CADDi hits $1.2B: your old drawings are the asset

The least glamorous asset in most businesses is a folder of PDFs nobody can search. CADDi just raised $114 million at a $1.2 billion valuation on the premise that making those files legible to software is a venture-scale business. For any operator sitting on twenty years of drawings, quotes, and spec sheets, that is worth a minute of your attention.

What actually happened

Fortune reported the Series D on September 15: $114M across eight new and returning investors, including Moore Strategic Ventures, Coreline Ventures, Toyota's Woven Capital, and Salesforce Ventures, with Atomico, Globis Capital Partners, and JPS Growth returning. Total raised is now $234 million. The valuation is up from $470 million in March 2025 — roughly 2.5x in eighteen months.

The technical shape matters. CADDi runs a proprietary model for product data — 2D drawings and 3D CAD files, formats general-purpose LLMs handle badly — and leans on general-purpose models for documents and spreadsheets. The product line is CADDi Explorer (the renamed Drawer, a manufacturing discovery engine) and CADDi Agent, which acts on manufacturing data inside a company's own operating context.

The traction numbers: operations in 22 countries, more than half of Japan's 100 largest manufacturers as customers, headcount around 900 (up from 600 in early 2025), and sales the company says are doubling year over year. It did not disclose revenue, so treat the growth rate as a company claim.

Why structured data matters for your business

Here is the part that generalizes past manufacturing. CADDi is not selling a smarter model. It is selling a semantic layer over documents that were never machine-readable — so that a person and an agent can work against the same data without a manual handoff in the middle. Investors valued that plumbing at $1.2B. That should recalibrate how you think about your own file share.

Most AI projects we see fail at exactly this point. Someone connects a chat interface to a document folder, gets plausible-sounding answers about the wrong revision of the wrong quote, and concludes AI is not ready. The model was fine. The retrieval had no idea that quote_final_v3_REALFINAL.pdf superseded four other files, or that the part number on page two is the one that matters.

The fix is not exotic. Pick your highest-value document class — past quotes, supplier specs, service records — and give it three things: a stable identifier, a version rule, and extracted fields you can query. Do it for one class, measure how often the answer is right, then do the next. That is the whole program. CADDi built a billion-dollar company doing it for drawings; you can do it for your quotes in a quarter.

Sources: Fortune, CADDi.