In recent years, major corporations have come to understand that unlocking the full value of AI agents requires redesigning business processes from scratch, rather than simply automating existing workflows. Now, manufacturers deploying robots powered by physical AI models are arriving at the same realization. Foundry, a startup less than a year old, is a prime example. The company exclusively told The Information that it has completed a new funding round totaling $100 million to build automated factories. The Series A round was led by Marlinspike Partners, with Khosla Ventures, which had previously led Foundry's seed round, participating as a follow-on investor.
Foundry designs its factories around AI-driven robots, rather than cramming robots into already-built production lines. The robots are organized into multiple work cells, each focused on skilled tasks such as wiring, fastening, and precise placement, rather than being dedicated to producing a specific product. Thanks to this flexibility, Foundry operates on a "manufacturing-as-a-service" business model, enabling it to produce a wide variety of products for customers within the same factory. It already has a small number of customers spanning the defense, energy, autonomous vehicle, and telecommunications industries.
Co-founder and CEO Adash Kulkarni said: "The value of an autonomous factory lies in the ability to produce many different products within the same facility." "A model that has thoroughly learned fastening tasks can be used to manufacture satellites, drones, robot dogs, and autonomous submersibles," he said. "We make drones on Monday, batteries on Tuesday, and satellites on Thursday."
Kulkarni acknowledged that this flexibility comes at the cost of some production efficiency. But for customers, the versatility of being able to rapidly iterate on products is more valuable. "Traditional large-scale automated factories are extremely efficient, but their efficiency is limited to a single objective: mass-replicating identical products with almost no room for design adjustments."
Khosla Ventures founder Vinod Khosla noted in an interview that, at least for the next several years, dedicated factories remain more suitable for high-volume production scenarios like iPhones and automobiles. However, Foundry's model can "adapt to market demand and rapidly switch production categories within extremely short cycles," creating an opportunity for American factories to compete with low-cost labor in China. "American factories must adopt a different approach: more capital-efficient, more product-adaptive, faster to respond to market demand, and close to the consumer market. A 60-day shipping cycle is no longer acceptable."
Kulkarni previously led the robotics division at Scale AI. Scale AI is a data-labeling startup partially owned by Meta, where Kulkarni oversaw third-party robotics data operations. His co-founder, Emmanuel Ramakia, previously managed the Model Y program at Tesla.
Foundry recently moved into its first factory in Emeryville, California, spanning approximately 130,000 square feet. The factory is still under construction but has already begun delivering products to paying customers. The facility will eventually deploy thousands of robots: primarily industrial robotic arms equipped with Foundry's proprietary tools, along with a fleet of transfer equipment to move semi-finished goods between work cells.
One type of robot is completely absent from the factory floor — bipedal humanoid robots. In recent years, the promise of humanoid robots enabling manufacturing automation has pushed the valuations of numerous humanoid robotics startups to astronomical levels. But Foundry's decision not to use humanoid robots suggests that the market size for humanoid robots may not be as large as some supporters claim. "I've worked on bipedal robots, and I'm bullish on the direction — the technology itself is stunning — but humanoid robots are inefficient in many scenarios," Kulkarni said. "Deploying humanoid robots in existing old factories with manual operations might make sense, but if you're designing a brand-new factory from scratch, there's absolutely no need to rely on humanoid solutions." (Khosla has predicted that robotics startup valuations will correct over the next two years, while also stating that in the long run, humanoid robots will become a "key component" of Foundry's business.)
Foundry's robots also have limitations. Kulkarni said that even after training for specific manufacturing tasks, AI robots still struggle with certain processes, such as operations requiring micron-level precision. In such cases, Foundry arranges for hundreds of human technicians to perform the work; data collected from human work processes is used to iteratively improve the AI models.
Foundry holds a major advantage over companies that sell robots to factories and warehouses: it can retain data generated across the production of various products. Many traditional factories are unwilling to let outside companies collect data from their facilities. "For us, negotiating data ownership isn't difficult," Kulkarni said. "This training data is a byproduct of helping customers streamline their manufacturing processes."
Khosla said that the ability of robots to continuously learn as they work is the core highlight. "This is a factory that learns on its own. Even if it learns to assemble defense equipment, the capabilities it acquires can be reused to assemble alarm clocks."
Comments