Know a company with years of records? Refer it and earn up to $100K
3D and design data

3D Model Catalog Licensing: Who Buys 3D Models for AI?

Who licenses whole 3D model catalogs for AI, how store terms limit it, pooled dataset deals versus direct licenses, and why no buyer publishes a price.

Short answerChecked

No buyer publishes a price per model, but whole 3D catalogs are being licensed to AI companies. One studio licenses 4,390 models and 7,794 material sets it fully owns, with an IP warranty and indemnity; another offers more than 90,000 models built to one standard over 20 years. Buyers are AI labs, robotics teams and data intermediaries. Store terms and third-party assets decide what you can license, so check them first.

A signed licensing agreement with a fountain pen and archive boxes
Key points
  • Single-owner catalogs with an IP warranty sell as training data; mixed-rights libraries are hard to license.
  • No marketplace or AI buyer publishes a per-model price.
  • Selling through a store pays a share of pooled dataset deals; licensing the catalog yourself keeps the terms in your hands.
  • Assets bought from stores such as KitBash3D cannot be passed on for AI training.

Who licenses 3D catalogs to AI companies today, and on what terms?

The public record and the store rules, as read on 19 November 2026:

Buyer or sourceStated terms or priceRecords asked forDate checked
CGAxis (3D studio), own catalogNo price; IP warranty and indemnification4,390 3D models and 7,794 PBR material sets it fully ownsEmail of 14 Jul 2026, read 19 Nov 2026
3D Molier (3D studio), catalog offerNo priceMore than 90,000 models from one studioPost of Jul 2026
TurboSquid (marketplace), data dealsPooled fund shared with contributors; opt-out availableContributor models included in curated datasetsFAQ of 29 Sep 2026
CGTrader seller termsRule: CGTrader may use products for machine learning unless the seller picks a No AI licenseModels listed for sale19 Nov 2026
Sketchfab Standard LicenseRule: a NoAI tag for free and paid modelsModels under the Standard License19 Nov 2026
KitBash3D licenseRule: assets cannot be used for AI trainingPurchased kit assets19 Nov 2026

Who buys 3D model catalogs and product scans for AI training?

Buyers are AI labs building 3D and material generators, robotics teams that need objects for simulation, and intermediaries that resell to both. The CGAxis founder put it this way: "Our store customers are 3D artists; data buyers are AI labs, robotics teams and brokers." (LinkedIn).

What they pay for is clean rights. CGAxis offers "7,794 human-authored PBR material sets and 4,390 3D models, fully owned by us, with an IP warranty and indemnification", and says "most 3D AI is trained on scraped datasets under contested licenses" (CGAxis). 3D Molier offers a catalog of "90,000+ models - is built by one studio, to one production standard, for over 20 years" (3D Molier). One owner, one standard and a long history are what both offers lead with.

The uses differ by buyer. Labs training models that turn text or images into 3D want many distinct objects with clean geometry and materials. Material and texture models want PBR sets with every map. Robotics teams want household and industrial objects at true scale, with parts that open, slide or come apart, to fill simulated rooms. A catalog strong in one of these should say so on its first page.

Product scans and manufacturer CAD catalogs are a close cousin. They carry real dimensions and materials, which robotics simulation needs. If your catalog came from client work, read product design consultancy CAD files first; for scanned sites and terrain, see aerial lidar and 3D data.

Game assets count too: Patronus AI lists the code, original assets and history of a released game among the material it licenses (our Patronus AI review).

How much are a company's 3D product models and CAD catalog worth to AI companies?

No buyer or marketplace publishes a price per model or per catalog, as of 19 November 2026. The studios licensing their catalogs say what they offer, not what they are paid. Any figure is set in negotiation.

Store dataset deals show how the money is split, not how much it is. TurboSquid says "earnings are pooled in a collective fund and distributed to contributors", and "Contributors will receive a share of the entire contract value paid by customers licensing datasets." (TurboSquid). A contributor gets a slice of a pooled deal; a studio that licenses directly negotiates the whole of it.

What sets a direct price is the same as in any data license: how hard the catalog would be to rebuild, how clean the rights are, whether the license is exclusive, and what the buyer may train. A catalog that took a team twenty years to build, with one owner and a warranty, is a different asset from a folder of mixed downloads, even at the same model count.

  • Rights: single owner, every artist under contract, no store assets mixed in.
  • Quality: clean topology, real-world scale, PBR materials, consistent naming.
  • Metadata: category, dimensions, material, polygon count and source files for each model.
  • Size and variety: thousands of distinct objects beat many versions of one.

For a first read on your own catalog, use the estimate.

Can a 3D asset studio license its model library to AI companies for training?

This is general information, not legal advice. Yes, for models the studio owns outright, and the store terms it accepted decide the rest. CGTrader's seller terms let it use products for machine learning "unless the Seller opted out by licensing the Product offered for purchase under the licenses marked with "No AI"" (CGTrader terms). In plain terms: models on CGTrader without the No AI license may already be in its own AI work, though you still own them.

Sketchfab says "we've expanded our Standard License so the NoAI tag can be applied to both free and paid models under our Standard License" (Sketchfab). And assets you bought cannot be passed on: KitBash3D's terms say "KitBash assets cannot be used for AI training or neural networks" (KitBash3D). Strip purchased kits, fonts, textures and scans out of any catalog you license.

Client work is the other limit. Models made for a client under a work-for-hire contract usually belong to the client. See licensing design agency files.

One marketplace a studio might once have listed with, Human Native, now belongs to Cloudflare; our Human Native review says what its site offers creators today.

