Physical Intelligence Valuation 2026: $5.6B

Physical Intelligence's confirmed valuation is $5.6 billion — the price a $600 million Series B set in November 2025. By March 2026, the company was already reportedly back in the market discussing another $1 billion round that would push it above $11 billion, which would nearly double its price in about four months. Neither number is for a robot the company built. Physical Intelligence doesn't make robots at all.

That last point is the whole thesis. While Figure AI and Unitree are racing to build the best humanoid chassis, Physical Intelligence is betting the durable value in robotics sits one layer up — in the AI model that tells any robot body what to do. It's the same wager OpenAI made on language before anyone had settled on which app would win.

What Is a Robot Foundation Model — and Why Is It a Separate Business From Building Robots?

Every humanoid robot needs two things: a body that can move, and a brain that decides how to move it. For most of robotics history, one company built both, because no general-purpose "brain" existed — every robot ran narrow, hand-coded software tuned to its specific hardware. A robot foundation model breaks that coupling. It's a single AI model trained on data from many robots and many tasks that can, in principle, run on any of them, the way GPT can power a coding assistant, a customer-service bot, or a spreadsheet tool without being rebuilt for each one.

Physical Intelligence is the clearest example of a company betting entirely on that model layer. It employs no factory, ships no chassis, and doesn't compete with Figure, Unitree, 1X, or Apptronik on hardware at all — its stated goal is for its models to eventually run on all of theirs. That's a fundamentally different business than a hardware company: the addressable market isn't "how many robots can we build," it's "how many robots, made by anyone, can we make competent."

Physical Intelligence's Valuation: From Seed to $5.6 Billion

Physical Intelligence was founded in 2024 in San Francisco by Karol Hausman (CEO, formerly a staff research scientist at Google DeepMind), Sergey Levine (Chief Scientist, a UC Berkeley professor whose lab pioneered deep reinforcement learning for robotics), along with Chelsea Finn, Brian Ichter, Lachy Groom, Adnan Esmail, and Quan Vuong. The company raised roughly $1.1 billion in total across its first two years, culminating in a $600 million Series B in November 2025 that set its current confirmed valuation at $5.6 billion.

That mark didn't stand still for long. In late March 2026, Bloomberg and TechCrunch reported Physical Intelligence was in talks for a fresh $1 billion round that would value the company at more than $11 billion — nearly double the November figure in about four months, a pace that tracks the broader robotics-funding acceleration of 2026. Peter Thiel's Founders Fund was reported to be participating, with Lightspeed Venture Partners in discussions alongside returning backers Thrive Capital and Lux Capital. As of this writing, that round has been reported as in-progress, not confirmed closed — worth flagging given how often "in talks" figures get repeated as settled fact before a deal actually prices.

Who Invested in Physical Intelligence?

Physical Intelligence's confirmed cap table reads like a list of parties with a strategic reason to want a model layer independent of any one robot maker: OpenAI, Jeff Bezos (through Bezos Expeditions), NVIDIA, Google's CapitalG, Thrive Capital, and Lux Capital. OpenAI and NVIDIA in particular are backing a company whose model could, if it wins, become a layer their own compute and infrastructure sit underneath — a similar dynamic to why NVIDIA and Microsoft back Figure even though Figure builds the hardware Physical Intelligence explicitly avoids.

The reported March 2026 round would broaden that syndicate with two large multi-stage generalist funds — Founders Fund and Lightspeed — rather than more strategics, which would be a signal that pure financial investors are now willing to underwrite the model-layer thesis at an $11 billion price, not just the AI labs and chipmakers with an obvious commercial stake in the outcome.

What π0 Actually Does

Physical Intelligence's flagship product is π0 ("pi-zero"), a vision-language-action (VLA) model that takes camera images and a natural-language instruction and outputs continuous robot motor commands, using a flow-matching architecture for generating those actions. The company open-sourced π0's code in February 2025, making it a reference architecture much of the rest of the robotics-AI research community now builds against — the same role GPT-2's release played for language models before the closed-source era took over.

By April 2025, its successor π0.5 demonstrated generalization to environments the model had never seen during training. Physical Intelligence pushed further on April 22, 2026, shipping Pi-0.5, which the company describes as the first robotics foundation model that generalizes across different robot embodiments — different arms, grippers, and chassis — without needing to be fine-tuned separately for each one. In demos, π0-family models have folded laundry straight from a dryer, cleared tables, flattened cardboard boxes, and bagged groceries — dexterous, real-world chores that are trivial for humans and were, until recently, well beyond general-purpose robot software.

What the headline misses

"$5.6 billion" and "$11 billion in talks" are both easy to conflate into one settled number, and they're not the same thing. One is a closed primary round; the other is a reported negotiation that, as of this writing, hadn't been confirmed as final. It's also worth being precise about what Physical Intelligence has actually shipped versus demoed: laundry-folding and table-clearing are compelling generalization proof points, not disclosed commercial revenue. The company has not published customer names or a revenue figure — its valuation is priced almost entirely on the model layer's promise, not on paid deployments the way Figure can point to BMW.

Physical Intelligence's Valuation Compared to the Rest of Robotics AI

At a confirmed $5.6 billion, Physical Intelligence is priced below both the hardware leaders and its closest model-layer competitor. Figure AI sits at $39 billion as the highest-valued humanoid hardware company, and 1X Technologies is reported around $10 billion. But the more relevant comparison is Skild AI, the other major "no hardware, just the brain" company, which is valued at over $14 billion — roughly 2.5x Physical Intelligence's confirmed mark — after a $1.4 billion SoftBank-led round in January 2026.

Should Investors Care About the Physical Intelligence vs. Skild AI Gap?

The valuation gap between Physical Intelligence and Skild AI is a genuine strategic divergence, not noise. Physical Intelligence is backed by AI labs and research-oriented capital — OpenAI, NVIDIA, CapitalG — and has spent its short life publishing open research and generalization benchmarks. Skild AI has spent the same period acquiring a warehouse-automation business from Zebra Technologies and signing partnerships with ABB Robotics, Universal Robots, MiR, and Foxconn — an enterprise go-to-market that's arguably why SoftBank, Samsung, LG, and Schneider Electric priced it higher. Whether the market or the lab wins the model layer is the open question for the whole category, and it's one we track alongside the rest of the sector on our Humanoid Robot Race tracker.

The Bottom Line

Physical Intelligence's $5.6 billion valuation is confirmed, primary, and backed by a cap table with real reasons to want the robot-model layer to exist independent of any one hardware maker. The reported $11 billion figure is a live negotiation, not a closed round, and the company still hasn't disclosed customers or revenue the way its hardware-building peers have started to. What's not in doubt is the strategic bet: Physical Intelligence is the purest public expression of the idea that the money in robotics won't ultimately be made by whoever builds the best arm — it'll be made by whoever builds the model that makes every arm work.