# Physical Intelligence **Source:** https://geo.sig.ai/brands/physical-intelligence **Vertical:** Robotics **Subcategory:** AI Foundation Models for Robotics **Tier:** Emerging **Website:** physicalintelligence.company **Last Updated:** 2026-04-14 ## Summary Foundation model for general-purpose robot control. $5.6B valuation, reportedly raising $1B more at $11B+. Multi-Scale Embodied Memory. Founded 2024, SF. ## Company Overview Physical Intelligence (pi) is a San Francisco-based robotics AI company founded in 2024 to build foundation models for general-purpose robot control. The company was founded by a team of leading robotics and AI researchers including Sergey Levine, Chelsea Finn, Karol Hausman, and Brian Ichter, who previously held senior positions at Google DeepMind and leading academic robotics labs. Physical Intelligence's mission is to develop a single AI model capable of controlling any robot hardware to perform any physical task — the robotics analog of what GPT-4 did for language understanding.\n\npi's core product is a general-purpose robot foundation model trained on diverse robot demonstrations across tasks, hardware types, and environments. The model supports Multi-Scale Embodied Memory (MSEM), an architecture designed to give robots persistent task memory across long-horizon manipulation sequences. Rather than training separate models for each robot type or task, pi's approach generalizes across hardware configurations, enabling faster deployment across the heterogeneous robot ecosystem that exists in warehouses, manufacturing facilities, and research labs.\n\nPhysical Intelligence achieved a $5.6B valuation shortly after its founding and is reportedly raising an additional round at a valuation exceeding $11B — a remarkable trajectory for a company less than two years old. The company has received backing from prominent investors including Sequoia and Jeff Bezos. pi represents the bet that physical world AI — controlling robots in the real world — is the next frontier after language and image models, and that a foundation model approach will ultimately outperform the task-specific robot programming paradigm that has dominated the industry. ## Frequently Asked Questions ### What does Physical Intelligence build? General-purpose AI foundation models controlling any robot for any task. ### What is the valuation? $5.6B, reportedly raising $1B more at $11B+. ### What are key technical advances? RL Token for fast RL and Multi-Scale Embodied Memory for complex multi-step tasks. ### Who founded it? AI and robotics researchers in SF, 2024. Backed by Thrive, Lux, Founders Fund, Lightspeed. ### What hardware does Physical Intelligence's models run on? Physical Intelligence's AI foundation models are designed to be hardware-agnostic — running on various robot platforms including manipulator arms, humanoid robots, and mobile manipulation systems. The company's stated goal is a single AI model that controls any robot for any task, similar to how GPT-4 runs on different inference hardware. Partnerships with robot manufacturers provide the hardware platforms for testing and validating the models. ### What commercial applications are Physical Intelligence's models being deployed for? Physical Intelligence is in early commercial deployment with select enterprise partners in manufacturing, logistics, and service robotics. The company prioritizes demonstrating generalization — showing that one model trained on diverse manipulation data handles novel tasks in new environments — over deploying specialized solutions for specific tasks. Early commercial work validates the foundation model approach before the technology reaches broad commercial licensing. ### How does Physical Intelligence's funding compare to other robotics AI companies? Physical Intelligence's $5.6B valuation and over $900M raised make it the most highly valued dedicated robotics AI company globally as of 2026. This exceeds the valuation of most humanoid robot hardware companies despite Pi being purely a software AI company. The premium reflects investor belief that the AI intelligence layer — not the hardware — will capture the majority of value in the robotics revolution, analogous to how iOS and Android captured more value than mobile hardware manufacturers. ### Who are Physical Intelligence's founders and key researchers? Physical Intelligence was co-founded by Sergey Levine (UC Berkeley robotics research professor), Karol Hausman (Google DeepMind research lead), Chelsea Finn (Stanford AI professor), and Brian Ichter (Google Brain researcher). This founding team represents the top tier of academic robotics AI research — all four have published defining papers in robot learning and manipulation. The team assembled at pi.ai represents an unusual concentration of frontier robotics research talent. ## Tags automation, hardware, manufacturing, b2b --- *Data from geo.sig.ai Brand Intelligence Database. Updated 2026-04-14.*