Physical Intelligence
Builds π (pi) general-purpose robot foundation models that run across many robot bodies
Scaleup · No humanoid status · Robot foundation model · San Francisco, California, United States · Founded 2024 · physicalintelligence.company
Series C amount and lead come from secondary trackers/filings; no company press release found. Prior round: $600M Series B Nov 2025 at $5.6B post (humanoid.observer). Source dates approximate.
Profile
Founded by robot-learning researchers from Google DeepMind, Stanford and UC Berkeley, Physical Intelligence trains vision-language-action models (π0, π0.5) that control many robot types. It sells models rather than robots. Its valuation rose from $5.6B (Nov 2025) to a reported ~$11B in a May 2026 Series C.
People
- Karol Hausman FounderCo-founder and Chief Executive Officer@hausman_kHandle linked from personal site
As of –. X handles are shown only when we could confirm they belong to the person.
Leadership on X
What leadership has posted about robotics, AI and tech in the last 30 days.
- He congratulated Ultra's Jon Miller Schwartz on OP1, a robot Ultra says is built for scaled deployment, and called Ultra an awesome partner.1
Replies are included only when they mention robots, robotics, humanoids or AI.
- Shared new work that pairs a small reinforcement-learning policy with a large robot foundation model, so the robot can react to newer camera images and steer toward better behavior.1
- Highlighted a hackathon built on Physical Intelligence's π-0.7 model, where teams ran it on unusual machines such as an excavator.2
Replies are included only when they mention robots, robotics, humanoids or AI.
- Shared a video of a Physical Intelligence robot working fully autonomously at a chocolate maker, and lessons on the gap between a one-off demo and hours of useful work.1
- Co-wrote a blog post on what robots need to be broadly useful, calling reliability one of the biggest open challenges in AI.2
- Shared Stanford research on reinforcement learning for fast, dynamic robot tasks, where a small real-time policy corrects a large vision-language-action model.3
- Promoted a new benchmark that tests whether AI models can spot which earlier papers could advance a research problem.4
- Shared Mulligan, a method that focuses human help during deployment on the starting conditions where a robot tends to fail, noting that standard DAgger wastes a person's time once robots succeed most of the time.5
Replies are included only when they mention robots, robotics, humanoids or AI.
Chatter on X
Social posts, not verified facts. What X is saying about this company, rated for how far it can be trusted. Shows the last 14 days, plus anything still unresolved.
Items shown have cleared our internal credibility screen.
- ConfirmedSupply chain / customerFirst seen Oct 9 · Updated Oct 10
Ultra raises $50M Series A and deepens its partnership with Physical Intelligence
Ultra leases AI-controlled warehouse packing robots and uses Physical Intelligence's models. Total funding is $62M.
Why this label- Announced by Ultra; Physical Intelligence's official account congratulated it
- Covered by Fortune and SiliconANGLE
Primary post on X · @physical_int, Oct 9Backing: Fortune·SiliconANGLE
- ConfirmedProduct demoFirst seen Oct 9 · Updated Oct 10
Physical Intelligence co-founder Chelsea Finn shares research on wasted effort in robot data collection
Her post says DAgger-style training becomes inefficient as robots get more reliable, and proposes focusing on failure-prone starting states. It is academic research, not a product announcement.
Why this label- First-hand post by a Physical Intelligence co-founder
Primary post on X · @chelseabfinn, Oct 9Backing: Post on X by @chelseabfinn
Labels
- Confirmed
- The company, an SEC filing or a major outlet backs it.
- Likely
- A credible first-hand source or named outlet reports it; nothing official yet.
- Unverified
- One source, or a source with a thin track record. Treat as a rumor.
- Disputed
- A credible party contests it.
Last checked Oct 10.
Flagship products
- π0 / π0.5 / π* AI model
Vision-language-action robot foundation models; π0 open-sourced
Funding and valuation
Latest round
$1.05B Reported
Series C · May 21, 2026
Led by Founders Fund
Also Lightspeed, Thrive Capital, Lux Capital.
Reported ~$1.0-1.05B; based on reports and filings, not a company press release
Valuation
$11B Reported
reported valuation (Series C) · May 21, 2026
Hiring signals
- Open roles
- 38all roles
- Top location
- San Francisco · 95%36 of 38 roles
- Top role area
- AI & foundation models12 roles · 35% of 34 engineering & ops
Where the roles are · share of 38 open roles (all roles)
- San Francisco, California, US95%36 roles
- Fremont, California, US5%2 roles
Locations are where the posting says the job is based, not where the robots are built. A posting that lists several sites is counted once, under the first.
