Physical Intelligence, Stripe veteran Lachy Groom’s latest bet, is building the brains of Silicon Valley’s coolest robots

From the street, the only sign I could find of the Physical Intelligence headquarters in San Francisco is the pi symbol, which is a slightly different color than the rest of the door. When I enter, I am immediately confronted with activity. There is no reception desk, no logo shining in the fluorescent lights.

Inside, the space is a giant concrete box made somewhat less austere by a random expanse of long, blond-wood tables. Some are clearly for lunch, dotted with boxes of Girl Scout cookies, jars of Vegemite (someone here is Australian), and little wire baskets stuffed with lots of condiments. The rest of the tables tell a completely different story. Many are loaded with screens, robot parts, tangles of black wires, and fully assembled robotic arms in various states to try to get a handle on mundane matters.

During my visit, one arm was folding a pair of black pants, or trying to. It’s not going well. Another method tries to turn the shirt upside down with some sort of design that suggests it will work eventually, but not today. The third method – and this method seems to have found its purpose – is to quickly peel the zucchini, after which the shavings are supposed to be placed in a separate container. The mulch is going well, at least.

“Think of it like ChatGPT, but for robots,” Sergey Levin told me, pointing toward the robotic ballet unfolding across the room. Levin, an associate professor at the University of California at Berkeley and one of the founders of Physical Intelligence, has the friendly, bespectacled demeanor of someone who has spent a lot of time explaining complex concepts to people who don’t immediately understand them.

What I’m seeing, he explains, is a continuous loop testing phase: Data is collected at robotics stations here and in other locations — warehouses, homes, wherever the team can set up shop — and this data trains general-purpose robotic foundation models. When researchers train a new model, it returns to stations like this for evaluation. The pants folder is someone’s experience. This is the shirt maker. Perhaps a zucchini peeler would test whether the model could generalize across different vegetables, learning the basic movements of peeling well enough to handle an apple or a potato like never before.

The company runs test kitchens in this building and elsewhere, including people’s homes, Levine says, using off-the-shelf equipment to expose the robots to different environments and challenges. There’s a sophisticated espresso machine nearby, and I assume it’s for the employees until Levin explains that no, it’s there for the robots to learn. Any foamy latte is data, not a feature of the dozens of engineers on site who are often poring over their computers or hovering over their automated experiments.

The hardware itself is intentionally unglamorous. These weapons sell for about $3,500, with what Levin described as a “huge premium” from the seller. If they manufactured them in-house, the cost of materials would drop to less than $1,000. A few years ago, he says, the robotics world would have been shocked that these things could do anything at all. But that’s the point – good intelligence makes up for bad hardware.

As Levin apologizes, Lachie Groom approaches me, moving through the space with the determination of someone who has six things going on at once. At 31, Groome still has the precocious qualities of Silicon Valley’s boy prodigy, a title he earned early, having sold his first company nine months after founding it when he was 13 in his native Australia (that explains Vegemite).

When I first contacted him earlier, as he was welcoming a small group of sweatshirt-clad visitors into the building, his response to my request to spend time with him was immediate: “Absolutely not, I have meetings.” Now he has ten minutes, maybe.

He found what he was looking for when he began following academic work coming out of the labs of Levin and Chelsea Finn, a former Berkeley PhD student of Levin who now runs her own lab at Stanford focused on machine learning. Their names kept popping up in everything happening in robotics. When he heard rumors they might start something, he tracked down Carol Hausman, a Google DeepMind researcher who also taught at Stanford and whom Groom knew was involved. “It was just one of those meetings where I walked out and it was like, ‘This is it.’

Groom never intended to become a full-time investor, he told me, though some might wonder why he didn’t given his track record. After leaving Stripe, where he was an early employee, he spent nearly five years as an angel investor, making early bets on companies like Figma, Notion, Ramp, and Lattice while searching for the right company to start or join. His first investment in robotics, Standard Bots, came in 2021 and brought him back to the field he loved as a child building Lego Mindstorms. As he jokes, he was “on vacation much more like an investor.” But investing was just a way to stay active and meet people, not the endgame. “I had been looking for a company for five years to start post-tape,” he says. “Good ideas at the right time with a good team – (that’s) very rare. It’s all about execution, but you can execute a terribly bad idea, and it’s still a bad idea.”

