Foundry Robotics, a company founded this year, just raised an oversubscribed $100 million Series A to build what it calls America's first autonomous factory for complex assembly3. The round, led by Marlinspike Partners with participation from Khosla Ventures, Red Glass Ventures, SE Ventures, Wing VC, Axiom Ventures, Hanabi, and Zero Shot Fund, is a direct wager that physical AI models controlling robotic cells can make domestic manufacturing economically viable in sectors where the United States is already straining its production capacity2. That thesis deserves scrutiny, not just enthusiasm. Foundry is not promising to match overseas labor costs through cheaper hands but to shrink the labor line item dramatically, substituting AI-directed robots and a manufacturing-as-a-service model for the workforce that does not exist at the scale defense and commercial customers need. Labor, however, is only one cost line. Foundry has disclosed no unit economics, and until it does, the claim that autonomous factories close the reshoring cost gap is a hypothesis, not a demonstrated result.
The money will fund buildout of a 130,000-square-foot facility in Emeryville, California, a footprint the company says represents a twentyfold expansion of its manufacturing space. Foundry will also use the capital to accelerate deployment of its AI models into real-world manufacturing and to grow its robotics research, engineering, and factory operations teams. The company currently serves defense and commercial industries, focusing on high-value, complex electromechanical products where production is constrained.
The team's composition tells you where the company thinks the hard problems are. CEO Adarsh Kulkarni was employee number four at Ghost Robotics, building legged robots for the U.S. military, and later became Scale AI's first Head of Robotics. Co-founder Emmanuel Lamacchia ran Tesla's Model Y program, which produced roughly 1.5 million cars a year. Former Tesla software leader Anup Karmakar is also on the team, alongside veterans of Raytheon and Northrop Grumman. The roster blends robotics AI expertise with high-volume automotive manufacturing experience and defense-industry domain knowledge, a combination calibrated for exactly the customer set Foundry is targeting: drone makers, satellite builders, and defense primes that need complex electromechanical assembly at rates American factories currently cannot sustain.
Foundry's framing as "manufacturing-as-a-service" is deliberate1. Rather than selling robots to existing factories, the company builds and operates the factory itself, combining AI-powered robotic cells with experienced manufacturing experts. This model shifts the customer's problem from a capital expenditure and workforce challenge to a procurement decision. For a defense prime that needs thousands of drones assembled quickly, the question becomes whether Foundry can deliver units on time and at an acceptable cost, not whether the prime can staff and equip a new production line. The framing, however, is not unique. Established contract manufacturers already offer outsourced manufacturing with significant automation and, critically, already hold defense facility clearances and production certifications that take years to obtain. Foundry's differentiation must rest on a specific technical claim: that its physical AI models can handle complex, variable electromechanical assembly tasks that existing contract manufacturers still perform with manual labor, at higher throughput or lower cost. That claim is plausible given the team's pedigree but unproven in production.
The defense application is where Foundry's pitch is most concrete. Building thousands of drones at a fast rate in America has been a major problem, and Foundry positions itself as a fix. The constraint is real and well-documented across defense procurement: the U.S. has design advantages in autonomous systems but repeatedly runs into production bottlenecks when it tries to scale from prototype to fleet quantities. Foundry's bet is that physical AI models, trained to handle the dexterous, variable assembly tasks that have historically required skilled human workers, can compress the timeline from factory design to volume production. But the defense beachhead carries a structural ambiguity. Defense customers pay premiums for domestic sourcing because national security overrides cost, exactly the condition under which autonomous factories do not need to be cost-competitive to win contracts. Success in defense proves Foundry can deliver units under permissive pricing; it does not, by itself, prove that physical AI closes the cost gap against overseas production in price-sensitive commercial markets.
The harder question is whether the economics work outside defense. Foundry says it serves commercial industries as well, but the company's initial focus on high-value, complex electromechanical products suggests it is starting where margins are fattest and tolerance for premium pricing is highest. That is a rational beachhead. But the manufacturing-as-a-service model's long-term significance depends on whether autonomous factories can eventually compete on cost, not just on speed and domestic sourcing, with conventional overseas production. The path from defense-premium pricing to commercial competitiveness is the gap the company has not yet addressed publicly.
ANALYSIS The oversubscription and the investor roster, spanning deep-tech venture (Khosla), corporate venture (SE Ventures), and defense-adjacent funds, show that multiple investor classes are willing to fund the reshoring thesis at scale. That is a signal about narrative appetite, not economic validation. Investors routinely back pedigreed teams pre-revenue; the round proves Foundry can raise, not that autonomous factories pencil out. Foundry does not need to prove that autonomous factories will replace all manufacturing, but it does need to prove, with throughput data, yield rates, and disclosed contract values, that they can serve the specific segments where domestic production is both strategically necessary and currently insufficient.
What matters now is execution. Foundry has capital, a facility, a team with relevant pedigree, and a customer base in sectors where demand for domestic production is politically and strategically durable. The open questions are operational.
Foundry's value proposition depends on physical AI models that can handle variable, complex assembly tasks. How quickly those models can be adapted to new product types will determine whether the manufacturing-as-a-service model scales or stalls; onboarding a new customer's product cannot require months of retraining if the factory is to serve multiple clients at volume.
Foundry says it currently serves defense and commercial industries, but no specific customers or contract values have been disclosed. Named customers, particularly in defense where contract vehicles are often public record, would validate the demand side of the thesis and distinguish Foundry from a well-funded team with a compelling deck but no production revenue.
Finally, the tariff environment for robotics components remains unresolved. Any final tariff schedule covering the machinery and sensors that go into Foundry's robotic cells will directly affect the unit economics of autonomous assembly.