Foundry Raises $100 Million for Autonomous Factories
Rocket Drew reports, as an exclusive from the company, that Foundry, a manufacturing startup founded less than a year ago, has raised a $100M Series A led by Marlinspike Partners, with seed lead Khosla Ventures participating. Foundry designs factories around AI-powered robots organised into skill-based work cells (cabling, fastening, precise placement) rather than product lines, so one plant can make many products under a manufacturing-as-a-service model. It has a handful of customers in military, energy, autonomous vehicles and telecoms, and has moved into its first factory, about 130,000 sq ft in Emeryville, California, which is still being set up but already ships to paying customers. CEO Adarsh Kulkarni (ex-head of robotics at Scale AI) and cofounder Emmanuel Lamacchia (ex-Tesla Model Y program lead) avoid humanoids, though Vinod Khosla expects humanoids to become a 'key component' later. Caveats: Kulkarni concedes the flexibility costs some efficiency, robots still struggle with micron-level precision, and Foundry employs hundreds of human technicians whose work feeds its models. No valuation or revenue is disclosed.
Why it matters
A fresh, well-sourced Series A for a sub-one-year-old physical-AI company, with lead investor, seed lead, founders' backgrounds and factory location, which is exactly the data Dealroom profiles need. It is also a useful counter-signal to humanoid valuations and a manufacturing-as-a-service model worth tracking in reshoring and defence-tech coverage. The newsletter's deals section adds other rounds worth logging (Physical Intelligence talks at ~$20B, Hex $1.2B valuation, Infisical, Machenta seed led by Creandum and USV).
Executive takeaways
- Round: $100M Series A, led by Marlinspike Partners; Khosla Ventures (seed lead) participated. Founded less than a year ago. No valuation given.
- Model: skill-based robot work cells instead of product lines, letting one factory build many products; sold as manufacturing-as-a-service to a handful of customers in military, energy, autonomous vehicles and telecoms.
- Footprint: first factory in Emeryville, California, about 130,000 sq ft, already shipping to paying customers; plans for thousands of robots, mostly industrial arms with in-house tools plus vehicles moving work between cells.
- Team: CEO Adarsh Kulkarni ran Scale AI's robotics division; cofounder Emmanuel Lamacchia ran Tesla's Model Y program.
- Data moat claim: Foundry owns the training data generated while making customers' products, which robot vendors selling into other factories often can't negotiate.
- Humanoids: deliberately absent for now (Kulkarni calls them inefficient for greenfield plants); Khosla expects capable humanoids in two years and sees them as a future component.
- Limits: hundreds of human technicians handle micron-precision tasks; versatility trades off efficiency versus single-product lines.
What The Information may be missing
The piece is company-sourced and gives no valuation, revenue, customer names, unit economics or capex per factory, so it's hard to judge how capital-intensive a multi-factory roll-out will be. It doesn't compare Foundry with other manufacturing-as-a-service or robotic-factory players (in the US or Europe), or say how the 'learning factory' data advantage will be measured. Defence customers are mentioned without detail on contracts or certifications. Marlinspike Partners itself isn't described.
