Arda Raises $160M Series A at a $1.5B valuation for agents that turn CAD files into factory instructions
Arda Industries, the manufacturing AI company started by former OpenAI chief research officer Bob McGrew, announced a $160 million Series A on October 20 at a $1.5 billion valuation. Altimeter, Accel and Khosla Ventures co-led, with XYZ, 8VC, SV Angel and Felicis participating. Including a $72 million seed, Arda has raised $232 million. In March, reports had it raising at $700 million, so the valuation roughly doubled in seven months. Link: https://arda.ai
The product takes a product design, the CAD file, and turns it into factory-ready manufacturing plans: assembly sequences, work instructions, and production processes that can be adjusted as the line changes. Anduril is the first named customer, and the number Arda leads with is that basic work instructions that once took 10 hours now take 15 minutes.
The interesting part is where this sits. Most agent money this year went to software that writes software. Arda is aiming at the translation layer between engineering and the factory floor, the documents that tell a technician what to bolt to what in which order. That work is slow, mostly done by hand, and redone every time the design changes, which makes it the same shape as the coding problem: a structured input, a structured output, and a checkable result at the end of the line.
McGrew leaving frontier research to build for factories is its own signal. The people who built the models increasingly think the money is in the last mile, in industries where nobody has written down the workflow yet. With a defense prime as the first customer, Arda also gets to learn in one of the few places where reshoring and speed are both national priorities.
← Back to all articles
The product takes a product design, the CAD file, and turns it into factory-ready manufacturing plans: assembly sequences, work instructions, and production processes that can be adjusted as the line changes. Anduril is the first named customer, and the number Arda leads with is that basic work instructions that once took 10 hours now take 15 minutes.
The interesting part is where this sits. Most agent money this year went to software that writes software. Arda is aiming at the translation layer between engineering and the factory floor, the documents that tell a technician what to bolt to what in which order. That work is slow, mostly done by hand, and redone every time the design changes, which makes it the same shape as the coding problem: a structured input, a structured output, and a checkable result at the end of the line.
McGrew leaving frontier research to build for factories is its own signal. The people who built the models increasingly think the money is in the last mile, in industries where nobody has written down the workflow yet. With a defense prime as the first customer, Arda also gets to learn in one of the few places where reshoring and speed are both national priorities.
Comments