Physical Intelligence company profile: valuation, funding & revenue

This Physical Intelligence company profile covers the company's valuation, revenue, funding rounds, investors and history. Physical Intelligence works in artificial intelligence, with a focus on foundation models for robots. It was founded in 2024 and is headquartered in San Francisco, California, United States. Its latest valuation on record is About US$11 billion (reported) (May 2026). Every figure links to a public source, and anything the company hasn't disclosed is marked as such.

Revenue

Funding history

Physical Intelligence timeline

Key dates from the sources on this page, oldest first. Metric snapshots are listed after the funding events.

• 2024 — Physical Intelligence is founded in San Francisco by Karol Hausman (CEO), Sergey Levine, Chelsea Finn, Brian Ichter, Lachy Groom, Adnan Esmail and Quan Vuong.
• 2024 — Equity round: US$400M, valuing the company at About US$2.4B. Investors: Jeff Bezos, OpenAI, Thrive Capital, Lux Capital, Bond Capital.
• Nov 2025 — Equity round: US$600M, valuing the company at About US$5.6B. Investors: CapitalG (lead); Lux Capital, Thrive Capital, Jeff Bezos, Index Ventures, Sequoia, T. Rowe Price.
• March 2026 — Reported talks: About US$1 billion at a valuation above US$11 billion, reported by Bloomberg; not confirmed as closed.

Investors

Overview

Physical Intelligence, often shortened to Pi, is building general-purpose AI "brains" for robots. Instead of making robots itself, it develops foundation models, large AI models trained on data from many robots and tasks, that can control different kinds of robot hardware to fold laundry, clean a kitchen, pack boxes or assemble items. The goal is to do for physical tasks what large language models did for text: one model that can be adapted to many jobs.

The company was founded in 2024 in San Francisco by a group of leading robotics and AI researchers: Karol Hausman, formerly of Google DeepMind, who is CEO; Sergey Levine of UC Berkeley; Chelsea Finn of Stanford; Brian Ichter and Quan Vuong, also from Google's robotics research; Adnan Esmail, formerly of Tesla and Anduril; and investor Lachy Groom. The founding team's research credentials are a large part of why it has raised so much so quickly.

What it has built

Physical Intelligence has released a series of robot foundation models, including π0 ("pi-zero") and later versions, and has published research and open-sourced some models. Its demonstrations show robots carrying out long, multi-step household tasks in unfamiliar homes, an important test because most robots struggle in environments they weren't trained in. The models use a "vision-language-action" design: they take in camera images and language instructions and output motor commands.

Why robot data is the bottleneck

Language models learned from the vast amount of text on the internet. There is no equivalent collection of robot data: every example of a robot successfully completing a task has to be collected, usually by people operating robots remotely or demonstrating tasks. Physical Intelligence has invested heavily in collecting data across many robot types and environments. Companies that solve the data problem, through scale, simulation or clever learning methods, are likely to lead the field.

Business model

Physical Intelligence hasn't detailed a commercial model publicly. Likely paths include licensing its models to robot manufacturers and companies deploying robots, and partnerships in logistics, manufacturing and home robotics. It doesn't publish revenue and is at an early commercial stage.

Funding

• 2024: about US$400 million at roughly US$2.4 billion, from investors including Jeff Bezos, OpenAI, Thrive Capital, Lux Capital and Bond Capital.
• November 2025: US$600 million at about US$5.6 billion, led by CapitalG, Alphabet's growth fund, with Lux Capital, Thrive Capital, Jeff Bezos, Index Ventures, Sequoia and T. Rowe Price.
• 2026: Bloomberg reported talks to raise about US$1 billion at a valuation above US$11 billion. We haven't seen confirmation the round closed, so we keep the November 2025 figure as the latest confirmed valuation.

Competitive landscape

Physical Intelligence competes with Google DeepMind's robotics models, Nvidia's robot AI platforms, humanoid makers that build their own AI such as Figure AI and Tesla, and other robot foundation model startups such as Skild AI. Its advantages are its research team, its hardware-agnostic approach and its data collection across many robot types.

Summary

Physical Intelligence is a San Francisco company founded in 2024 that builds AI foundation models for robots. It raised about US$400 million at roughly US$2.4 billion in 2024 and US$600 million at about US$5.6 billion in November 2025, led by CapitalG. Reports in 2026 described talks at more than US$11 billion.

Software versus hardware in robotics

Robotics has historically been dominated by hardware companies that built machines for specific tasks. Physical Intelligence is betting that, as in computing, value will increasingly sit in software that works across many devices. If one AI model can run on robots from many makers, the company that owns it could earn software-like margins while manufacturers compete on hardware. The alternative view, held by vertically integrated companies like Figure and Tesla, is that the best results come from designing the AI and the robot together. Which approach wins will shape the industry's structure.

Points for due diligence

• How well do the models perform in new environments they weren't trained in?
• How much data does the company have, and how fast is it growing?
• Which hardware partners and customers use the models in real deployments?

The research team as an asset

Several of Physical Intelligence's founders are among the most cited researchers in robot learning. Sergey Levine's lab at UC Berkeley and Chelsea Finn's lab at Stanford have produced influential work on teaching robots through reinforcement learning and from demonstrations. Karol Hausman, Brian Ichter and Quan Vuong worked on Google's robotics models, including systems that connected language models to robot control. In a field where a small number of people understand how to train these systems, assembling such a team is itself a competitive advantage, and it helps attract other top researchers.

The company has also published papers and released some models openly. That builds credibility in the research community and encourages others to build on its work, though it means competitors can learn from it too. The balance between openness and keeping commercial advantages is one many AI companies are still working out.

Where general robot models could be used first

Early commercial uses are likely to be in settings where tasks vary but conditions are somewhat controlled: warehouses and logistics centres, light manufacturing and assembly, laboratories, and commercial kitchens. Homes are the long-term prize but are the hardest environment, because every house is different and safety expectations are very high. A model that can handle many tasks with only a little extra training for each new site would sharply reduce the cost of deploying robots, which is the main barrier today.

Latest developments (October 2026)

Bloomberg reported in March 2026 that Physical Intelligence was in talks to raise about US$1 billion at a valuation above US$11 billion, with Founders Fund, Lightspeed, Thrive Capital and Lux Capital involved, according to Bloomberg Law, and trackers list a round of about US$1.05 billion at US$11 billion closing in May. That would roughly double its November 2025 valuation of about US$5.6 billion shown above. The company has about 80 employees and has not set a commercialisation timeline. See our robotics guide.

What to watch

Watch for confirmation of a new round, commercial partnerships, new model releases, and evidence of robots running Pi models in real workplaces.

Reader notes

Robotics companies may be able to license general models rather than building AI from scratch. Founders should note how research credibility attracts capital in deep tech. Investors are backing a long-term platform bet; commercial proof will take time.

Sources

Every figure on this page comes from the public sources below. Where sources disagreed, we used the company's own announcement or a primary news report. Private-company data changes; check the date beside each figure.

1. Bloomberg Law
2. PYMNTS
3. Wikipedia