Guides · Published
Best AI training data licensing companies (2026)
Compare six AI training data licensing options for 2026, from marketplaces to rights-cleared libraries. See who each suits and what to check before you sign.
The best AI training data licensing options in 2026 depend on what you have to license. Marketplaces such as Troveo pay rights holders for video, audio and business data, Opendatabay and Depositphotos sell licensed datasets, and Smelted matches US companies' internal business records with AI buyers at no cost to the company. Expert-data vendors such as Mercor and micro1 sell human work rather than licensing records, so they serve a different need.
We make Smelted, so it's on this list. We judged it on the same criteria as the others, using only facts we can support.
How we picked
Each entry is judged on the same five criteria:
- What is licensed: the type of data or work the service deals in.
- Who the seller is: rights holders, companies or a vendor selling its own datasets.
- Cost to the seller: whether the seller pays anything to take part.
- Control: whether the seller chooses what is included and who gets it.
- Fit: the kind of organisation it suits best.
We use the pages' own descriptions and the figures they publish. Where a price or result is not public, we leave it out.
The list
1. Smelted
Smelted is a free matching service that lists anonymised internal business records, such as support tickets, CRM history, SOPs, project files, code, CAD and lab data, for AI data companies to license.
Best for: US companies with 50 to 500 employees and several years of records in tools like Zendesk, Jira, GitHub, Salesforce, HubSpot, Bullhorn, AutoCAD, Revit or Benchling.
- Strengths: The company keeps ownership, since a buyer licenses a copy. It chooses what is included and which buyers it meets, and names, contact details and IDs are removed before anything is shared. The buyer pays Smelted's fee, so the company pays nothing.
- Watch for: Value depends on headcount, years of records and how connected the data is, so the company needs a buyer review of a sample before it knows the figure. Payment usually comes a few weeks to two months after signing. Companies under 30 employees should expect smaller offers, because the buyers who take smaller companies pay less.
2. Troveo
Troveo runs a licensing marketplace that connects rights holders with AI buyers across video, audio, gaming, robotics and business data.
Best for: Rights holders who want a single marketplace to license to many AI buyers.
- Strengths: Troveo says it has paid more than 20 million dollars to over 7,000 rights holders (Troveo). It describes how marketplaces handle the long tail of small licensors that no buyer would negotiate with one by one.
- Watch for: Its published figures cover its own marketplace, so check the current terms before you list.
3. Opendatabay
Opendatabay publishes a guide that compares commercial and general licences for licensed AI training data, for both buyers and sellers.
Best for: Teams that need to understand licence types before buying or selling datasets.
- Strengths: Its guide sets out how licence type changes what a buyer may do with a dataset.
- Watch for: Read the licence on each dataset. A guide explains the options but does not replace a legal review.
4. Depositphotos
Depositphotos sells rights-cleared image and media datasets for AI training and offers custom plans.
Best for: Buyers who need media assets with clear rights for model training.
- Strengths: It is built around rights clearance. Its guide defines AI data licensing in 2026 as the legal authorisation of datasets used to train machine learning and generative AI models (Depositphotos).
- Watch for: It focuses on media. A company with support tickets, code or CAD records will find little here that fits.
5. Mercor
Mercor is one of the expert-data providers that third-party guides list for human feedback and specialist data work.
Best for: AI labs that need trained people to produce or review data.
- Strengths: It supplies human expertise, not existing records.
- Watch for: A company that wants to license its own past records should not expect this kind of work to pay for them.
6. micro1
micro1 is another expert-data provider that guides group with Mercor for expert data and human feedback.
Best for: Labs that want specialist people to generate or grade training data.
- Strengths: Useful when the need is human work on a new task.
- Watch for: As with Mercor, this is a service for buying work, not a licence for existing company records.
Comparison
| Option | What is licensed | Who sells | Cost to seller | Seller controls what is shared | Best fit |
|---|---|---|---|---|---|
| Smelted | Anonymised internal business records | Companies (50 to 500 staff) | Free; buyer pays | Yes: chooses data and buyers | US companies with years of SaaS or engineering records |
| Troveo | Video, audio, gaming, robotics and business data | Rights holders | Not stated on the page we reviewed | Listed by rights holders | Rights holders with varied content |
| Opendatabay | Licensed datasets | Dataset sellers | Not stated on the page we reviewed | Per licence | Buyers and sellers who need licence guidance |
| Depositphotos | Rights-cleared media | Depositphotos | Not applicable | Not applicable | Buyers needing cleared images and media |
| Mercor | Human expert work | Vendor | Not applicable | Not applicable | Labs buying human feedback |
| micro1 | Human expert work | Vendor | Not applicable | Not applicable | Labs buying specialist data work |
FAQ
Are AI companies running out of training data?
AI companies still buy data, and the market has moved toward licensing. Troveo reports that deals range from a few million dollars a year for mid-size publishers to $250 million over five years for News Corp and OpenAI (Troveo). Quartz also describes the News Corp deal as worth more than $250 million, the largest disclosed AI content licensing agreement (Quartz).
How do you sell your data to AI companies?
A company can license a copy of its records through a matching service or a marketplace. With Smelted, the company starts with a 2-minute form at smelted.ai/get-started. Smelted then matches the company with buyers, gives a ballpark range and introduces it to two or three buyers. Names, contact details and IDs are removed first, and the company chooses what is included and who it meets. Companies with 50 or more employees can also book a short call at calendly.com/smelted/intro-call.
How much money do companies make selling their data?
The amount depends on the data. Smelted gives no figure until a buyer reviews a sample, and the value depends on headcount, years of records and how connected the data is. Public deals vary widely, from a few million dollars a year to hundreds of millions for large publishers, so they are not a guide for most companies.
How much is one person's data worth?
Smelted works with companies, not individuals. Personal data raises separate consent and privacy questions. Companies licensing their own internal records should review what those records contain before they share them.