Physical Intelligence announced a $600M Series B led by CapitalG (Alphabet's growth fund) and Lux Capital, with participation from Bond, Redpoint, Sequoia, Jeff Bezos, and OpenAI. The round values the 2024-founded company at $5.6B — one of the highest valuations in the robotics foundation model space.
The funding will expand data collection, strategic partnerships, and hiring as the company scales its vision-language-action (VLA) model. Version 0.6, released this month, uses the RECAP (RL with Experience & Corrections via Advantage-conditioned Policies) approach to train robots through demonstration, human corrections, and autonomous learning. Physical Intelligence claims this doubled throughput on tasks like espresso machine assembly and reduced failure rates compared to imitation learning alone.
The technical architecture runs 3B–5B parameter transformer models that process RGB-D camera streams and movement history, predicting the next 50 steps in ~100ms. Users issue plain-language commands like "make a flat white" or "pack chocolates into this box." The model outputs robot-agnostic tokens that a hardware abstraction layer converts into joint-specific commands within force and speed limits.
Physical Intelligence is one of several well-funded entrants racing to build generalist robot foundation models. Dyna Robotics raised $120M in September, Archetype AI raised $35M for "physical agents," and Skild AI is developing its Skild Brain. AgiBot has deployed real-world RL systems in manufacturing pilots, while 1X Technologies is using teleoperation to train its NEO humanoid. The market is moving quickly — and consolidation risk is high as early technical leads compound.


