If you run a general contractor, an architecture practice, a design-build firm or a specialty contractor, the models, drawings and project records your firm has built up over years of completed work are what AI labs want their models to learn from. Buyers tell us they're asking for CAD and project files from construction and architecture right now, and few firms know their archive is worth anything. Read on to find out why labs want construction work, what buyers are asking for, who owns the drawings, what in your firm you can license, and how the process runs.
Why AI labs want construction and architecture work
A building is a long run of decisions with a physical result. Someone draws it, someone checks the loads, someone prices it, someone asks what the drawing meant when it met the site, and someone answers. Almost none of that is on the public internet. The models are in Revit and AutoCAD, the calculations are in your engineers' files, and the record of how the design changed and why is in your RFIs, submittals and change orders. Your firm holds years of it, and every project ends with a result anyone can check.
OpenAI's GDPval benchmark, which measures AI models on real professional work, lists CAD design files among the work product it tests and names an engineering blueprint as one example of that work. Labs want models that can read a drawing, reason about a structure and follow a project from design to closeout, and the only place that training data exists is inside firms like yours. Our guide to who buys company data for AI training covers the five kinds of buyer and how you reach them.
What buyers are asking for from construction firms
Buyers are specific about this industry, and three things come up every time.
They want complete projects. A finished building with the architectural, structural and mechanical, electrical and plumbing work for the same project is worth far more than a folder of unrelated drawings. The disciplines together show how one decision constrained the next.
They want native files. A Revit model or a DWG file carries the layers, the objects, the parameters and the history that a flattened PDF loses. Native BIM and CAD, with the drawing sets exported from them, is what buyers ask for first. A PDF set on its own is the floor.
They want records that link. A drawing tied to the calculation behind it, the specification it has to meet, the RFI that questioned it, the change order that revised it and the as-built that recorded what was built is a complete record of how construction work happens. Buyers are also asking for structured records more than text across every industry, so your project management system, your model files and your schedules matter more than your inbox.
Firms with that kind of archive are the ones buyers describe when they ask us for construction data. Our guide to how engineering firms can license their data to AI labs covers the design and test side of the same work, and our guide to how manufacturers can license their data to AI labs covers the other industry where buyers ask for assemblies and engineering notes.
Who owns the drawings
Ownership in construction is set by contract, and it decides what you can license before anything else does. Under the standard architect's agreements, the architect keeps ownership of the drawings and models and gives the owner a license to use them for that project. Under most construction contracts, the contractor holds the drawings and the project records under the owner's and the architect's terms. Subcontractors own their shop drawings in many cases, and manufacturers own the product models they supply. Your firm's position depends on which seat it sat in and what each contract says.
Two things follow from that. An architecture or engineering practice usually owns most of its design work and can license it, subject to any contract that says otherwise. A general contractor owns its own records (estimates, schedules, logs, cost reports, the reasoning in its RFIs and submittals) and needs to check each project contract before the drawings themselves are in scope. Our guide to what agencies and consultancies can license to AI labs covers how to read a client contract with licensing in mind.
What a construction or architecture firm can license
| What you hold | What it shows | What has to come out first |
|---|---|---|
| BIM models and CAD drawings (Revit, AutoCAD, ArchiCAD, IFC exports) | How a building was designed, element by element, and how the design changed through the phases | Anything your contract gives to the client, and manufacturer product models you downloaded under a license |
| Drawing sets by phase and record drawings | What was issued for bid, for permit, for construction, and what was built | Client logos and identifying details on the sheets |
| Structural, energy and other calculations | The reasoning behind the design | Licensed software templates and third-party reports you don't own |
| Specifications and submittals | What the design required and what the contractor proposed to meet it | Vendor pricing and submittals covered by a confidentiality agreement |
| RFIs, change orders and meeting minutes | How the design met the site and what changed, with the reason | Names and personal details, and anything about a dispute |
| Schedules, estimates and cost reports | How a project was planned and priced, and how it ran against the plan | Subcontractor bids and unit prices under a confidentiality agreement |
| Daily logs, inspection reports and progress photos | What happened on site, day by day, and how the work was checked | Names of workers, and anything that identifies a person |
The value is highest where these connect. A model linked to its calculations, its specifications, the RFIs that questioned it and the record drawings that closed it out is the dataset buyers describe. A single discipline from a single project is still worth something. The whole project is worth more.
