# AI data licensing explained: license, sale or revenue share

URL: https://sellmydatatoai.net/guides/ai-data-licensing-explained
Reviewed: 2026-11-08

AI data licensing means granting an AI company the right to use a copy of your data for agreed purposes, such as training or evaluation, while you keep ownership. Most deals are one-off license fees, exclusive or not. Some pay a revenue share each time data is resold. Outright sales of the data itself are rare outside bankruptcy and closures.

## Key takeaways

- A license grants use rights. You keep ownership unless the contract assigns it.
- The main structures are a one-off fee, an upfront fee plus revenue share, a pure revenue share and milestone payments.
- Polyshares' four published deals were all perpetual exclusive licenses. Endless Data licenses non-exclusively to several labs.
- The rights that matter most are permitted uses, rights in derived assets such as fine-tuned models, exclusivity, term and deletion.
- Once data trains a model it cannot practically be removed, so the end of a license rarely undoes the training.

When companies say they sold their data to an AI company, they usually mean they licensed it. The difference decides what you keep, what the buyer can do and whether you can make money from the same records again. This guide explains each structure with examples from buyers' published terms.

## What is AI data licensing?

AI data licensing is a contract in which you give an AI company permission to use a copy of your data for defined purposes, in return for payment. The purposes are typically training, fine-tuning or evaluating models. You keep ownership of the data and anything the license does not grant.

A [data license](https://sellmydatatoai.net/glossary/data-license) is the same legal tool used for software, music and photographs. What makes AI licenses different is that the buyer's use creates new assets, such as trained models, and the data cannot be taken back out of them afterward.

The license is also where the price is really set. Two offers with the same headline figure can be worth very different amounts once you compare the records covered, the permitted uses, the exclusivity and the term. A lower fee for a narrow, time-limited, non-exclusive license can beat a higher fee for a perpetual exclusive one.

## Is selling data to AI the same as licensing it?

Selling data to AI is usually not a sale in the legal sense. Most company deals are licenses: the buyer gets rights to use a copy, and you keep the original and the ownership.

Outright sales do happen, mainly when a company is closing. The Spirit Airlines data is being sold as an asset in bankruptcy, with Google's $10 million bid subject to court approval, according to MIT Technology Review and The Next Web. Even there, the deal carried conditions: The Next Web reported that the buyer must publicly commit not to re-identify the data and may pass it on only to recipients who make the same commitment. Our guide to [selling data from a closed company](https://sellmydatatoai.net/guides/selling-data-from-a-closed-company) covers those cases.

## How are AI data deals structured?

AI data deals are structured in five main ways, and each published buyer tends to favor one. The table shows the structures, how you are paid and a buyer that uses each one, taken from published terms compiled by Don't Do It Yourself in October 2026 unless stated.

| Structure | How you are paid | Example | Main trade-off |
| --- | --- | --- | --- |
| One-off non-exclusive license | Fixed sum, paid after transfer | Endless Data, which licenses to several labs | You can license the same data again, but each fee is lower |
| Perpetual exclusive license | Fixed sum after approval, net 30 to 60 days | Polyshares, exclusive by default | Higher fee, but you cannot license the data to anyone else, ever |
| Upfront fee plus revenue share | Fee at signing, plus a share of later use | Labelbox Prism | Ongoing income depends on the buyer's sales |
| Revenue share through an agent | A share each time data is licensed | Troveo, which says its datasets sell an average of four times | No upfront certainty, but repeated income |
| Milestone payments | Staged payments set in each agreement | Handshake AI | Payment tied to delivery and acceptance stages |

micro1, checked November 2026, pays in full within 30 days of acceptance after a review period of up to 30 days. Its contracting stage includes limits on reselling data before personal information is anonymized, which is a sensible clause to look for from any buyer.

