Physical Intelligence, a two-year-old San Francisco startup founded by AI researchers and former Google DeepMind staff, is in talks to raise roughly $1 billion in new funding at a valuation exceeding $11 billion. If the deal closes at those terms, it will have doubled its valuation in four months. Its previous round, a $600 million raise in November 2025, valued the company at $5.6 billion. The round before that, in late 2024, put it at around $2.4 billion. At this rate, the company is compounding its paper value faster than it can build robots.
That is not a criticism. It is a measure of how dramatically sentiment around physical AI has shifted. For years, robotics was the unglamorous cousin of the software AI boom: capital-intensive, slow, full of hardware failure modes that made investors nervous. The robots could not generalise. Each new environment required fresh training, bespoke software, and months of calibration. Investors had been burned enough times to be cautious.
What changed is the same thing that changed everything else: foundation models. Physical Intelligence's core research bet is that the same techniques underpinning large language models, trained on vast amounts of demonstration data, can produce robots that generalise across tasks and environments without needing to be programmed for each one. In principle, a model trained on enough footage of human hands performing tasks could transfer that knowledge to a robotic gripper attempting tasks it has never seen before. The early results suggest this is not just theoretical.
The investor base around the company reflects exactly how seriously the big players are taking this. Amazon founder Jeff Bezos was involved in earlier rounds. OpenAI put money in. Alphabet's CapitalG has backed it. These are not casual angel cheques; they are strategic bets from organisations that understand what general-purpose robot intelligence would mean for logistics, manufacturing, and warehousing at scale.
The timing of the Bloomberg report is notable. This week also saw CNBC reporting on Google's new partnership with Agile Robots, another humanoid robotics company, via DeepMind. Google is not simply investing in the space; it is actively providing its AI research capability as infrastructure for robot manufacturers to build on. The pattern across the industry is consistent: the largest AI labs are positioning themselves as the software layer for a coming generation of physical machines, much as they became the intelligence layer for software applications in the past two years.
What is less clear is whether the underlying robotics technology is keeping pace with the valuations. Physical intelligence, the actual capability, is genuinely hard. Robots still struggle with basic physical reasoning tasks that a three-year-old handles without effort: understanding object permanence, predicting how materials will deform, or recovering gracefully when something unexpected happens. The gap between impressive demo video and reliable warehouse deployment remains substantial.
Investors are, in effect, betting that this gap will close on a timeline compatible with their fund structures, and that Physical Intelligence specifically will be among the winners when it does. That may well be right. But $11 billion is a significant multiple on a company that has not yet shipped a commercial product at scale. The valuation is a claim about the future, not a measure of the present. In the current climate, that is not unusual. It is, however, worth keeping in mind the next time a demo video goes viral.
The deeper question the robotics funding wave raises is not whether humanoid robots will eventually be useful. They almost certainly will be. The question is what happens to the labour market when they are. Physical intelligence as a research programme is fascinating. Physical intelligence as an economic force, applied at scale in logistics and manufacturing, is a different and more complicated story, one that the investment community tends to leave for later.