Most firms hold this across a few systems. The models live in Revit or AutoCAD and a cloud platform like Autodesk Construction Cloud, the project records live in a tool like Procore or Bluebeam, the schedules live in Primavera P6 or Microsoft Project, and the estimates and cost reports live in your accounting system. Our guide to licensing company documents from Google Drive, OneDrive and Dropbox covers the drawing sets and reports that ended up in shared folders instead.
What stays out
You decide the exclusions before anything is exported:
- Anything your contracts assign to the client or restrict you from reusing.
- Projects with security restrictions, including government, defense, critical infrastructure and secure facilities, and any drawing a client asked you to keep confidential.
- Subcontractor bids, vendor pricing and anything else received under a confidentiality agreement.
- Licensed content, like manufacturer product models, code and standards subscriptions, and software template libraries.
- Current bids and active projects you want to protect. Many firms start with completed and closed-out projects.
- Names and personal details of staff, workers, inspectors and clients, which are replaced with pseudonyms to a documented standard.
- Safety incident records that identify a person, and anything about a dispute or a claim.
- Credentials and access details.
This isn't legal advice, and your counsel should sign off on the scope before anything is exported. Firms that do most of their work under a client's contract should read the client-work exclusions first.
Work that gets built can be checked
Buyers tell us that the more objectively a piece of work can be checked, the more readily they license it. Construction is checked at every stage. The structure stood or it didn't. The permit was issued. The RFI was answered and the answer was built. The schedule was met or the daily logs show why it wasn't. Records that carry their outcome with them are worth more than records that don't, and construction records carry theirs by default. Our guide to how much your company's data is worth to AI labs covers the six factors behind the price.
What it's worth
Typical full-company licensing deals start at six figures, and buyers pay a premium for work in the industries they're asking for. There's no rate card, and a real number takes a scoping call. One public reference point is Spirit Airlines, whose operating records drew bids of 10 million and 12.5 million dollars in bankruptcy court, and that package included years of maintenance records on physical assets. Our breakdown of the Spirit Airlines data deal covers what was in it.
How the process runs
Licensing through us runs in five steps, and then you're paid:
- Assessment. You take the free assessment, we hold a short call, and you decide what's in and what's out.
- Agreement. You review and sign standard terms covering your license and your earnings.
- Verification. You show us the data exists and matches what you described. Screenshots of your systems are enough, and nothing gets connected or installed.
- Snapshot. We build an anonymized profile of your data and put it in front of buyers, without your firm's name.
- Data extraction. When a buyer wants to go further, you export the files with our guidance, and we replace names and personal details with pseudonyms and document what's included in a manifest.
- Get paid. Buyers evaluate the snapshot without knowing whose it is, the first sale closes, and you're paid on every sale after it.
You keep ownership the whole way through, and your firm is never named publicly. Our guide to how long it takes to license your company's data covers the timeline and what buyers check, and our owner's guide to licensing company data for AI covers the full process.
Where to start
The quickest way to find out what your firm's projects could license for is to take our free data value assessment (ten questions about your systems, history and industry, about five minutes). You get an estimate, and a scoping call after that gets you a real number.
Where Troveo fits
Troveo is the marketplace that puts your data in front of all five kinds of buyer. We work with more than 40 active buyers and we've paid rights holders more than 20 million dollars. You sign one license, you decide what's in and what's out, we verify and profile your data, we handle the packaging and every buyer conversation, and you're paid on every sale with no fees and no deductions. Start with the data value assessment, read our business data page for how the process works, or talk to us about what your firm holds.