## What rights does an AI data license grant?

An AI data license grants only the rights written into it, so the scope clauses decide the deal. Nixon Peabody's October 2026 alert lists the questions a license must answer:

- Which models, systems and affiliates may use the data
- Whether use covers training only, or also testing, product development and commercial deployment
- Whether the data may be combined with other datasets or shared with subprocessors or research partners
- Who owns or may use derived artifacts, such as cleaned datasets, annotations, embeddings, evaluation sets, fine-tuned models, weights and outputs
- Whether the buyer must stop outputs reproducing confidential or personal information
- Retention, deletion, security, audit and incident response obligations

Derived assets are the point sellers most often miss. A license that allows training also lets the buyer keep the trained model, unless it says otherwise. If you care about a competitor benefiting, restrict the field of use rather than relying on deletion.

## What happens when an AI data license ends?

When an AI data license ends, the buyer usually has to stop using and delete its copies of your raw data. Models already trained on that data normally continue to exist.

Frankfurt Kurnit Klein and Selz states that once data is used to train a model, it cannot practically be pulled back out. Nixon Peabody adds that deleting source files may not remove the data's influence from a trained model, and recommends saying in the contract whether deletion, retraining or model retirement applies if rights expire or prohibited material surfaces. Phyllo lists withdrawal terms, including what happens to models already trained, as one of six things a training data license should cover.

> **Price the license as if it were permanent** Because training cannot be undone, a five-year license and a perpetual license often have similar practical effects on models trained in year one. Negotiate the price on that basis.

## How long do AI data licenses last?

AI data licenses last anywhere from a single training run to forever. Polyshares' published deals were all perpetual, which means no end date. Troveo describes its exclusivity as time-limited. Annual licenses are common for packaged datasets, and Phyllo reports professional annual licenses at $25,000 to $50,000 and enterprise annual licenses at $50,000 to $500,000 as market ranges.

A [perpetual license](https://sellmydatatoai.net/glossary/perpetual-license) is not automatically bad. It suits archives you will never add to, such as records from a closed business line. For data you keep creating, a fixed term with an option to renew lets you reprice later.

## What is the difference between a research and a commercial license?

The difference between a research and a commercial license is whether the buyer may use your data to build products it sells. Phyllo reports research licenses at $5,000 to $15,000 and commercial licenses at $25,000 to $500,000 or more, and notes that research licenses usually do not permit production training.

Academic and lab research teams sometimes ask for a research license first. If you accept one, define research narrowly, ban use in commercial models and require deletion at the end of the project. Otherwise a cheap research license can become a commercial one in practice.

## What should an AI training data license cover?

An AI training data license should cover the rights you are granting and the evidence that you can grant them. Phyllo lists six things a training data license should cover, which work as a checklist for sellers too:

1. Rights documentation for each asset
2. Explicit permission for AI training
3. A documented chain of acquisition
4. Clearance for each layer of content, since video can contain music, images, performances and trademarks
5. Evidence that rights holders are paid where they need to be
6. Withdrawal terms, including what happens to models already trained

Buyers will expect you to back the first three with documents. A data manifest, the contracts that cover each dataset and a note of your exclusions usually satisfy them.

## Who carries the risk under an AI data license?

The risk under an AI data license sits wherever the warranties and indemnities put it. Nixon Peabody lists the warranties a buyer will ask for: authority, privacy notices, contractual restrictions, IP rights, regulatory compliance, litigation holds, security and completeness of disclosed exclusions.

As the seller, you want those warranties limited by your knowledge and by the exclusions you disclosed, with a cap on liability. Nixon Peabody also mentions holdbacks, where part of the price is kept back for a period, and insurance. Agree these before you share the full data, not after.

## When does a revenue share make sense?

A revenue share makes sense when your data can be licensed to many buyers and you can wait for the income. A [revenue share](https://sellmydatatoai.net/glossary/revenue-share) pays you a percentage of what the licensing party earns from your data, rather than a single fee.

Troveo says its datasets sell an average of four times, which is the case for revenue sharing in one number. The risks are the opposite of a fixed fee: you do not know the total, you depend on the agent's reporting and you need audit rights to check it. Labelbox Prism's model, an upfront fee plus a revenue share, splits the difference.

## Will the buyer say publicly that it used your data?

The buyer may have to disclose that it used licensed data, depending on where its models are offered. California's AB 2013, in force from January 1, 2026, requires developers of generative AI systems offered to the public to post a high-level summary of their training data on their website, including the sources or owners of the datasets and whether they were purchased or licensed, according to Perkins Coie.

