Physical Intelligence, the latest venture from Stripe veteran Lachy Groom, is quietly becoming one of Silicon Valley’s most talked-about robotics startups, building what its team calls the “ChatGPT of robots.” Located in a nondescript San Francisco office marked only by a slightly off-color pi symbol on the door, the company’s headquarters is a raw, open space filled with the hum of activity. Long wooden tables host a mix of snacks and high-stakes robotics experiments—robot arms attempting to fold pants, turn shirts inside out, and peel zucchinis with robotic precision. These aren’t demos; they’re part of a continuous loop of data collection, model training, and real-world testing. At the heart of the operation is a bold vision: to create general-purpose robotic foundation models that can learn and adapt across different tasks and hardware platforms. Sergey Levine, an associate professor at UC Berkeley and co-founder, likens the system to large language models—except instead of processing text, it learns physical tasks through real-world experience. Data gathered from these test stations, in warehouses, homes, and even a dedicated test kitchen, trains the models. When a new model is developed, it’s sent back to these stations to be tested under real conditions. The zucchini peeler, for example, isn’t just peeling—it’s learning the motion of peeling so it can generalize to apples, potatoes, or other foods. The hardware itself is deliberately unimpressive—off-the-shelf robotic arms priced at around $3,500, with material costs under $1,000 if built in-house. But as Levine points out, the intelligence matters more than the machine. “Good intelligence compensates for bad hardware,” he says. The goal isn’t flashy robots, but adaptable, intelligent systems that can be deployed on any platform. Lachy Groom, the 31-year-old co-founder, embodies the restless energy of Silicon Valley’s most driven innovators. Having sold his first company at 13 in Australia, he spent years as an angel investor, backing early-stage startups like Figma, Notion, and Ramp. But his real passion lies in robotics—a field he first fell in love with as a kid building Lego Mindstorms. When he learned that Levine and Chelsea Finn were planning to start a robotics company, he reached out to Karol Hausman, a Google DeepMind researcher, and found what he was looking for: a rare convergence of brilliant minds, a compelling problem, and the right moment. Physical Intelligence has raised over $1 billion, with backing from Khosla Ventures, Sequoia Capital, and Thrive Capital, and is valued at $5.6 billion. Unlike many startups, it doesn’t give investors a clear timeline for profitability. “I don’t give answers on commercialization,” Groom says. “It’s not something people usually tolerate, but they do.” The strategy hinges on building robust, general-purpose intelligence first—so that when new hardware emerges, the model can be deployed with minimal retraining. The company is already testing its systems with real-world partners in logistics, grocery, and manufacturing. Co-founder Quan Vuong, formerly of Google DeepMind, emphasizes the power of cross-embodiment learning: knowledge gained from one robot can transfer to another, drastically reducing the cost of onboarding new platforms. The race is heating up. Skild AI, a Pittsburgh-based rival, recently raised $1.4 billion at a $14 billion valuation and has already begun commercial deployment, claiming $30 million in revenue. Skild argues that most robotics foundation models are just vision-language models in disguise, lacking true physical understanding. Physical Intelligence, by contrast, is betting on pure research and data diversity—delaying commercialization to build something more fundamental. Groom remains undeterred. “It’s such a pure company,” he says. “A researcher has a need, we go and collect data to support it. It’s not externally driven.” The team has already outpaced its initial 5-to-10-year roadmap, and while growth is planned, Groom insists it will be slow. “Hardware is just really hard,” he says. “It breaks. It arrives late. Safety is a constant challenge.” As the zucchini shavings pile up and the pants remain unfolded, the question lingers: can general-purpose robot intelligence become a reality? The answer may take years. But in Silicon Valley, the belief in people like Groom—driven, patient, and willing to bet on the long game—remains strong.