Buy 3D and CAD data for AI training
Last updated How we research
In short
3D and CAD data is licensed geometry, from parametric CAD models and BIM files to point clouds and meshes, used to train generative design, 3D generation and spatial AI models. No buyer publishes a price per model. Shutterstock lists more than 1 million 3D models for AI licensing, and Mercor's published listings for manufacturers run from $100,000 to $1.5 million per company archive.
Key takeaways
- AI teams use 3d and cad data for (1) Text-to-CAD and generative design, (2) 3D asset generation for games, film and simulation and (3) point cloud segmentation for robotics, mapping and construction.
- Usually priced per model or per company archive. Not published per model.
- Common delivery formats: Native CAD files plus STEP or IGES exports, IFC files for BIM and LAS, LAZ or E57 point clouds.
- Check before you license 3d and cad data: (1) US copyright: work made for hire for staff designs, written assignments for consultants, (2) US: EAR deemed exports and ITAR technical data for controlled designs and (3) customer drawings under NDA excluded.
- Every seller we shortlist passes 8 vetting checks before you see a sample.
3D and CAD data: buyer's spec
- Used for
- Text-to-CAD and generative design
- 3D asset generation for games, film and simulation
- Point cloud segmentation for robotics, mapping and construction
- Engineering assistants that read drawings and BIM models
- Delivery formats
- Native CAD files plus STEP or IGES exports
- IFC files for BIM
- LAS, LAZ or E57 point clouds
- OBJ, glTF, USD or FBX meshes with textures
- Usually priced
- Per model or per company archive. Not published per model
- Check before you sign
- US copyright: work made for hire for staff designs, written assignments for consultants
- US: EAR deemed exports and ITAR technical data for controlled designs
- Customer drawings under NDA excluded
- Library parts, fonts and stock textures removed
- Title blocks, stamps and site coordinates stripped
- US: BIPA if scans capture face geometry
- Native files with feature history, not meshes only
- UK and EU: UK export licensing and EU Regulation 2021/821
Why AI teams work with us
- 1
Every seller of 3d and cad data we shortlist has passed eight vetting checks, from authority to sign through to a re-identification test.
- 2
Each shortlist comes with a vetting file per seller: rights evidence, provenance record, de-identification method, data manifest and proposed license terms.
- 3
Samples are drawn at random across the archive and labeled with the sampling method, so you can judge the whole package rather than its best rows.
- 4
You contract directly with the seller, and we help agree acceptance tests, staged payments and remedies before any data is delivered.
What is 3D and CAD data?
3D and CAD data is any digital description of shape. CAD models describe parts with exact curves, surfaces and design features. BIM files describe buildings as objects with properties. Point clouds are millions of measured points from laser scanners or LiDAR, and meshes are surfaces made of triangles used in games, film and simulation.
Each kind trains different models, so tell sellers which one you need. Our seller guide to CAD and design files covers the same data from the owner's side.
What do AI teams use 3D and CAD data for?
AI teams use 3D and CAD data for generative design, 3D asset generation and spatial understanding. Native CAD with feature history shows a model how engineers build a part step by step, which an exported mesh cannot.
| Data type | Common formats | Typical training use |
|---|---|---|
| Parametric CAD | Native files, STEP, IGES, Parasolid | Text-to-CAD and design assistants |
| BIM | IFC and native authoring files | Building design and code-checking models |
| Point clouds | LAS, LAZ, E57, PLY | Segmentation for robotics, mapping and construction |
| Meshes and assets | OBJ, glTF, USD, FBX | 3D generation for games, film and simulation |
What open 3D datasets exist?
Open 3D datasets are large but mostly built from public material. The ABC dataset contains one million CAD models, each a collection of explicitly parametrized curves and surfaces, and Objaverse-XL contains more than 10 million deduplicated 3D objects from sources including manually designed objects and photogrammetry scans.
Production engineering records, such as tolerances, assemblies, specifications and revision history, usually sit inside the companies that made them. That is where licensed archives add something open sets do not.
Who sells 3D and CAD data?