Models offered in the EU face a similar duty. Article 53 of the EU AI Act requires providers of general-purpose AI models to publish a sufficiently detailed summary of the content used for training, using a template the European Commission published in July 2025. If confidentiality about the deal matters to you, agree in the license how you will be described. Our guide to [the EU AI Act and training data transparency](https://sellmydatatoai.net/guides/eu-ai-act-training-data-transparency) explains the template.

## Who signs the license?

You sign the license directly with the buyer. When we broker a deal, Sell My Data To AI is not a party to the license and does not take custody of your raw data. Full data moves only after you sign.

Before you sign, work through our [AI data licensing agreement checklist](https://sellmydatatoai.net/guides/ai-data-licensing-agreement-checklist) with your own lawyer. We are a broker, not a law firm. To compare structures for your own records, [book a free consultation](https://sellmydatatoai.net/book-a-consultation?side=seller).

## Frequently asked questions

### Do I still own my data after licensing it to an AI company?

You still own your data after a license, unless the contract assigns ownership to the buyer. A license grants the right to use a copy for defined purposes. Read the grant clause carefully, because some contracts called licenses contain wording that transfers rights in practice, especially over derived assets and trained models.

### What is the difference between an exclusive and a non-exclusive license?

An exclusive license stops you licensing the same data to anyone else for the agreed scope and term. A non-exclusive license lets you license it again. AIData.Broker, quoted by Don't Do It Yourself, says an exclusive deal fetches about 2.5 to 5 times a non-exclusive one.

### Can I license the same data to more than one AI company?

You can license the same data to several AI companies if every license is non-exclusive. Endless Data licenses non-exclusively to several labs, and Troveo says its datasets sell an average of four times. Check that no earlier license contains an exclusivity or first-refusal clause before you sign another.

### Can the buyer resell my data?

The buyer can only resell or sublicense your data if the license allows it. Nixon Peabody recommends stating whether data may be shared with affiliates, subprocessors or research partners. micro1's process includes limits on reselling data before personal information is anonymized. Ask for onward transfer to be prohibited or tightly defined.

### Is a revenue share better than a fixed fee?

A revenue share is better when your data can be licensed many times and you can wait for income. A fixed fee is better when you want certainty or are closing the business. Troveo pays a share each time data resells, while Polyshares and micro1 pay one sum after approval.

## Sources

- [How to sell your company's data to AI companies](https://ddiy.co/sell-company-data-to-ai/), Don't Do It Yourself, accessed November 2026
- [How much is company data worth to AI? (2026 prices and real deals)](https://ddiy.co/how-much-is-company-data-worth-to-ai/), Don't Do It Yourself, accessed November 2026
- [Company data partnerships](https://www.micro1.ai/data-partnerships), micro1, accessed November 2026
- [Shopping for AI training data: five diligence questions before you buy](https://www.nixonpeabody.com/insights/alerts/2026/10/26/shopping-for-ai-training-data-five-diligence-questions-before-you-buy), Nixon Peabody, accessed November 2026
- [Selling company data for AI training: what to know before you say yes](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes), Frankfurt Kurnit Klein and Selz, accessed November 2026
- [Best public data providers for AI training and grounding](https://www.getphyllo.com/post/best-public-data-providers-ai-training-grounding), Phyllo, accessed November 2026
- [AI labs are buying data from dead startups to train their models](https://www.technologyreview.com/2026/10/13/1145755/ai-labs-buying-data-from-dead-startups-to-train-models), MIT Technology Review, accessed November 2026
- [Spirit's privacy ombudsman backs Google's $10m data deal](https://thenextweb.com/news/google-spirit-airlines-data-ombudsman-approval), The Next Web, accessed November 2026
- [AB 2013: new California AI law mandates disclosure of GenAI training data](https://legacy.perkinscoie.com/insights/update/ab-2013-new-california-ai-law-mandates-disclosure-genai-training-data), Perkins Coie, accessed November 2026

---

Book a free consultation: https://sellmydatatoai.net/book-a-consultation