A manufacturing AI startup just raised $114 million to make old blueprints searchable by shape. Construction has the same knowledge-walkout problem

CADDi, a Japanese manufacturing-AI startup, raised a $114 million Series D on September 15-16 at a $1.2 billion valuation — money aimed at expanding into North America and pushing further into a very specific problem: manufacturers can't search their own drawing archives. CADDi's answer, a tool called CADDi Drawer, reads a company's stockpile of 2D drawings and 3D CAD files and makes every part in it searchable by actual geometry — shape, dimensions, GD&T tolerances — not just filenames. A construction firm with fifteen years of shop drawings and submittals sitting in PDF folders has the identical problem, just with a different name on the door.

What is CADDi actually selling?

Manufacturers accumulate decades of engineering drawings, and most of what's useful in them — how a part was toleranced, who designed it, what it cost to source, whether something nearly identical already exists — lives in the head of whichever engineer has been there longest. CADDi Drawer applies OCR and AI models to extract text, dimensions, and geometric shape data from that drawing archive, then lets anyone search it the way you'd search a parts catalog instead of digging through folders. One customer named in CADDi's own announcements, mobile-crane manufacturer Tadano, is using the platform to cut duplicate part designs and streamline procurement across a group that's grown through acquisitions — the exact "we probably already built something like this" question that wastes engineering hours. CADDi has also built an ITAR-compliant tier for U.S. aerospace and defense manufacturers, where the same problem shows up as a security issue: when a senior engineer retires, the reasoning behind old design decisions can leave with them, undocumented.

Why does a manufacturing funding round belong in a construction brief?

Because the bridge here isn't hypothetical — it's the same failure mode wearing different clothes. A 20-person mechanical sub's estimator knows, from memory, that a specific riser detail was priced and built on a job three years ago, and roughly what it cost. When that estimator leaves, that knowledge goes with them, because it was never captured as searchable data — it was captured as a PDF in a project folder named after a job number nobody remembers. CADDi just got funded at $1.2 billion for building the fix to that problem in manufacturing. Nobody has built the construction-industry equivalent at the same depth yet.

What would a construction version need to do differently?

CADDi's search works because part drawings are relatively standardized — dimensions, tolerances, GD&T symbols follow consistent conventions. Construction drawings are messier: hand-marked redlines, inconsistent CSI-section tagging across firms, scanned as-builts of varying quality. A tool built for this industry would need to:

Is this something to buy, or just something to watch?

Nothing yet — CADDi doesn't sell to contractors, and no AEC vendor has shipped true shape-based drawing search at this depth. The useful move right now is preparation, not procurement: if your firm's shop drawings, submittals, and as-builts are filed by job number instead of tagged by CSI division and assembly type, that's the gap. Whichever CAD, BIM, or document-control vendor closes it first — by adding real drawing-to-drawing search instead of keyword search — will be selling directly into the knowledge-walkout problem every firm with senior staff nearing retirement already has.

It's the same bottleneck on the fabrication side, too: a UK machine-shop AI startup just raised $20 million to kill the manual CNC-quoting bottleneck — another case of a shop's historical drawing and quoting knowledge sitting in one estimator's head instead of searchable data.