From the road, the one indication I’ve discovered Physical Intelligence’s headquarters in San Francisco is a pi image that’s a barely totally different coloration than the remainder of the door. When I stroll in, I’m instantly confronted with exercise. There’s no reception desk, no gleaming emblem in fluorescent lights.
Inside, the area is an enormous concrete field made barely much less austere by a haphazard sprawl of lengthy blonde-wood tables. Some are clearly meant for lunch, dotted with Girl Scout cookie packing containers, jars of Vegemite (somebody right here is Australian), and small wire baskets full of one too many condiments. The remainder of the tables inform a distinct story fully. Many extra of them are laden with screens, spare robotics components, tangles of black wire, and absolutely assembled robotic arms in varied states of making an attempt to grasp the mundane.
During my go to, one arm is folding a pair of black pants, or making an attempt to. It’s not going nicely. Another is making an attempt to show a shirt inside out with the form of dedication that means it can finally succeed, simply not in the present day. A 3rd — this one appears to have discovered its calling — is shortly peeling a zucchini, after which it is alleged to deposit the shavings right into a separate container. The shavings are going nicely, at the least.
“Think of it like ChatGPT, but for robots,” Sergey Levine tells me, gesturing towards the motorized ballet unfolding throughout the room. Levine, an affiliate professor at UC Berkeley and one in all Physical Intelligence’s co-founders, has the amiable, bespectacled demeanor of somebody who has spent appreciable time explaining complicated ideas to individuals who don’t instantly grasp them.
Image Credits:Connie Loizos for TechCrunch
What I’m watching, he explains, is the testing section of a steady loop: information will get collected on robot stations right here and at different areas — warehouses, houses, wherever the group can arrange store — and that information trains general-purpose robotic basis fashions. When researchers practice a brand new mannequin, it comes again to stations like these for analysis. The pants-folder is somebody’s experiment. So is the shirt-turner. The zucchini-peeler is perhaps testing whether or not the mannequin can generalize throughout totally different greens, studying the basic motions of peeling nicely sufficient to deal with an apple or a potato it’s by no means encountered.
The firm additionally operates a take a look at kitchen on this building and elsewhere utilizing off-the-shelf {hardware} to show the robots to totally different environments and challenges. There’s a complicated espresso machine close by, and I assume it’s for the workers till Levine clarifies that no, it’s there for the robots to study. Any foamed lattes are information, not a perk for the handfuls of engineers on the scene who’re principally peering into their computer systems or hovering over their mechanized experiments.
The {hardware} itself is intentionally unglamorous. These arms promote for about $3,500, and that’s with what Levine describes as “an enormous markup” from the seller. If they manufactured them in-house, the fabric price would drop beneath $1,000. A couple of years in the past, he says, a roboticist would have been shocked these items may do something in any respect. But that’s the purpose — good intelligence compensates for unhealthy {hardware}.
Techcrunch occasion
Boston, MA
|
June 23, 2026
As Levine excuses himself, I’m approached by Lachy Groom, transferring via the area with the purposefulness of somebody who has half a dozen issues occurring directly. At 31, Groom nonetheless has the fresh-faced high quality of Silicon Valley’s boy surprise, a designation he earned early, having offered his first firm 9 months after beginning it at age 13 in his native Australia (this explains the Vegemite).
When I first approached him earlier, as he welcomed a small gaggle of sweatshirt-wearing guests into the building, his response to my request for time with him was quick: “Absolutely not, I’ve got meetings.” Now he has 10 minutes, possibly.
Groom discovered what he was on the lookout for when he began following the educational work popping out of the labs of Levine and Chelsea Finn, a former Berkeley PhD pupil of Levine’s who now runs her personal lab at Stanford targeted on robotic studying. Their names stored showing in every part attention-grabbing occurring in robotics. When he heard rumors they is perhaps beginning one thing, he tracked down Karol Hausman, a Google DeepMind researcher who additionally taught at Stanford and who Groom had realized was concerned. “It was just one of those meetings where you walk out and it’s like, This is it.”
Groom by no means supposed to develop into a full-time investor, he tells me, regardless that some may surprise why not given his monitor document. After leaving Stripe, the place he was an early worker, he spent roughly 5 years as an angel investor, making early bets on firms like Figma, Notion, Ramp, and Lattice whereas looking for the proper firm to begin or be a part of himself. His first robotics funding, Standard Bots, got here in 2021 and reintroduced him to a discipline he’d cherished as a child building Lego Mindstorms. As he jokes, he was “on vacation much more as an investor.” But investing was only a method to keep energetic and meet individuals, not the endgame. “I was looking for five years for the company to go start post-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 like hell on a bad idea, and it’s still a bad idea.”
