See if your data qualifies

Guides · Published

Best AI training data licensing companies (2026)

Compare AI training data licensing companies for 2026: marketplaces, matching services and annotation providers, judged on who pays, control and effort.

AI training data licensing companies either match your existing business records with AI buyers, run marketplaces where rights holders list content, or sell annotation work. For most companies with years of internal records, a matching service or a marketplace is the right type to compare first. This list ranks five options on the same criteria: Smelted, Troveo, Opendatabay, Depositphotos and Ficstar.

We make Smelted, so it is on this list. We judged it on the same criteria as the others, using only the facts we have about our own service.

How we picked

Every entry is judged on six criteria:

  1. What you license. Internal business records, media, or labelled datasets built to order.
  2. Who pays you. Whether the company receives money, and whether the buyer or the company pays the platform.
  3. Control. Whether you choose what is included, who the buyers are, and whether names and identifiers are removed.
  4. Ownership terms. Whether you keep ownership or grant a licence, and whether exclusivity is part of the deal.
  5. Effort and timing. How much work the company asks of you, and how long payment takes.
  6. Fit. Which company sizes, industries and data types the service is built for.

Where a page did not state a point, we say so in the comparison table rather than guess.

The list

1. Smelted

Smelted is a free matching service. It lists anonymized internal business records, such as support tickets, CRM history, SOPs, project files, code, CAD and lab data, for AI data buyers.

Best for: US companies with 50 to 500 employees, founded in 2022 or earlier, with years of records in tools like Zendesk, Jira or GitHub, Salesforce or HubSpot, Bullhorn, AutoCAD or Revit, or Benchling.

Strengths: The company pays nothing. The buyer pays Smelted's fee. The company keeps ownership and grants a licence to a copy, chooses what is included and which buyers it meets, and has names, contact details and IDs removed before anything is shared. Smelted finds matching buyers, gives a ballpark range and introduces the company to two or three of them.

What to watch for: The value depends on headcount, years of records and how connected the data is, so the first step is a short form, not a price. A firm figure comes only after a buyer reviews a sample. Plan on one to three months from first call to payment. Companies under 30 employees should know that the buyers who take smaller companies pay less.

2. Troveo

Troveo runs a marketplace where rights holders license content for AI training. It describes itself as having paid more than 20 million dollars to over 7,000 rights holders across video, audio, gaming, robotics and business data (Troveo).

Best for: Rights holders with media or video assets who want a marketplace that handles many small licences at once.

Strengths: The marketplace model solves the problem of negotiating thousands of separate contracts. Troveo reports that 95 percent of its licensors have signed exclusively, which shows how common exclusive terms are on the platform (Troveo).

What to watch for: Because exclusivity is so common there, read the term length and scope before you sign. Its figures come from its own resource pages, so check them against your own review.

3. Opendatabay

Opendatabay is a data marketplace with a guide to AI training licences. Its guide compares commercial and general licences for both buyers and sellers (Opendatabay).

Best for: Sellers who want to understand licence types before listing a dataset.

Strengths: The guide asks sellers to check whether the data has other beneficiaries, whether it was collected with consent, and whether they can stand behind ownership. That checklist is useful whichever provider you choose.

What to watch for: The guide is a reference, not a statement of pricing or payment terms. Confirm those directly before listing.

4. Depositphotos

Depositphotos publishes a guide to rights-cleared AI data licensing, covering how deals work and how to reduce legal risk (Depositphotos).

Best for: Image and media owners who want to learn what a rights-cleared dataset needs before they license one.

Strengths: The focus on rights clearance is clear, and the guide explains the risks in plain terms.

What to watch for: It is aimed at media and images, so it is a weaker fit for internal business records such as tickets, CRM history or lab data.

5. Ficstar

Ficstar lists the leading AI training data providers in 2026. Its list includes HumanSignal, which offers Label Studio as open source plus managed services for image, video, text, audio and time series data (Ficstar).

Best for: Teams that need labelled datasets built or annotated to order, rather than a licence to records they already hold.

Strengths: It covers annotation providers in one place, which helps when you are comparing data work rather than licensing.

What to watch for: Annotation is paid work, not a licence to existing records. Its list does not show the licensing terms or who keeps ownership of the output.

Comparison

CompanyTypeBest forWho pays the companyCompany keeps control of what is sharedEffort on your side
SmeltedMatching serviceUS companies, 50 to 500 staff, with years of business recordsCompany pays nothing; buyer pays Smelted's feeYes: chooses included records and buyers; IDs removedShort form, then a call and sample review
TroveoMarketplaceRights holders with media and videoRights holders are paid; figures on its siteExclusivity common (95% of licensors)Not stated on the page we reviewed
OpendatabayData marketplaceSellers comparing licence typesNot stated on the guide we reviewedGuide asks sellers to confirm rightsNot stated on the guide we reviewed
DepositphotosGuide on rights-cleared dataImage and media ownersNot stated on the guide we reviewedNot stated on the guide we reviewedNot stated on the guide we reviewed
FicstarDirectory of annotation providersLabelled datasets built to orderPaid work, not a licenceNot stated on the list we reviewedDepends on the provider

Why licensing is growing

The market for licensed training data is large and still growing. The global AI training dataset market is projected to reach 16.3 billion dollars by 2033, growing at 22.6 percent a year from 2026, according to Grand View Research (Troveo). Epoch AI has projected that the stock of high-quality public text could be effectively exhausted for training between 2026 and 2032, one reason labs look for non-public data (Troveo).

Deal sizes vary widely. News Corp and OpenAI signed a licensing deal worth more than 250 million dollars over five years, announced in May 2024 (Quartz). Most company-level deals are far smaller, and their value depends on the data itself.

FAQ

How do you sell your data to AI companies?

You either list it on a marketplace or work with a matching service that finds buyers for it. Before you sign, decide what you will include, who you will meet, and whether you keep ownership or grant a licence. Smelted works this way for internal business records: the company chooses what is included, names and contact details are removed, and the company pays nothing.

How much money do companies get for selling data?

The range is very wide. A large publisher deal can reach 250 million dollars over five years, while smaller deals are far less (Quartz). For Smelted, the value depends on headcount, years of records and how connected the data is, and a firm figure comes only after a buyer reviews a sample.

How much is 100 GB of data worth?

There is no reliable per-gigabyte price, because value depends on what the data is and who wants it. A licence to a set of support tickets, CRM history or lab records is valued by its use to a buyer, not its size. Ask a provider for a sample review before comparing offers.

Sources