Physical Intelligence: Inside the $5.6B Robotics Startup Building 'ChatGPT for Robots'

In a nondescript San Francisco building marked only by a subtle pi symbol on the door, one of Silicon Valley's most ambitious bets on the future of robotics is taking shape. Physical Intelligence, the $5.6 billion startup backed by Stripe veteran Lachy Groom, is racing to build what they call "ChatGPT for robots"—general-purpose AI that can make any robot do almost anything.

Inside Physical Intelligence's Robot Laboratory

Walking into Physical Intelligence's headquarters, there's no gleaming reception desk or fluorescent logo. Instead, you'll find a giant concrete box filled with long blonde-wood tables, some scattered with Girl Scout cookies and jars of Vegemite (a nod to Groom's Australian roots), while others are laden with monitors, spare robotics parts, and—most importantly—robotic arms attempting to master everyday tasks.

The Robots in Action

During a recent TechCrunch visit, the scene included:

• The pants folder: A robotic arm struggling but determined to fold black pants
• The shirt turner: An arm attempting to turn a shirt inside out with notable persistence
• The zucchini peeler: Perhaps the star performer, quickly peeling vegetables and depositing shavings into a container

There's even a sophisticated espresso machine—but it's not for the staff. It's there for the robots to learn from.

"Think of it Like ChatGPT, But for Robots"

That's how co-founder Sergey Levine, an associate professor at UC Berkeley, describes what Physical Intelligence is building. The company is developing general-purpose robotic foundation models that can be trained on diverse data and then applied across different robots and tasks.

The Training Loop

Physical Intelligence operates on a continuous cycle:

1. Data Collection: Robots at stations in San Francisco, warehouses, homes, and other locations gather experience
2. Model Training: Data trains general-purpose robotic foundation models
3. Evaluation: New models return to test stations for evaluation
4. Iteration: Results feed back into the next training round

The Founders: Academic Stars and Stripe Talent

Physical Intelligence brings together some of the biggest names in robotics AI:

Sergey Levine

• Associate Professor at UC Berkeley
• Pioneer in robotic learning research
• Known for explaining complex robotics concepts

Chelsea Finn

• Former Berkeley PhD student under Levine
• Now runs her own robotics learning lab at Stanford
• Her name appears in "everything interesting happening in robotics"

Karol Hausman

• Former Google DeepMind researcher
• Also taught at Stanford
• Brought key AI research expertise

Quan Vuong

• Also from Google DeepMind
• Co-founder focused on cross-embodiment learning

Lachy Groom

• Former Stripe early employee
• Sold first company at age 13 (!) in Australia
• Angel investor in Figma, Notion, Ramp, Lattice
• Spent five years searching for "the right company" after leaving Stripe

The $1 Billion+ War Chest

Physical Intelligence has raised over $1 billion at a $5.6 billion valuation, with backing from Silicon Valley's most prestigious firms:

Where the Money Goes

Groom is quick to point out the company doesn't actually burn that much money—most spending goes toward compute resources. As he told TechCrunch: "There's no limit to how much money we can really put to work. There's always more compute you can throw at the problem."

The Unusual Business Strategy: No Commercialization Timeline

What makes Physical Intelligence particularly unusual is what Groom doesn't give investors: a timeline for making money.

"I don't give investors answers on commercialization. That's sort of a weird thing, that people tolerate that."

Yet investors not only tolerate it—they're pouring in billions. The bet is that getting the fundamental technology right is more important than rushing to market.

The Philosophy

Physical Intelligence is betting that resisting near-term commercialization will enable them to produce superior general intelligence. It's a patient, research-first approach that prioritizes scientific breakthroughs over quarterly revenue.

Cross-Embodiment: The Key Innovation

What sets Physical Intelligence apart is their focus on cross-embodiment learning. Co-founder Quan Vuong explains the concept:

If someone builds a new robot hardware platform tomorrow, they won't need to start data collection from scratch. They can transfer all the knowledge the model already has to the new platform.

