PALO ALTO — In one of the most prominent venture capital bets on physical artificial intelligence to date, Benchmark has led an $85 million Series B financing round in robotics platform Foundry, valuing the Palo Alto startup at $620 million post-money. Benchmark general partner Peter Fenton will join Foundry’s board of directors, alongside existing backer Founders Fund and strategic industrial investor Siemens. The capital injection marks a watershed moment in the commercial evolution of autonomous manipulation from rigid factory pre-programming to general-purpose spatial intelligence.
For decades, industrial automation has remained tethered to hard-coded kinematics. Traditional industrial robotic arms excel at high-speed repetitive tasks—such as spot-welding car chassis in structured automotive lines—but fail catastrophically when confronted with unstructured environments, component variations, or tactile uncertainty. Foundry has developed what it calls a “Large Physical Model” (LPM): a multi-modal foundation architecture trained on over 650,000 hours of synchronized tactile, stereoscopic vision, and force-torque telemetry collected across commercial electronics manufacturing lines and semiconductor assembly cleanrooms.
The Technical Architecture: Large Physical Models and Tactile Feedback Loops
At the center of Foundry’s breakthrough is an end-to-end spatial transformer model named “Foundry-Core-1.” Unlike conventional robotics software stacks that divide perception, motion planning, and actuator control into discrete, fragile software pipelines, Foundry’s neural policy operates directly from sensor inputs to continuous 7-degree-of-freedom motor commands at an unprecedented 500Hz control loop frequency.
By integrating high-density optical tactile sensors into its proprietary compliant grippers, Foundry’s robots can perceive micro-surface textures and dynamic slip in real time. This allows the system to manipulate deformable objects—such as flexible ribbon cables, micro-screws, and delicate printed circuit boards—with dexterity rivaling human assembly line workers. The company’s cross-embodiment training framework enables policies trained in high-fidelity physics simulations (utilizing custom NVIDIA Isaac Sim kernels) to transfer zero-shot into real-world factory hardware without catastrophic distribution shifts.
“Generative AI taught software how to think and write. Physical AI is teaching machines how to interact with the physical laws of our universe. The industrial manufacturing sector represents a $16 trillion addressable market, and Foundry is building the definitive operating system for physical labor.”
— Peter Fenton, General Partner at Benchmark
Unit Economics: Hardware-as-a-Service and 9-Month Payback Horizons
Foundry’s commercial traction has accelerated dramatically by bypassing capital-intensive hardware sales in favor of an SLA-backed Hardware-as-a-Service (HaaS) deployment model. Electronics manufacturing services (EMS) providers and tier-one automotive suppliers deploy Foundry robotic workcells with zero upfront equipment capital expenditure, paying an hourly operational fee pegged directly to successful assembly throughput.
Because Foundry workcells can be reprogrammed for new product introductions (NPI) in hours via zero-shot natural language prompts rather than weeks of roboticist programming, client payback periods have compressed from an industry standard of 36 months down to less than 9 months. In high-turnover assembly environments across Texas, Germany, and Taiwan, Foundry robots have logged over 2.4 million operating hours with an uptime reliability metric exceeding 99.7%.
The Geopolitical Reshoring Dynamic
The broader macro tailwind powering Foundry’s valuation surge is the aggressive push by Western governments and corporate boards to reshore critical supply chains. Driven by geopolitical friction and the CHIPS and Science Act, multinational technology firms are building sprawling domestic fabrication and assembly facilities across the United States and Europe. However, severe labor shortages in skilled manufacturing have threatened to stall factory commissioning schedules.
Foundry bridges this structural labor gap by providing drop-in automated dexterity that does not require specialized roboticists on the factory floor. Assembly operators without engineering degrees can supervise fleets of 20 to 50 Foundry robots using intuitive tablet interfaces, retraining autonomous workcells for new product SKUs using simple conversational demonstrations.
Key Takeaways
- Venture Round Scale: Benchmark leads $85M Series B in Foundry at a $620M valuation, with partner Peter Fenton taking a board seat alongside Founders Fund.
- End-to-End Physical Models: Foundry-Core-1 runs 500Hz direct sensor-to-actuator control loops, replacing fragile legacy robotics software stacks with continuous spatial transformers.
- Tactile Dexterity: Proprietary compliant grippers and optical tactile sensing enable zero-shot manipulation of deformable cables, micro-fasteners, and delicate silicon dies.
- Commercial Model: Hardware-as-a-Service (HaaS) pricing eliminates upfront CapEx, compressing enterprise payback periods to under 9 months with 99.7% industrial uptime.