Physical Intelligence

Physical Intelligence (pi) is a San Francisco robotics foundation model company that raised over $1 billion including a $600 million Series B at a $5.6 billion valuation led by Alphabet CapitalG alongside Lux Capital, Bond, Redpoint, and Sequoia Capital. Founded in 2023 by former Google Brain researchers Sergey Levine, Chelsea Finn, and Karol Hausman, the company builds a generalist AI brain for robots rather than robot hardware itself.

Its flagship model pi0 (pi-zero) is a Vision-Language-Action flow model trained on data from diverse robot types and tasks, enabling robots to perform activities they were never explicitly programmed for. Physical Intelligence open-sourced pi0 and the autoregressive pi0-FAST variant under an Apache 2.0 licence through its openpi repository in February 2025, later adding pi0.5 weights and fine-tuning code.

Its pi*0.6 model, published in November 2025, doubled throughput on dexterous manipulation tasks including espresso machine filter insertion, folding previously unseen laundry, and cardboard box assembly. The company hardware-agnostic approach means pi0 runs on existing robots from multiple manufacturers, allowing robot makers to add generalist intelligence without building their own foundation models.

Physical Intelligence is positioned as the foundation model layer of the robotics stack — analogous to what OpenAI is for language AI — with plans to deploy models across robot fleets via SaaS licensing. The RECAP training method behind that result — reinforcement learning with experience and corrections via advantage-conditioned policies — is what separates Physical Intelligence from imitation learning alone: a robot is taught by demonstration, coached through human corrections, and then improves from its own autonomous experience, so failure rates fall over hours of continuous operation instead of staying flat.

Commercially it is the odd one out in this category because it ships no robot. Figure, Apptronik, Agility, 1X and Boston Dynamics each pair their own hardware with their own or a partner's model; Physical Intelligence's addressable market is every robot those companies ship, and its open releases mean a team can benchmark it on hardware it already owns before committing — something none of the hardware vendors here allow.

The company was reported during 2026 to be in talks for a further raise of around $1 billion at a valuation near $11 billion; that round is press-reported rather than company-confirmed and should be treated as such. Best fit: robot OEMs and research groups that want a generalist manipulation policy without funding a foundation-model programme of their own.