Role mix · 34 engineering & ops roles
Show 4 more areas
Roles are sorted into a fixed list of areas by job title. Percentages are of engineering & ops roles. The other 4 open roles (finance, HR, legal, sales, facilities and similar) are in the total but not shown here.
Notable postings
- JobSenior role in AI & foundation models, one of the company's largest hiring areasAnnotations Portfolio LeadSan Francisco, California, US
- JobSenior role in Manufacturing engineering, one of the company's largest hiring areasNPI Technical Program ManagerSan Francisco, California, US
- JobTypical role in Actuators & mechanical, one of the company's largest hiring areasPrototype EngineerSan Francisco, California, US
- JobTypical role in Quality & test, one of the company's largest hiring areasSoftware Engineer, Data QualitySan Francisco, California, US
- JobTypical role in Software & infrastructure, one of the company's largest hiring areasSoftware Engineer, Robot InterfacesSan Francisco, California, US
As of . Scope: all roles. Counts are our reading of public job postings, not company statements.
Scope note All open Physical Intelligence postings, including non-engineering roles.
Trademarks
No US filings since 2025 (last checked ).
Source: USPTO Trademark Search, matched on the company’s legal names. Rechecked on the 1st and 15th of each month. A filing shows intent to use a name in the US, not a launched product.
Competitors
Companies with overlapping products. Direct = same customers and same job; Adjacent = overlapping technology or market, different focus.
Skild AI
Direct competitorBoth build general robot AI models that other companies' robots can run: Skild AI builds Skild Brain, a foundation model meant to run on any robot body, while Physical Intelligence builds the general-purpose π robot foundation models.
Rhoda AI
Direct competitorBoth build general robot AI models that other companies' robots can run: Rhoda AI is building a robot brain pre-trained on internet video, while Physical Intelligence builds the general-purpose π robot foundation models.
FieldAI
Direct competitorBoth build general robot AI models that other companies' robots can run: FieldAI builds Field Foundation Models for robots working in unstructured sites, while Physical Intelligence builds the general-purpose π robot foundation models.
Show 5 more
Genesis AI
Direct competitorBoth build general robot AI models that other companies' robots can run: Genesis AI builds a robot-brain model and now its own hand and the Eno robot, while Physical Intelligence builds the general-purpose π robot foundation models.
Google DeepMind (Gemini Robotics)
Direct competitorBoth build general robot AI models that other companies' robots can run: Google DeepMind develops the Gemini Robotics models, used by partners such as Apptronik, while Physical Intelligence builds the general-purpose π robot foundation models.
NVIDIA
Direct competitorBoth build general robot AI models that other companies' robots can run: NVIDIA supplies Jetson Thor compute and the open GR00T humanoid foundation models, while Physical Intelligence builds the general-purpose π robot foundation models.
Sanctuary AI
Adjacent competitorBoth sell robot AI to other makers: Sanctuary AI built the Phoenix humanoid with hydraulic hands and now also licenses its physical-AI software, while Physical Intelligence builds the general-purpose π robot foundation models.
Toyota Research Institute
Adjacent competitorBoth train large behavior models for robots: Toyota Research Institute builds the Large Behavior Models that drive Boston Dynamics' Atlas, while Physical Intelligence builds the general-purpose π robot foundation models.
Recent news
- — Reported ~$1B Series C at ~$11B valuation led by Founders Fund The Robotics Media Reported
- — In talks to raise ~$1B at $11B valuation Tech Funding News Reported
Sources
- Physical Intelligence funding: every round, investor, and valuation (Revenue Memo) (May 21, 2026)
- Physical Intelligence hits $11B valuation in $1B round (The Robotics Media) (May 21, 2026)
- Physical Intelligence eyes $1B raise at $11B valuation (Tech Funding News) (Apr 15, 2026)
- Who is paying for humanoids: 69 funding rounds (humanoid.observer) (Sep 6, 2026)
- Karol Hausman, co-founder and CEO of Physical Intelligence (The Generalist) (Mar 17, 2026)
- X profile @brian_ichter (bio) (Oct 8, 2026)
- X profile @QuanVng (bio) (Oct 8, 2026)
- Physical Intelligence, a Specialist in Robot A.I., Raises $400 Million (The New York Times) (Nov 4, 2024)
- A peek inside Physical Intelligence, the startup building Silicon Valley's buzziest robot brains (TechCrunch) (Jan 30, 2026)
- X profile @adnan_esm (bio) (Oct 8, 2026)
Confidence: medium.