The $11 Billion Bet on a Robot That Can Learn Anything
For decades, industrial robots have been extraordinarily good at doing one thing. A robotic arm on a car assembly line can weld the same joint ten thousand times without complaint — but ask it to pack a box or fold a shirt and it is utterly lost. That rigidity has been robotics' defining constraint, and it is precisely the constraint that San Francisco-based startup Physical Intelligence is trying to demolish.
The company is reportedly in advanced talks to raise $1 billion in new funding, a round that would more than double its valuation from $5.6 billion to over $11 billion — a figure assessed just four months ago. The speed of that re-rating is a measure of how seriously investors are taking the company's proposition: that general-purpose AI models can power robots to learn and adapt to almost any physical task, the way large language models reshaped what software can do with text.
If Physical Intelligence is right, the implications stretch far beyond factory floors. Logistics, healthcare, manufacturing, and domestic services could all be reshaped by machines that do not need to be reprogrammed for every new job.
The Details
Physical Intelligence was founded by AI academics and former Google DeepMind researchers — a pedigree that signals this is foundational research, not a product sprint. The company's co-founder Sergey Levine has described the ambition plainly: they are building "ChatGPT, but for robots." The analogy is instructive. ChatGPT did not need to be trained separately for every language task; it developed broad, transferable capabilities. Physical Intelligence wants robots to work the same way — learning from experience and generalising to new situations rather than being locked into a fixed repertoire.
"ChatGPT, but for robots."— Sergey Levine, Co-founder, Physical Intelligence
Realising that vision requires enormous computational resources. Training AI models capable of controlling physical systems in the real world — with all its unpredictability, texture, and consequence — is vastly more demanding than text prediction. That is the core reason behind Physical Intelligence's appetite for capital at this scale: the company needs infrastructure commensurate with its foundational ambitions.
The funding round, if completed, would cement Physical Intelligence as one of the most richly valued robotics companies in the world, and would mark a striking acceleration from its last valuation set only months ago. That trajectory reflects a broader shift in investor sentiment: the question is no longer whether general-purpose AI can be applied to robotics, but who will get there first and own the platform that powers the next generation of machines.
The Bigger Picture
Physical Intelligence is not operating in isolation. The broader robotics industry is undergoing a structural shift, moving away from highly specialized machines toward AI-driven systems capable of flexible, adaptive behavior. What Physical Intelligence is attempting — and what the reported funding round suggests investors believe — is that the same architectural leap that made large language models transformative can be replicated in the physical world.
The stakes are considerable. A robust general-purpose AI model for robotics would not just be a product; it would be infrastructure — the kind of foundational layer that other companies build on top of. That is the scale of ambition reflected in an $11 billion valuation for a company still in foundational R&D, and the reason this funding round is worth watching closely.
