Physical Superintelligence raises $58M seed — and it is not a robot company

Physical Superintelligence, a Cambridge, Massachusetts company, launched on September 2, 2026 with a $58 million seed round led by Breakthrough Energy Ventures. Dragon Global, Robot Ventures, SV Angel and others participated, along with angel investors from OpenAI and NVIDIA. The founders are Matt Pines, Alex Klokus and Dr Alexander Wissner-Gross.

The product is Emmy, which the company describes as an AI-native physics platform — “virtual physicists” running curated simulations to break research problems into hypotheses that can be checked. The first commercial target the company names is optimising the design and operation of data centres, terrestrial and orbital, across power, cooling, networking and compute. That is a large seed for a company that has not yet published results.

A naming collision worth pinning down

“Physical AI” has, until this year, meant one thing in this industry: models that control bodies — arms, legs, wheels. This round uses the phrase for something else entirely: AI applied to physics as a science, with no actuator anywhere in the stack. Both are defensible readings of the words. Only one of them is what a robotics investor means when they use the term.

The reason to hold the line is that funding tallies are already blurring. Aggregate “physical AI” totals for 2026 will be materially larger if simulation-and-discovery companies are counted alongside robot makers, and the two categories have different customers, different unit economics and different failure modes. Readers comparing this round with a humanoid Series B are not comparing like with like.

Where it does touch embodied AI

The overlap is real, though narrower than the name suggests. Robotics depends on simulators that are physically faithful enough for a policy trained inside them to survive contact with a real object — the problem covered here in sim-to-real. And the ambition to learn the dynamics of a system well enough to predict it is the same ambition behind the world-model bet now being placed by several robotics labs. A company that genuinely improves how simulated physics is constructed and verified would eventually matter to both.

“Eventually” is doing work in that sentence. Emmy is a seed-stage platform with no published benchmark, no disclosed customer and no peer-reviewed result attached to the announcement. The claim of discovering new physical laws is a statement of intent, not a demonstration. The strongest thing that can be said today is that a serious energy investor and technically literate angels underwrote the team.

The read

The choice of data centres as a first market is the most informative detail. It is a domain with expensive, measurable inefficiencies, where a simulation that gets thermal and power interactions slightly less wrong has an immediate dollar value — and where a customer can validate the claim within a quarter. That is a sensible way to make an open-ended research bet legible to a buyer. It is also a long way from robots, and the company has not claimed otherwise; the conflation is ours to avoid, not theirs.