Physical Intelligence Burns $1B Teaching Robots to Fold Laundry
What happens when you give a Stripe veteran $1 billion and tell him to build robot brains with no timeline for making money?
You get Physical Intelligence (π), the most peculiar bet Silicon Valley has made in years. Co-founded by Lachy Groom, this 80-person team is teaching cheap robotic arms to fold laundry, turn shirts inside out, and peel zucchini. Their hardware costs around $3,500 per unit—with materials under $1,000 when built in-house.
The kicker? They're valued at $5.6 billion without a single dollar in revenue.
The ChatGPT Moment for Hardware
While competitors like Skild AI are already reporting $30M in revenue from warehouse deployments, Physical Intelligence is playing a different game entirely. Co-founder Sergey Levine calls it "ChatGPT but for robots"—foundation models that can transfer knowledge across different robot platforms without starting from scratch.
<> "We're working with people who've been working on this problem for decades and who believe the timing is finally right," Groom told investors, refusing to provide any monetization timeline despite raising over $1 billion./>
Their secret sauce is cross-embodiment learning. Train the model on diverse robot types and tasks, then deploy it anywhere. Need a robot for logistics? Same brain. Want one for grocery automation? Same brain. It's an ambitious bet on software intelligence compensating for cheap hardware.
The Numbers Don't Lie (Or Do They?)
Physical Intelligence's progress has been staggering:
- 18 months to achieve what their roadmap projected for 5-10 years
- Data collected from warehouses, homes, test kitchens, and even their office espresso machine
- Partnerships testing systems with logistics companies, grocery chains, and a local chocolate maker
- Majority of their budget allocated to compute power, not flashy hardware
Their first model, π0, combines multi-task and multi-robot data with a new network architecture designed for dexterous manipulation. The demo videos show robots performing tasks that would have required months of custom engineering just years ago.
The Skild Smackdown
Not everyone is buying the hype. Skild AI, Physical Intelligence's Pittsburgh-based rival, has been throwing shade publicly. With their own $1.4B raise at a $14B valuation, Skild claims Physical Intelligence relies too heavily on "disguised vision-language models" that lack true physical common sense.
Skild's criticism cuts deep: these models are just internet-trained systems pretending to understand physics, rather than learning from real simulation and physical data. While Physical Intelligence demos laundry folding, Skild is deploying robots in security, warehouses, and manufacturing—generating actual revenue.
Hot Take
This is either the smartest long-term play in robotics or the most expensive tech demo ever built.
Groom's refusal to discuss monetization isn't refreshing transparency—it's a red flag. Sure, Sequoia and Bezos are backing them, but $5.6B for a company that treats revenue discussions like state secrets?
The technical approach is sound. Cross-embodiment learning could genuinely commoditize robot intelligence. But betting $1B on "pure research" while competitors ship products feels dangerously academic.
Physical Intelligence has 18 months to prove their foundation model approach scales beyond YouTube-worthy demos. Otherwise, they'll become Silicon Valley's most expensive reminder that even the smartest CTOs need customers, not just cute robots making espresso.