Should we sell through asset stores or license the whole catalog?

Through a store, you share in deals the store makes. TurboSquid adds that contributors can opt out: "We have an option in the contributor account settings that will allow you to opt your content out of future dataset opportunities." (TurboSquid). That suits an artist with a few hundred models.

A studio with thousands of models it owns has another choice: license the catalog directly, set the field of use, and offer a warranty that a store cannot. That is the route CGAxis and 3D Molier have taken. It takes more work: a data card, a sample and a contract. Both routes can run at once, if the store terms allow it and the direct license is non-exclusive.

The choice also depends on how the catalog was built. A studio whose models were made in house by staff on employment contracts can warrant them. A catalog assembled from freelancers, bought packs and client leftovers usually cannot, and a store's pooled deal may be the safer way to earn from it.

What do 3D data buyers ask to see first?

A description, then a sample. The description is one page: how many models and materials, which categories, file formats, real-world scale, polygon ranges, how many artists made them over how many years, and who owns them. The CGAxis offer leads with counts, ownership and the warranty, which is the order a buyer reads in.

The sample is a few hundred models across categories, with their metadata table and renders, sent under an evaluation license that bars training and requires deletion. Buyers check topology, UVs and materials in their own tools, and they check the rights column against your contracts. A sample with gaps in the rights column ends most talks early, so clean it first.

What should a 3D studio check before it signs a catalog license?

Check these before any model leaves the studio:

  • Every store agreement you accepted, and whether any grants the store its own machine learning rights.
  • Artist and contractor contracts: who owns each model, and whether former staff assigned their rights.
  • Client work: models made for a client usually belong to the client, even if you kept the files.
  • Third-party content inside models: purchased kits, textures, HDRIs, fonts and brand logos.
  • The warranty and indemnity you offer, with a cap and a list of models it covers.
  • Exclusivity, field of use, resale, deletion and what happens to models already trained.

Should we take the money, or are we selling off our own future with our 3D catalog?

It is a fair worry. A model trained on your catalog may one day generate assets that compete with your store sales. You can limit that, but you cannot undo training once it has happened.

Weigh it against what the catalog earns today. Store income from older models often falls each year, while a license pays now. The question is whether the license price covers the store income you might lose, and whether the terms leave you free to keep selling and to license again.

  • License non-exclusively, so you can license again and keep selling in stores.
  • Limit the field of use: for example, robot simulation and research, but not a 3D generator sold to artists.
  • Bar resale and sublicensing, and require deletion of the data at the end of the term.
  • Keep source files and new work out of scope, so the catalog you sell next year is still yours alone.

Price the license against what you give up, not against headline ranges. The data offer check helps you read an offer.

Which companies pay for 3D models and scans to train generative AI?

The public record shows AI labs, robotics teams and intermediaries as buyers, studios like CGAxis and 3D Molier licensing direct, and marketplaces like TurboSquid pooling contributor models into datasets. None publishes a price. Ask any buyer who the end user is, whether the license is exclusive, and whether your models can be resold.

Intermediaries can find buyers a studio would not reach alone, but they add a layer between you and the lab. Read whether the intermediary licenses your catalog on to several labs, what it keeps, and whether you can see who ends up using the models. A direct license with a known lab is simpler to police.

Origin is an independent data partner. We help you license your data. We start from a description of the catalog and its rights, and the studio signs any license itself. See what your catalog could be worth.

Origin Data Partners works with companies that own data and helps get it bought. We are independent of the services on this page, and a buyer may pay us when a deal closes (how Origin is paid). See what your data is worth.

Common questions

A data company offered to buy our 3D catalog outright. Is that legit and what should we check before we sign?

It can be. Ask who the end buyer is and why the company wants to own the catalog rather than license it. A sale ends your store income and your right to license again, so it should be priced for that. Check that every model in scope is yours, with no purchased kits or client work. The data offer check lists the questions to send back.

I am the lawyer for a 3D asset studio. What should its catalog license for AI training include?

A schedule of the models in scope; a warranty limited to models the studio owns, with an indemnity cap; the field of use; exclusivity and a ban on resale and sublicensing; delivery format and metadata; deletion at the end of the term; and what happens to trained models. Check the store terms the studio accepted, since some grant the store its own machine learning rights.

Can we license 3D models we bought from asset stores as part of our catalog?

Usually no. A purchase license lets you use the asset in your own work, not pass it on as training data, and some sellers say so outright: KitBash3D bars AI training on its assets. Remove purchased models, textures, HDRIs and scans from the catalog before you license it, and keep a list of what you removed.

What metadata makes a 3D catalog easier to license for AI?

One row per model: name, category, real-world dimensions, materials, polygon count, file formats, creator, creation date and rights status. Add renders from fixed angles and the source files. Buyers want to filter and check a catalog without opening every file, and a clean rights column answers the first question their lawyers ask.

Does an IP warranty or indemnity matter to buyers of 3D training data?

Yes. Much 3D AI has been trained on scraped data under contested licenses, so buyers pay attention to catalogs that come with a warranty. CGAxis leads its offer with an IP warranty and indemnification. Offer one only for models you own outright, cap it, and leave out anything you cannot trace to an artist under contract.

Do product scans and CAD catalogs count as 3D model catalogs for AI buyers?

They are close, and often more useful for robotics. Scans and CAD carry real dimensions and materials, which simulation needs. The same rules apply: you need to own them, and CAD made for clients usually belongs to the client. Describe the products, scan method, accuracy and formats, and keep customer names and drawings out.

Sources and further reading

What is your company’s data worth?

AI companies pay $20K to $5M for company records. Six questions, no files.