when it licenses pooled data | Shutterstock says contributors earn a 20% average royalty on revenue it receives for data licenses, split by how much of each contributor's content is in the licensed datasets (Shutterstock). |
Source: TechCrunch: Reddit says it has made $203M so far licensing its data (Feb 2024)
Data licensing examples by sector
Data licensing is old. Exchanges, listing services and publishers have licensed data for decades; AI has added a new kind of buyer. Four sectors show the range.
| Sector | What is licensed | How the license is shaped | Source |
|---|---|---|---|
| Financial market data | Prices, quotes and order-book depth from exchanges | Subscription fees per user, with enterprise licenses for firms that distribute widely; fees differ for professional and non-professional users | Nasdaq filing (2022) |
| Real estate data | Property listings held by multiple listing services (MLSs) | Separate license agreements by intended use: public listing display (IDX), virtual office websites (VOW), back-office use and broker-only data; MLSs may charge fees for back-office feeds | MIBOR; CCIAOR |
| News and content for AI | Current and archived articles, posts and images | Multi-year contracts covering training, and in some deals use of content to answer user questions | AP; TechCrunch; Bloomberg Law |
| Company records for AI | Internal emails, chats, documents and operational records, de-identified | Usually a one-time license or sale of a defined snapshot; personal data removed before delivery | KLTV / InvestigateTV |
Market data licensing
Stock exchanges are among the oldest data licensors. They charge per user and per use: one 2022 Nasdaq filing lists TotalView enterprise licenses of $25,000, $100,000 and $500,000 a month depending on how widely a firm distributes the data, with professional and non-professional subscriber fees added to the first two (Federal Register).
The model shows how far a data license can be sliced: by type of user, by display or machine use, by internal or external distribution. The same thinking applies to any dataset that updates continuously, including company records that keep growing.
Source: Federal Register: Nasdaq TotalView enterprise license filing (Jul 2022)
Real estate data licensing
Listing data is licensed by use. The MIBOR Broker Listing Cooperative in Indianapolis offers four types of data licensing agreements "based on the intended use of the data": IDX, VOW, back-office and broker-only (MIBOR). MLS guidelines also allow reasonable licensing provisions and fees for back-office feeds (CCIAOR).
The lesson for any data owner is that one dataset can carry several licenses at different prices, each tied to a defined use.
Source: MIBOR BLC: data licensing agreements (IDX, VOW, back office, broker-only)
AI data licensing: news, content and company records
AI developers license data for training and for answering questions. News Corp struck a multiyear deal with OpenAI covering both training and answers to user questions, with access to fresh and archived material; the Wall Street Journal said it could be worth more than $250 million over five years (AP). Reddit disclosed $203.0 million of data licensing contracts in its IPO filing (TechCrunch), and Shutterstock's AI licensing business produced $104 million of revenue in 2023 (Bloomberg Law).
Ordinary company records are the newer category. In 2026 Google won a bankruptcy auction for Spirit Airlines' internal business data at $10 million; the records at issue include about 100 million emails and 500 million workplace messages, and Google said the data could help improve its products and AI models. As of November 3, 2026 it was still a proposed sale, while a flight attendants union and more than 100 members of Congress pushed for stronger privacy protections for former employees (KLTV).
Most operating companies are not publishers or bankrupt airlines, and their deals are smaller and private. The AI data licensing deals tracker lists the public ones with sources, and how to license company data covers what AI buyers look for in company records.
Source: AP: OpenAI to use News Corp content for training and answers (May 2024)
Data licensing vs selling data vs data sharing
People use "sell data" loosely. In practice most commercial data deals are licenses, and the differences matter for price, control and law.
- Privacy law may treat a license as a sale. California's CCPA defines "sell" to include making a consumer's personal information available to a third party for monetary or other valuable consideration (Civil Code 1798.140). Removing personal information first avoids most of that.
- For how an outright sale works and when it makes sense, see how to sell data.
| Data licensing | Selling data outright | Data sharing | |
|---|---|---|---|
| Ownership | Stays with the owner | Passes to the buyer, often with all rights | Stays with the owner |
| Payment | Fee, subscription, per use or revenue share | Usually one purchase price | Often none, or exchanged for services or other data |
| Can the owner reuse it? | Yes, unless the license is exclusive | Usually not | Yes |
| Control after delivery | Use limits, audit and deletion terms in the contract | Little, unless reserved in the sale terms | Set by a data sharing agreement or open license |
| Typical setting | Market data, listings, AI training data | Bankruptcy and wind-down sales of a company's data | Research partnerships, supply chains, open data |
What to check before licensing your data
A buyer will ask three questions before it pays: do you hold the rights, what personal data is inside, and do your contracts allow it. Answering them yourself first saves weeks.
- Rights: confirm the company created or acquired the data and that no customer, vendor or platform contract reserves it. Know where every source came from (data provenance).
