Physical Intelligence, Stripe veteran Lachy Groom’s latest bet, is building Silicon Valley’s buzziest robot brains

Tucked away in San Francisco, the headquarters of Physical Intelligence offers little indication of the groundbreaking work happening within. Upon entering, there’s no flashy reception area, just a stark concrete space filled with an eclectic mix of tables, monitors, and robotics components. Some tables are cluttered with lunch items like cookies and Vegemite, while others serve as testing grounds for robots attempting everyday tasks. During a recent visit, I observed a robotic arm struggling to fold a pair of black pants, another diligently turning a shirt inside out, and a third expertly peeling a zucchini. Sergey Levine, one of the co-founders and an associate professor at UC Berkeley, likened the operation to a 'ChatGPT for robots'. This analogy highlights their aim to create general-purpose robotic models that can learn from diverse environments, feeding data back into their systems for continuous improvement. The facility houses various testing environments, including a kitchen where robots learn to perform tasks like making espresso — not for staff enjoyment, but as data points for their training. The robotic arms used in these tests are surprisingly affordable at around $3,500, a price that Levine notes is marked up significantly compared to in-house manufacturing costs. Lachy Groom, a notable entrepreneur and investor in his early thirties, is at the helm of Physical Intelligence. His previous success as an early employee at Stripe and angel investor has set the stage for his latest venture. Groom became intrigued by the research emerging from Levine's team and sought to contribute to the robotics field he loved as a child. With over $1 billion raised since its inception just two years ago, Physical Intelligence is focused on developing robust robotic intelligence without rushing to commercialization. Groom admits he does not provide investors with a timeline for profitability, a rarity that seems to be tolerated by backers including Khosla Ventures and Sequoia Capital, who have valued the company at a staggering $5.6 billion. The company's strategy involves cross-embodiment learning, which enables their models to adapt to new hardware without starting from scratch. This approach positions them favorably in the competitive landscape of robotics, where other firms like Skild AI are also vying for leadership with different methodologies. While Physical Intelligence prioritizes research and model training, Skild AI has already made strides in commercial applications, generating significant revenue in its early months. This dichotomy poses an interesting debate in the industry: should the focus be on immediate deployment or on advancing foundational intelligence? As Groom prepares to move on to his next engagement, I reflect on the ongoing challenges of robotics, especially in hardware development, which often complicates progress. Amidst these hurdles, the company's vision remains clear and ambitious, aiming to redefine what is possible in robotic automation. Questions linger about the practicality of robots performing household tasks and the broader implications of their integration into daily life. Nonetheless, Groom’s confidence in his team and their mission suggests that he is undeterred by uncertainty. In Silicon Valley, the backing of visionary thinkers like Groom often leads to groundbreaking advancements, even when the path forward is not immediately clear.