Image Credits:Connie Loizos for TechCrunch
The two-year-old firm has now raised over $1 billion, and once I ask about its runway, he’s fast to make clear it doesn’t truly burn that a lot. Most of its spending goes towards compute. A second later, he acknowledges that underneath the proper phrases, with the proper companions, he’d elevate extra. “There’s no limit to how much money we can really put to work,” he says. “There’s always more compute you can throw at the problem.”
What makes this association significantly uncommon is what Groom doesn’t give his backers: a timeline for turning Physical Intelligence right into a money-making endeavor. “I don’t give investors answers on commercialization,” he says of backers that embody Khosla Ventures, Sequoia Capital, and Thrive Capital amongst others which have valued the corporate at $5.6 billion. “That’s sort of a weird thing, that people tolerate that.” But tolerate it they do, and so they could not at all times, which is why it behooves the corporate to be well-capitalized now.
So what’s the technique, if not commercialization? Quan Vuong, one other co-founder who got here from Google DeepMind, explains that it revolves round cross-embodiment studying and numerous information sources. If somebody builds a brand new {hardware} platform tomorrow, they gained’t want to begin information assortment from scratch — they’ll switch all of the information the mannequin already has. “The marginal cost of onboarding autonomy to a new robot platform, whatever that platform might be, it’s just a lot lower,” he says.
The firm is already working with a small variety of firms in several verticals — logistics, grocery, a chocolate maker throughout the road — to check whether or not their methods are adequate for real-world automation. Vuong claims that in some circumstances, they already are. With their “any platform, any task” method, the floor space for fulfillment is giant sufficient to begin checking off duties which can be prepared for automation in the present day.
Physical Intelligence isn’t alone in chasing this imaginative and prescient. The race to construct general-purpose robotic intelligence — the inspiration on which extra specialised functions might be constructed, very like the LLM fashions that captivated the world three years in the past — is heating up. Pittsburgh-based Skild AI, based in 2023, simply this month raised $1.4 billion at a $14 billion valuation and is taking a notably totally different method. While Physical Intelligence stays targeted on pure analysis, Skild AI has already deployed its “omni-bodied” Skild Brain commercially, saying it generated $30 million in income in only a few months final 12 months throughout safety, warehouses, and manufacturing.
Image Credits:Connie Loizos for TechCrunch
Skild has even taken public photographs at rivals, arguing on its weblog that the majority “robotics foundation models” are simply vision-language fashions “in disguise” that lack “true physical common sense” as a result of they rely too closely on internet-scale pretraining fairly than physics-based simulation and actual robotics information.
It’s a fairly sharp philosophical divide. Skild AI is betting that industrial deployment creates a knowledge flywheel that improves the mannequin with every real-world use case. Physical Intelligence is betting that resisting the pull of near-term commercialization will allow it to provide superior common intelligence. Who’s “more right” will take years to resolve.
In the meantime, Physical Intelligence operates with what Groom describes as uncommon readability. “It’s such a pure company. A researcher has a need, we go and collect data to support that need — or new hardware or whatever it is — and then we do it. It’s not externally driven.” The firm had a 5- to 10-year roadmap of what the group thought can be potential. By month 18, they’d blown via it, he says.
The firm has about 80 workers and plans to develop, although Groom says hopefully “as slowly as possible.” What’s probably the most difficult, he says, is {hardware}. “Hardware is just really hard. Everything we do is so much harder than a software company.” Hardware breaks. It arrives slowly, delaying exams. Safety concerns complicate every part.
As Groom springs as much as rush to his subsequent dedication, I’m left watching the robots proceed their observe. The pants are nonetheless not fairly folded. The shirt stays stubbornly right-side-out. The zucchini shavings are piling up properly.
There are apparent questions, together with my very own, about whether or not anybody truly desires a robot of their kitchen peeling greens, about security, about canines going loopy at mechanical intruders of their houses, about whether or not the entire money and time being invested right here solves sufficiently big issues or creates new ones. Meanwhile, outsiders query the corporate’s progress, whether or not its imaginative and prescient is achievable, and if betting on common intelligence fairly than particular functions is sensible.
If Groom has any doubts, he doesn’t present it. He’s working with individuals who’ve been engaged on this downside for many years and who consider the timing is lastly proper, which is all he must know.
Besides, Silicon Valley has been backing individuals like Groom and giving them a variety of rope because the starting of the business, understanding there’s a superb likelihood that even with out a clear path to commercialization, even with out a timeline, even with out certainty about what the market will seem like once they get there, they’ll determine it out. It doesn’t at all times work out. But when it does, it tends to justify a variety of the occasions it didn’t.