- Personal data: under the GDPR, personal data is any information relating to an identified or identifiable person, which covers most emails, chats and tickets. Plan to remove or mask it before delivery. See de-identifying company data before licensing and data licensing under GDPR and CCPA.
- Contracts: read customer terms, NDAs, software and cloud agreements, and employee policies for limits on use or disclosure. Regulated records (health, tax, legal) have extra rules.
- Buyer checks: expect questions about volume, history, format, rights and security. See data licensing due diligence.
- Terms: decide in advance which uses you will allow (training, retrieval, evaluation), whether you will grant exclusivity and what happens at the end of the term.
Source: GDPR Article 4: definitions
How to start licensing your company's data
Start with a description, not an export. List the systems you use, the years of history each holds, rough volumes and what each record shows about how work gets done. No files need to move at this stage.
The free company record map helps you lay that out one workflow at a time. When you have a rough picture, see what your data is worth from a few questions about the company.
Origin Data Partners is an independent data partner that helps companies get their data bought. It is free to the company that owns the data; a buyer may pay Origin when a deal closes (how Origin is paid). Any introduction, any records access and the license terms remain separate decisions for the owner.
Common questions
What is data licensing in simple terms?
It is renting out permission to use your data. The owner keeps the data and signs a contract that lets another company use it for a set purpose, period and price. The contract says what the user may and may not do with it.
What is the difference between a data license and a data licensing agreement?
The license is the permission itself: the right to use the data in certain ways. The data licensing agreement is the contract that grants it and sets the details, including permitted uses, exclusivity, term, territory, privacy duties, deletion and payment.
What is licensed data?
Licensed data is data used under written permission from whoever holds the rights to it, usually for a fee. AI companies use the term for training data they obtained through license deals, as opposed to data scraped from the public web.
What are examples of data licensing?
Stock exchanges license price and order-book feeds to trading firms for monthly fees per user. Multiple listing services license property listings separately for public display, virtual office websites and back-office use. Reddit, News Corp and Shutterstock license content to AI companies, and in 2026 Google won an auction for Spirit Airlines' internal business data for $10 million.
How do companies make money from data licensing?
Through four common models: a flat fee or fixed contract, a subscription or feed, a price per record or per use, and a revenue share of what a platform earns from pooled data. Published examples include Reddit's $203.0 million of two to three year contracts, $100 per de-identified record in a Tempus AI filing and a 20% average royalty paid by Shutterstock to contributors.
What is AI data licensing?
AI data licensing is licensing data for use in building or running AI systems. The license states whether the data may be used to train or fine-tune a model, to retrieve answers at question time, or only to evaluate models, and what happens to trained models when the license ends.
Is licensing data the same as selling data?
No. In a license the owner keeps the data and grants defined rights, and can usually license it again. In an outright sale ownership passes to the buyer. Privacy laws such as the CCPA can still treat licensing personal information for payment as a sale, so personal data is usually removed first.
What are CC BY and ODbL licenses?
They are open licenses. CC BY 4.0 lets anyone copy, share and adapt material for any purpose, including commercially, if they give credit. The ODbL, used by OpenStreetMap, also requires that publicly used adapted databases be shared under the ODbL. They are built for sharing, not for earning license fees.
Can a small or private company license its data?
Yes, if it holds the rights and the data shows something buyers cannot get elsewhere, such as years of operational records from a specialized field. Volume, history and how cleanly personal data can be removed all affect whether a buyer is interested and at what price.
Do I need a lawyer to license data?
For a commercial license, yes, it is worth having counsel review the agreement, especially the permitted uses, exclusivity, privacy terms, warranties about rights and what happens to trained models when the license ends.
Sources and further reading
Origin Data Partners is independent and not affiliated with any buyer named on this page.
- Cornell Law School LII: license
- Creative Commons: Attribution 4.0 International (CC BY 4.0)
- Open Data Commons: ODbL summary
- OpenStreetMap: copyright and license
- TechCrunch: Reddit says it has made $203M so far licensing its data (Feb 2024)
- SEC EDGAR: Reddit, Inc. Form S-1 (Feb 2024)
- AP: OpenAI to use News Corp content for training and answers (May 2024)
- Bloomberg Law: Shutterstock's AI licensing business generated $104 million (Jun 2024)
- Shutterstock: data licensing and the Contributor Fund
- SEC EDGAR: Tempus AI Form 8-K (Oct 23, 2026)
- Stock Titan: text of the Tempus AI 8-K (Oct 2026)
- Federal Register: Nasdaq TotalView enterprise license filing (Jul 2022)
- MIBOR BLC: data licensing agreements (IDX, VOW, back office, broker-only)
- CCIAOR: MLS data distribution guidelines
- InvestigateTV via KLTV: privacy protections sought in Spirit data sale to Google (Nov 3, 2026)
- GDPR Article 4: definitions
- California Civil Code 1798.140 (CCPA definitions, including "sell")