The now two-year-old company has sparked excitement More than $1 billionWhen I asked him about his runway, he was quick to explain that it doesn’t actually burn that much. Most of its spending goes toward computing. After a moment, he acknowledged that under the right conditions, and with the right partners, he would raise more. “There’s no limit to the amount of money we can actually put into the business,” he says. “There’s always more computing you can use to solve a problem.”

What makes this arrangement particularly unusual is what Groom did not offer his supporters: a timeline for turning physical intelligence into a money-making endeavor. “I don’t give investors answers about marketing,” he says of backers that include Khosla Ventures, Sequoia Capital, and Thrive Capital, among others that have valued the company at $5.6 billion. “That’s a strange thing, that people tolerate that.” But they tolerate that, and they may not always do that, which is why the company has to be well capitalized now. Not because it has to be, but because it enables the team to make long-term decisions without any compromises.

Quan Vuong, another co-founder who came from Google DeepMind, explains that the strategy revolves around learning across embodiment and diverse data sources. If someone built a new hardware platform tomorrow, they wouldn’t need to start collecting data from scratch — they could transfer all the knowledge the model already has. “The marginal cost of integrating autonomy into a new robotic platform, whatever that platform is, is much lower,” he says.

The company is already working with a small number of companies in different sectors — logistics, grocery, the chocolatier across the street — to test whether its systems are good enough for real-world automation. Fung claims that in some cases this is indeed the case. With an “any platform, any task” approach, the space for success is large enough to start verifying automation-ready tasks today.

Physical intelligence is not alone in chasing this vision. The race to build general-purpose machine intelligence – the foundation on which more specialized applications can be built, much like the MBA models that took the world by storm three years ago – is heating up. Pittsburgh-based Skild AI, which was founded in 2023, just this month raised $1.4 billion. Value: $14 billion It takes a markedly different approach. While physical intelligence is still focused on pure research, Skild AI has already deployed its “full-body” Skild Brain commercially, saying it generated $30 million in revenue in just a few months last year across security, warehouse and manufacturing.

Skild even publicly photographed competitors, He argues on his blog That most “robot foundation models” are merely “disguised” vision language models that lack “true physical common sense” because they rely heavily on Internet-scale prior training rather than physics-based simulations and real robotics data.

It’s a very sharp philosophical dichotomy. Skild AI is betting that commercial deployment creates a data flywheel that improves the model with each real-world use case. Physical Intelligence is betting that resisting the pull of marketing in the near term will enable it to produce superior general intelligence. Who is “correct” will take years to solve.

Meanwhile, physical intelligence operates with what Groom describes as extraordinary clarity. “It’s pure business. A researcher has a need, we go and collect data to support that need — or new devices or whatever — and then we do it. It’s not externally driven.” The company had a 5-10 year roadmap of what the team believed was possible. By the 18th month, he says, it was over.

The company has about 80 employees and plans to grow, although Groom says he hopes it will be “as slowly as possible.” The most challenging thing, he says, is the hardware. “Hardware is really hard. Everything we do is much harder than a software company.” Hardware breaks. Arrives slowly, delays tests. Safety considerations complicate everything.

As the groom rushes off to his next commitment, I watch the robots continue their training. The pants are still not completely tucked in. The shirt stubbornly stays right side out. Zucchini mulch accumulates well.

There are obvious questions, including my own, about whether anyone really wants a robot in their kitchen peeling vegetables, about safety, about dogs going crazy over mechanical intruders in their homes, and about whether all the time and money invested here is solving big enough problems or creating new ones. Meanwhile, outsiders wonder how far the company has come, whether its vision is achievable, and whether betting on general intelligence rather than specific applications makes sense.

If the groom has any doubts, he does not show them. He’s working with people who have been working on this problem for decades and believe the timing is finally right, and that’s all he needs to know.

Moreover, Silicon Valley has been supportive of people like Groome and giving them a lot of support since the beginning of the industry, knowing that there’s a good chance that even without a clear path to market, even without a timeline, even without certainty about what the market will look like when they get there, they’ll figure it out. It doesn’t always work out, but when it does work, it tends to excuse a lot of the times it doesn’t work out.