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

From the street, the only indication that I’ve found Physical Intelligence’s San Francisco headquarters is a pi symbol that’s a slightly different color than the rest of the door. When I walk in, I am immediately confronted with activity. There is no reception, no shiny logo in fluorescent lighting.

Inside, the space is a gigantic concrete box made slightly less austere by a random series of long, blond-wood tables. Some are clearly meant for lunch, littered with Girl Scout cookie boxes, jars of Vegemite (someone here is Australian) and little wire baskets filled with a little too many spices. The rest of the tables tell a completely different story. Many more of them are laden with monitors, spare robotics parts, tangles of black wire, and fully assembled robotic arms in various states of trying to master the mundane.

During my visit, one arm folds, or tries to, a pair of black pants. Things aren’t going well. Another tries to turn a shirt inside out with the kind of determination that suggests it will work eventually, but not today. A third – this one seems to have found his calling – quickly peels a zucchini, after which he has to deposit the scraps in a separate container. In any case, the chips are going well.

“Think of it as ChatGPT, but for robots,” says Sergey Levine, gesturing to the motorized ballet unfolding across the room. Levine, an associate professor at UC Berkeley and one of the co-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 looking at, he explains, is the testing phase of a continuous loop: data is collected at robotic stations here and at other locations (warehouses, homes, wherever the team may be based) and that data trains general-purpose robotic foundation models. When researchers train a new model, it comes back to these types of stations for evaluation. The trouser folder is someone’s experiment. This also applies to the shirt turner. The zucchini peeler could test whether the model can be generalized to different vegetables, learning the basic movements of peeling well enough to handle an apple or a potato he has never encountered before.

The company also operates a test kitchen in this building and elsewhere, using off-the-shelf hardware to expose the robots to different environments and challenges. There’s a high-end espresso machine nearby, and I assume it’s for the staff, until Levine makes it clear that no, the robots have to learn. All the foam lattes are data and not a perk for the dozens of engineers on site who are usually peering into their computers or hovering above their mechanized experiments.

The hardware itself is deliberately unglamorous. These guns sell for about $3,500, which is what Levine describes as “a huge profit margin” for the seller. If they manufactured them in-house, the material cost would drop below $1,000. A few years ago, he says, a roboticist 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 Levine apologizes, I am approached by Lachy Groom, who moves through the space with the purposefulness of someone making half a dozen things happen at once. At 31, Groom still has the fresh-faced look of Silicon Valley’s boy wonder, a title he earned early on, having sold his first company nine months after founding it in his native Australia at age 13 (this explains the Vegemite).

When I first approached him earlier, as he welcomed a small group of sweatshirt-wearing visitors into the building, he immediately responded to my request to spend time with him: “Absolutely not, I have meetings.” Now he has maybe ten minutes.

Groom found what he was looking for when he started following the academic work coming out of the labs of Levine and Chelsea Finn, a former Berkeley PhD student of Levine’s who now runs her own lab at Stanford focused on robot learning. Their names continued to appear in everything interesting in the field of robotics. When he heard rumors that they might be starting something, he tracked down Karol Hausman, a Google DeepMind researcher who also taught at Stanford and whom Groom had learned was involved. “It was just one of those meetings where you walk out and think, this is it.”

Groom never intended to become a full-time investor, he says, although some may wonder why he didn’t, given his track record. After leaving Stripe, where he was an early employee, he spent about five years as an angel investor, making early bets on companies like Figma, Notion, Ramp, and Lattice as he looked for the right company to start or join. His first investment in robotics, Standard Bots, came in 2021 and reintroduced him to a field he loved as a child building Lego Mindstorms. As he jokes, he was “on vacation much more as an investor.” But investing was just a way to stay active and meet people, not the end game. “I’ve been looking forward to it for five years for the company to start after Stripe,” he says. “Good ideas at a good time with a good team — [that’s] extremely rare. It’s all execution, but you can execute a bad idea like crazy, and it’s still a bad idea.

