The History of Physical Intelligence

01Founding

Physical Intelligence — often written π — was founded in San Francisco in early 2024 by seven co-founders: CEO Karol Hausman, Chief Scientist Sergey Levine, Research Lead Chelsea Finn, Brian Ichter, COO Lachy Groom, Adnan Esmail, and Quan Vuong. The founding team's robotics pedigree is dense even by AI-startup standards. Hausman, a Staff Research Scientist at Google DeepMind and Stanford adjunct professor before co-founding the company, was central to RT-2 and SayCan, two of the most-cited systems connecting large language models to physical robot control. Levine, a UC Berkeley professor, and Finn, a Stanford assistant professor, are both widely published robot-learning researchers who advised or collaborated with much of the rest of the founding team during their own PhDs.

The company's framing, as Levine put it to TechCrunch in January 2026, is deliberately blunt: "Think of it like ChatGPT, but for robots." In the same profile he argued that "good intelligence compensates for bad hardware" — the company's rationale for building a general-purpose brain rather than its own robot body, unlike hardware-first rivals such as Figure or Unitree. COO Lachy Groom, an early Stripe employee turned robotics-focused angel investor before joining as a co-founder, brought startup-operating experience the largely academic founding team otherwise lacked.

02Funding

Physical Intelligence's funding accelerated unusually fast. A $70 million seed round, led by Thrive Capital, closed and was announced alongside the company's public launch in March 2024, reportedly valuing the young company at around $400 million — a figure worth distinguishing from the separate $400 million Series A that followed eight months later, since the two are coincidentally the same number for very different things (a valuation versus a raise amount). That Series A closed in November 2024, backed by Jeff Bezos, Thrive Capital, Lux Capital, and the OpenAI Startup Fund, at a valuation reported as either $2 billion or $2.4 billion depending on the outlet. A $600 million Series B closed in November 2025, led by Alphabet's growth-investing arm CapitalG, valuing the company at $5.6 billion — meaning total funding through that round came to roughly $1.07 billion.

By late March 2026, Physical Intelligence was reported to be in talks for a further roughly $1 billion round that would value the company above $11 billion — more than double its Series B valuation in about four months, with Founders Fund confirmed as a participant. As of this writing, that round hadn't been confirmed as closed in any source reviewed; the company's actual, current valuation should be read as "reportedly in talks for over $11B," not a finalized figure.

03The model line

Physical Intelligence doesn't build its own hardware. Its models are trained across other companies' robot platforms — UR5e arms, Trossen's bimanual arms, mobile manipulators, and the open ALOHA/DROID research platforms — and released as software. π0, the company's first generalist robot policy, launched October 31, 2024, demonstrating tasks like folding laundry and bussing tables across eight distinct robot configurations. Physical Intelligence open-sourced the base π0 model's code and weights in February 2025, alongside fine-tuned checkpoints for specific tasks — an unusual move for a company sitting on more than $1 billion in funding, and one that let outside teams (including Hugging Face, which built an independent PyTorch implementation) build on the work directly.

π0.5, released April 2025, extended the approach to open-world generalization with chain-of-thought planning, tested by mobile manipulators doing chores in three real homes the model had never seen during training. By November 2025, π*0.6 added reinforcement learning on top of demonstration data, and the most recent model as of this writing, π0.7 (April 2026), demonstrated what Physical Intelligence calls compositional generalization — Levine has described a version that learned to operate an air fryer from just two training examples by recombining previously learned skills, while also acknowledging its limits: "you can't tell it, 'Hey, go make me some toast'" without step-by-step guidance.

04No hardware, no customers, no revenue

Physical Intelligence has been unusually candid that it doesn't have a commercialization plan yet. COO Lachy Groom told TechCrunch in January 2026, "I don't give investors answers on commercialization. That's sort of a weird thing, that people tolerate that." The company has run pilot testing with unnamed partners in logistics, grocery, and — per the same profile — a chocolate manufacturer, but has disclosed no customer names, no revenue figures, and no product for sale. Headcount estimates vary sharply by source and date, from roughly 80 employees (TechCrunch, PitchBook) to more than 200 (Tracxn, Forbes) in reporting from the first half of 2026 — a spread wide enough that any single figure should be treated as approximate.

That research-first posture sits in explicit contrast with Physical Intelligence's most direct rival, Skild AI, which has emphasized commercial deployments and disclosed roughly $30 million in 2025 revenue. Press coverage routinely frames the two companies as taking opposite bets on the same underlying thesis — that a single foundation model can generalize across robot bodies — with Physical Intelligence betting research breadth will eventually convert into product, and Skild betting commercialization now builds the data and revenue to fund further research.

05The OpenAI relationship

OpenAI's venture arm has invested in Physical Intelligence across both its Series A and, per some reporting, Series B rounds — while OpenAI itself has separately begun building its own embodied-AI and robotics-hardware division, expanding hiring into roles spanning actuator design, 3D-printing lab technicians, and data-acquisition engineering. That's a familiar pattern in this sector: OpenAI backed Figure AI's Series B in February 2024 before Figure ended their model-development partnership roughly a year later, with Figure's CEO saying outsourcing AI "doesn't work for real-world embodied robotics." No comparable break has been reported between OpenAI and Physical Intelligence as of this writing, but the structural tension — a major investor building a potentially competing product — is the same one that already played out with a different portfolio company.

06FAQ

07Sources