Why This Matters

• Lower marginal costs: Onboarding autonomy to new robot platforms becomes cheaper
• Any platform, any task: The same AI can work across different robot types
• Faster deployment: No need to train from scratch for each new robot

The Hardware Philosophy: Cheap and Unglamorous

The robotic arms at Physical Intelligence cost about $3,500—and even that includes "an enormous markup" from the vendor. In-house manufacturing would drop material costs below $1,000.

As Levine notes, a few years ago, roboticists would have been shocked these cheap arms could do anything useful at all. But that's precisely the point: good intelligence compensates for bad hardware.

Early Commercial Testing

Despite the research-first approach, Physical Intelligence is already working with a small number of companies:

• Logistics companies: Testing warehouse automation
• Grocery operations: Exploring retail applications
• Local businesses: Including a chocolate maker across the street from their HQ

Vuong claims that in some cases, their systems are already good enough for real-world automation.

The Competition: Skild AI's $14 Billion Challenge

Physical Intelligence isn't alone in the race for robot brains. Pittsburgh-based Skild AI just raised $1.4 billion at a $14 billion valuation—nearly triple Physical Intelligence's current valuation.

Different Approaches

The Philosophical Divide

Skild has taken public shots at competitors, arguing that most "robotics foundation models" are just vision-language models "in disguise" that lack "true physical common sense."

Physical Intelligence counters that rushing to commercialize creates inferior technology. Who's right will take years to determine.

The 80-Person Team

Physical Intelligence has about 80 employees and plans to grow—though Groom says hopefully "as slowly as possible." The biggest challenge? 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 considerations complicate everything.

Blowing Past the Roadmap

In perhaps the most telling detail, the company had a 5- to 10-year roadmap when they started. By month 18, they'd blown through it entirely.

This accelerated progress is what has investors excited despite the lack of commercialization timeline. When fundamental research advances faster than expected, the money tends to follow.

What Physical Intelligence's Robots Are Learning

The current focus areas include:

Household Tasks

• Folding clothes
• Food preparation (peeling, cutting)
• General manipulation

Industrial Applications

• Warehouse logistics
• Manufacturing assembly
• Package handling

Service Tasks

• Food service (hence the espresso machine)
• Retail operations
• Delivery assistance

The Open Questions

Several important questions remain:

1. Does anyone want robots peeling vegetables at home? Consumer applications remain uncertain
2. Safety concerns: Robots in homes with pets and children present challenges
3. Economic viability: Can general-purpose robots compete with specialized automation?
4. Timeline to deployment: Without a commercialization schedule, when will we see results?

Why Silicon Valley Believes

Groom's track record helps explain investor confidence:

• Early Stripe employee: Part of one of fintech's biggest success stories
• Successful angel investor: Early bets on Figma, Notion, Ramp all paid off
• Pattern recognition: "Good ideas at a good time with a good team — that's extremely rare"

As Groom told TechCrunch, he spent five years after Stripe looking for the right opportunity. Physical Intelligence was it.

The Founder's Conviction

Groom shows no doubts about the mission. He's working with researchers who've been working on robotics for decades and who believe the timing is finally right.

Silicon Valley has been backing people like this since the beginning of the industry—giving them runway even without clear paths to commercialization. It doesn't always work out. But when it does, it justifies a lot of the times it didn't.

What This Means for the Future of Robotics

If Physical Intelligence succeeds, the implications are enormous:

For Consumers

• Robots that can help with household tasks
• More capable home automation
• Potential for elderly care assistance

For Industry

• General-purpose robots replacing specialized machines
• Faster deployment of automation
• Lower costs for robotics implementation

For AI

• Proof that foundation model approaches work for physical tasks
• New paradigms for robot learning
• Cross-embodiment as a key capability

The Bottom Line

Physical Intelligence represents one of Silicon Valley's biggest bets on the future of robotics. With over $1 billion raised, world-class founders, and a research-first approach that prioritizes getting the technology right over rushing to market, they're attempting something ambitious: building the AI that will power the next generation of robots.

Whether the pants-folding arm ever masters its task—or whether the zucchini peeler graduates to more complex vegetables—the company's progress over the next few years will help determine whether general-purpose robot brains become reality.

For now, in that unassuming San Francisco building, the robots keep practicing. And the world keeps watching.

Stay tuned to PromptGalaxy for the latest news on AI robotics and breakthrough startups.