Key Takeaways
- Foundry Robotics raised a $100 million Series A to scale its 130,000-square-foot autonomous manufacturing facility in Emeryville, California.
- The company targets the high-mix, high-volume quadrant of manufacturing, a space incumbents dismiss as impossible because flexible setups usually kill scale.
- Instead of building custom assembly lines for each customer, Foundry standardizes robotic primitives like cabling, fastening, and precise component placement.
- CEO Adarsh Kulkarni argues that factory data and unit economics will prevent hardware startups from bringing manufacturing in-house, even at a volume of one million units per year.
- Decoupling hardware design from physical plant setup mirrors the shift software underwent when AWS replaced on-premise server racks.
The High-Mix High-Volume Quadrant
Traditional manufacturing forces you into a trade-off. You can produce a low mix of parts at massive scale, like Foxconn assembling iPhones, or you can run a high mix of custom parts at low volume in a local machine shop. The fourth corner of the matrix—high mix and high volume—is usually grayed out. If you change what an assembly line makes every week, downtime and retooling expenses eat your margins.
Foundry Robotics claims software and robotic autonomy solve this problem. At their 130,000-square-foot facility in Emeryville, the team assembles defense drones, autonomous submarines, and robot dogs under one roof.
“Fundamentally, right, because we build many things and we're flexible. I can build drones on Monday, batteries on Tuesday, and satellites on Thursday,” (play from 1:44:36) Kulkarni says. “That means to answer your question, it's not a power law product, but instead more like a power law set of primitives.” (play from 1:44:42)
Those primitives are common physical operations: threading a cable, driving a fastener, seating a printed circuit board, and inspecting a seam. By treating physical assembly as reusable software calls executed by arms and vision systems, Foundry switches production lines without gutting the building. Customers buy manufacturing capacity the same way software engineers buy compute cycles.
Why Running Your Own Factory Fails
Hardware founders often assume that once they hit scale, owning the factory is the obvious next step. If your order book expands to a million defense drones, vertical integration feels like the path to fatter gross margins.
Kulkarni says that calculation is dead wrong because it misunderstands where your company creates value:
When you build your own plant, your CapEx is locked into one product generation. If your drone redesign requires a different airframe shape or wire harness routing, you have to write down your tooling or spend months retraining workers. In contrast, an autonomous factory amortizes software upgrades, camera systems, and gripper designs across dozens of clients.
Kulkarni expects this shift to trigger a wave of new hardware companies. “The speed of hardware design innovation and new products being conceived of prototyped produced mass produced distributed will skyrocket the way the pace of software innovation also skyrocketed when we saw the cloud sort of take over as the main form of how compute is handled.” (play from 1:50:41)
What to Do With This
Audit your product roadmap for physical assembly constraints. If your hardware team is designing around the limitations of a contract manufacturer's fixed tooling or budgeting capital to buy your own CNC and pick-and-place lines, stop. Redefine your bill of materials around standard mechanical primitives (standard screw drives, accessible wire channels, pickable flat surfaces) so an autonomous plant can build your hardware through software instructions.