The two-year-old company has now implemented an increase over $1 billionand when I ask about the runway, he quickly makes it clear that it doesn’t actually burn that much. Most of the spending goes on computers. A little later he acknowledges that he would bring in more under the right conditions and with the right partners. “There’s no limit to how much money we can really put to work,” he says. “There is always more computing power you can apply to the problem.”

What makes this arrangement particularly unusual is what Groom doesn’t give his supporters: a timeline for turning physical intelligence into a money-making venture. “I don’t give investors answers about commercialization,” he says of the likes of Khosla Ventures, Sequoia Capital and Thrive Capital, which have valued the company at $5.6 billion. “That’s quite strange that people tolerate that.” But they do tolerate it, and that may not always be the case. Therefore, the company must now be well capitalized.

So what is the strategy, if not commercialization? Quan Vuong, another co-founder of Google DeepMind, explains that it’s all about cross-embodiment learning and diverse data sources. If someone builds a new hardware platform tomorrow, they won’t have to start collecting data from scratch; he can transfer all the knowledge that the model already has. “The marginal cost of onboarding autonomy to a new robot platform, whatever that platform may be, is just a lot lower,” he says.

The company is already working with a small number of companies in different industries – logistics, supermarkets, a chocolatier across the street – to test whether their systems are good enough for true automation. Vuong claims that this is already the case in some cases. With their “any platform, any job” approach, the window for success is large enough that you can start checking off tasks ready for automation today.

Physical Intelligence is not alone in pursuing this vision. The race to build general-purpose robot intelligence – the foundation on which more specialized applications can be built, much like the LLM models that captivated the world three years ago – is heating up. Founded in 2023, Pittsburgh-based Skild AI raised $1.4 billion this month at a $14 billion valuation and is taking a noticeably different approach. While Physical Intelligence continues to focus on pure research, Skild AI has already deployed its ‘omni-body’ Skild Brain commercially, saying it generated $30 million in revenue in just a few months last year across security, warehouses and manufacturing.

Skild has even publicly shot at competitors, argument on his blog that most “basic robotics models” are just “disguised vision language models” that lack “real physical common sense” because they rely too heavily on internet-scale prior training rather than physics-based simulation and real robotics data.

It’s a pretty sharp philosophical divide. Skild AI is betting that commercial deployment creates a data flywheel that improves the model with every real-world situation. Physical Intelligence is betting that resisting the pull of commercialization in the short term will allow it to produce superior general intelligence. Whoever is ‘more right’ will need years to find a solution.

Meanwhile, Physical Intelligence operates with what Groom describes as unusual clarity. “It’s such a pure business. A researcher has a need, we’re going to collect data to support that need – or new hardware or whatever it is – and then we do it. It’s not externally driven.” The company had a five- to 10-year roadmap of what the team thought would be possible. By month 18, they had blown through it, he says.

The company has about 80 employees and plans to grow, although Groom says hopefully “as slowly as possible.” What’s most challenging, he says, is the hardware. “Hardware is just really hard. Everything we do is so much harder than a software company.” Hardware breaks. It arrives slowly, delaying tests. Safety concerns complicate everything.

As Groom gets up to rush to his next appointment, I watch the robots continue their practice. The pants are still not completely folded. The shirt stubbornly remains right side out. The zucchini shavings pile up nicely.

There are obvious questions, including mine, about whether anyone actually wants a robot in their kitchen that peels vegetables, about safety, about dogs going crazy about mechanical intruders in their homes, about whether all the time and money invested here is solving enough problems or creating new ones. Meanwhile, outsiders question the company’s progress, whether its vision is feasible and whether it makes sense to bet on general intelligence rather than specific applications.

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

Furthermore, Silicon Valley has backed people like Groom and given them a lot of rope since the industry’s inception, knowing that even without a clear path to commercialization, even without a timeline, even without certainty about what the market will look like when they get there, there’s a good chance they’ll figure it out. It doesn’t always work. But when it does, it often justifies the times when it didn’t.