3D and CAD data is sold by stock libraries, creator platforms and companies with design archives. Shutterstock's data licensing page lists more than 1 million 3D models for spatial, simulation and physical-world applications, and Wirestock licenses creators' 3D assets and 2D game art. See the Shutterstock data licensing and Wirestock profiles.
Company archives reach the market through brokers and training-data companies. Mercor's published listings, compiled by ddiy.co, run from $100,000 to $1.5 million for manufacturers with 20 to 2,000 staff and from $115,000 to $1 million for construction companies with 50 to 2,000 staff, whose records include drawings. Our page for engineering and architecture firms covers that sector.
How much does 3D and CAD data cost?
3D and CAD data has no published price per model from any AI buyer or marketplace we checked. Shutterstock asks buyers to contact its AI data team, and the Mercor figures are payouts for whole company archives, not prices per file.
Native files with history, linked specifications and test results are worth more than meshes. Our guide to AI training data pricing models covers per-unit and archive pricing.
Which US legal checks apply to CAD data?
The main US legal checks for CAD data are copyright ownership and export controls. Under the US Copyright Office's Circular 30, designs made by employees within their jobs are works made for hire owned by the company, while files by outside engineers and architects stay theirs unless assigned in writing.
Export rules apply to designs of controlled items. Under 15 CFR 734.13(b), any release of controlled technology to a foreign person in the United States is a deemed export, and ITAR's definition of technical data in 22 CFR 120.33 includes blueprints, drawings, photographs and plans for defense articles. Your own foreign staff or contractors can trigger these rules, so ask the seller for the export classification of each product line.
- Customer drawings and specifications under NDA: excluded
- Library parts from component makers: removed
- Title blocks, engineer stamps and file properties: stripped
- Site coordinates in civil and BIM models: stripped where they locate private property
- Scans of people: reviewed, since BIPA lists a scan of face geometry as a biometric identifier
If you hold UK or EU data
If designs come from the UK or EU, export controls still apply. The UK Export Control Joint Unit says emailing controlled technology abroad, or uploading it to the cloud where it is accessed overseas, needs an export license, and EU Regulation 2021/821 controls dual-use items, including technology. UK GDPR or EU GDPR applies to personal data in file properties and scans.
How do you source vetted 3D data?
You source vetted 3D data by naming the format, sector and volume you need, then checking chain of title on a sample. Every seller we shortlist passes our eight-point vetting standard. Book a free consultation or submit a data requirement.
Frequently asked questions
Are STL files good enough for AI training?
Do export controls apply to buying CAD data?
Can I train commercially on open 3D datasets?
Do point clouds contain personal data?
Who owns CAD files made by contractors?
Sources
- Data licensing, Shutterstock, accessed November 2026
- Wirestock homepage, Wirestock, accessed November 2026
- How to sell company data to AI companies (2026 guide), ddiy.co, accessed November 2026
- ABC: A Big CAD Model Dataset For Geometric Deep Learning, arXiv (Koch et al., 2018), accessed November 2026
- Objaverse-XL: A Universe of 10M+ 3D Objects, arXiv (Deitke et al., 2023), accessed November 2026
- Works Made for Hire (Circular 30), US Copyright Office, accessed November 2026
- 15 CFR 734.13: export, Legal Information Institute, Cornell Law School, accessed November 2026
- 22 CFR 120.33: technical data, Legal Information Institute, Cornell Law School, accessed November 2026
- Biometric Information Privacy Act, 740 ILCS 14, Illinois General Assembly, accessed November 2026
- Exporting military or dual-use technology: definitions and scope, Export Control Joint Unit, GOV.UK, accessed November 2026
- Regulation (EU) 2021/821 setting up a Union regime for the control of exports, brokering, technical assistance, transit and transfer of dual-use items, EUR-Lex, accessed November 2026
For AI teams
Ready to source?
Book a consultation to brief us. We shortlist vetted sellers, each with a vetting file covering rights evidence, provenance and a representative sample.
Related guides
- How to sell CAD and design files for AI training
Data types
- How to buy AI training data
Buying data
- Data provenance explained for AI teams
Buying data
- AI training data due diligence checklist
Buying data
- AI training data pricing models
Pricing and value