He sent robots into mine tunnels. Now he wants a factory that builds almost anything.
Founder and CEO of Foundry Robotics - the San Francisco startup chasing the "Everything Factory," an AI-first robotic contract manufacturer built to put American assembly back to work.
Kulkarni, on a hillside with the San Francisco skyline at his back - the same view his factories aim to fill with American-made goods.
The Pitch
Most factories are good at exactly one thing. A line that stamps car doors cannot suddenly assemble drones, and retooling it costs weeks and fortunes. Adarsh Kulkarni's bet with Foundry Robotics is that this constraint - the thing every manufacturer treats as a law of physics - is really just a software problem waiting for the right robots.
Foundry calls it the "Everything Factory." The idea is an AI-and-robotics-first, assembly-focused, dual-use contract manufacturer: a single facility, built around Vision-Language-Action foundation models, that can switch from one product to the next without the traditional retooling overhead. The first targets are deliberately unglamorous - products in high demand from both commercial customers and the Department of Defense, most of which are made abroad today.
It is an audacious goal, and Kulkarni did not arrive at it by accident. He arrived by way of mine tunnels, legged war-robots, self-driving trucks, and one of the most closely watched AI companies of the decade.
Origins
Before the company, before the funding, there was a dark hole in the ground. At the University of Pennsylvania's GRASP Laboratory - one of the most storied robotics labs in the country - Kulkarni earned a Robotics MSE in 2021 and worked on something genuinely strange: coordinating packs of four-legged robots as they explored abandoned mine tunnels for the DARPA Subterranean Challenge. The work became a peer-reviewed paper, "Mine Tunnel Exploration Using Multiple Quadrupedal Robots," published in IEEE Robotics and Automation Letters and presented at ICRA 2020.
It is the kind of problem that teaches you humility fast. Underground, there is no GPS, no map, no margin for a robot that freezes when it sees something it was not trained on. You learn to build autonomy that survives the real world, not the demo.
Kulkarni took that hard-won autonomy into industry. He was an early autonomy engineer at Ghost Robotics, the maker of rugged quadruped robots, where he worked hand-in-hand with the Department of Defense on deploying machines for defense and industrial use. Ghost Robotics was later acquired for roughly $400 million - the kind of outcome that turns an early engineer into someone who has seen the whole arc, from prototype to procurement.
From there he moved to TuSimple, building machine-learning infrastructure for autonomous trucking - the unglamorous plumbing that lets a self-driving system learn from oceans of road data. His GitHub bio still wears that lineage like a badge: "ML Infra @ TuSimple | Ex- Ghost Robotics | Ex- UPenn GRASP."
His next stop was Scale AI, where he joined as a software engineer and rose to Head of Robotics. Scale sits at the center of the modern AI supply chain - the data, the evaluation, the infrastructure - and running its robotics effort put Kulkarni at the exact intersection that Foundry would later exploit: frontier AI models meeting physical machines that have to do real work, reliably, at scale.
The Thesis
There is a reason a robotics lifer would walk away from a senior role at a marquee AI company to go build factories. The pieces finally fit. Vision-Language-Action models can now take a high-level instruction and translate it into physical action. Robot hardware is cheaper and more capable than it has ever been. And the political and economic appetite for reshoring - making things in America again, especially things the defense industrial base depends on - has rarely been stronger.
Foundry's plan is to fuse all three: VLA foundation models, modern robotics, and manufacturing software, so a single cell-based factory can generalize across products instead of being purpose-built for one. If it works, the economics of "made in America" change, because the cost of switching what you make collapses toward zero.
In early 2026, Foundry Robotics announced a seed round led by Garuda Ventures, with participation from Khosla Ventures and a roster of other investors - capital aimed squarely at turning the Everything Factory from a thesis into a building you can walk into.
Illustrative reading of the tailwinds Foundry cites - not a measured index.
The Arc
"The Everything Factory" isn't a slogan. It's a dare to every assembly line that only knows how to make one thing.
FOUNDRY ROBOTICS / THE FOUNDING THESIS
The Long Game
Somewhere along the way, the United States outsourced the muscle memory of making things. Foundry Robotics is Kulkarni's wager that AI can hand that muscle memory back - not by reopening yesterday's factory, but by inventing a new kind that bends to whatever needs building, whether it ships to a hardware store or a forward operating base. It is robotics in service of something concrete: shelves, supply chains, and the unfashionable but essential work of assembly. He spent his twenties teaching machines to navigate the dark. The next act is teaching them to build in the light.