Physical Superintelligence Emerges From Stealth With $58M to Build an AI Physics Lab — and an Interstellar Mission

On September 1, 2026, a Cambridge, Massachusetts startup called Physical Superintelligence (PSI) emerged from stealth with $58 million in seed funding and one of the most unusual founding charters in the current AI boom: to “industrialize the discovery of new physics” at machine scale. While most AI capital chases chatbots, coding agents, and video generation, PSI is betting that the same agentic reasoning wave can be pointed at physics itself — designing data centers today, and planning humanity’s first spacecraft to another star.

The round was led by Breakthrough Energy Ventures, the climate-tech fund backed by Bill Gates, with participation from Dragon Global, Robot Ventures, Solari, Susa, Ron Conway’s SV Angel, and Valkyrie. Angel investors include Balaji Srinivasan and Anthony Scaramucci, along with individuals from OpenAI, NVIDIA, SoftBank Energy, Oracle, Hugging Face, JUMP Capital, and the a16z Scout Fund — a coalition that reads like a deliberate bridge between the AI frontier and the energy-to-compute infrastructure world.

A physics lab staffed by virtual physicists

PSI was co-founded by Matt Pines (CEO), Alex Klokus, and Dr. Alexander Wissner-Gross, a Harvard-trained physicist and computer scientist who has spent years writing about the coming cascade of AI-driven scientific discovery. The company describes itself as an AI-native physics research lab, organized as a public benefit corporation, with team roots at Google, OpenAI, Harvard, Meta, NVIDIA, MIT, Stanford, Oxford, Johns Hopkins, Cambridge, the Institute for Advanced Study, and the Perimeter Institute.

Its core platform is Emmy, named after the mathematician Amalie Emmy Noether, whose theorems connect symmetry and conservation laws — arguably the most foundational result in modern physics. According to the launch announcement, Emmy is a team of virtual physicists built on a “sovereign reasoning engine” and a large curated inventory of simulations. Where a human physicist carries one mental model of a system at a time, Emmy constructs higher-fidelity world models, decomposes research problems into trees of verifiable hypotheses, and tests them in parallel.

The first commercial application is deliberately pragmatic: AI data centers and AI factories, both terrestrial and orbital. Emmy applies physics-native reasoning and simulation to multiphysics design problems spanning power, cooling, network, and compute — optimizing infrastructure before construction and retrofitting gains into facilities already running. It is a shrewd wedge: the AI industry’s own compute buildout is arguably the most physics-constrained engineering problem on Earth, with heat, power density, and thermodynamics now gating scaling as much as chip supply.

“PSI aims to industrialize the discovery of new physics,” Pines said in the announcement. “Today, that means giving our customers a measurable edge to design and run their data centers more efficiently. Tomorrow, it means going after physics problems that have been untouched for decades.”

Get Physics Done, open-sourced

PSI has also developed Get Physics Done (GPD), described as the first open-source agentic AI physicist, and is open-sourcing it. Released earlier this year and hosted on GitHub, GPD is designed for long-horizon research work that requires rigorous verification, structured research memory, and multi-step analysis: it scopes a problem, derives equations, sets up simulations, and produces a structured research workflow end-to-end. For a field where “AI for science” has mostly meant literature search and hypothesis ranking, an agent that attempts the full arc of a physics investigation is a notable step — and the open-source release gives the community a concrete artifact to audit.

The Fermi Explorer: an AI-planned mission to Alpha Centauri

The most eye-catching item in the launch is PSI’s role as founding technical partner of the Fermi Explorer Mission, a 501(c)(3) nonprofit billed as the first AI-planned interstellar mission to Alpha Centauri. The mission’s stated objectives are radical in their frugality: launch humanity’s first spacecraft toward another star, reaching at least 99% of the way to Alpha Centauri within 80,000 years; carry at least a 1 kg payload inside a 10×10×10 cm volume; launch before the end of 2029; and cost less than $15 million to design, build, launch, and operate.

PSI says it validated the mission’s physics, identified a substantially more efficient trajectory within the mission’s mass and budget constraints, and will contribute additional scientific instrumentation to the flight. The work is documented in a July 2026 technical feasibility assessment prepared by PSI, which makes an extraordinary claim: every quantitative result in the assessment was produced by PSI’s autonomous physics-research platform end-to-end, under staged independent audit — with separate verification agents re-deriving each headline number and an adversarial proof audit applied to theorem-grade claims. The document also carries an explicit notice that it did not undergo comprehensive human peer review, a transparency choice that is itself a data point about how AI-generated science may be vetted going forward.

Analysis: physics as the next agent frontier

PSI’s launch lands at a moment when the industry is asking where agents go after code and customer support. The company’s thesis maps to a broader pattern: verifiable domains — mathematics, formal proofs, simulation-predictable engineering — are where agentic AI can self-check and compound fastest. Physics, with its hard constraints and unforgiving ground truth, is both the hardest and the most credible arena for that bet.

The risks are equally clear. “$58 million” is pocket change next to the billions flowing into frontier model labs, and “discovering new physical laws” is a goal measured in decades, not funding cycles. The data-center wedge gives PSI a plausible revenue engine while the long-horizon research matures. And the Fermi Explorer’s 80,000-year timeline won’t be judged in any investor’s lifetime — but the 2029 launch deadline and $15 million budget will be.

What PSI has demonstrably shipped so far — an open-source agentic physicist, an audited feasibility study for an interstellar probe, and a simulation-grounded reasoning platform — makes it one of the more substantive stealth debuts of the year. Whether Emmy becomes a factory for physical superintelligence or a very sophisticated data-center consultancy, the launch signals where the next wave of AI ambition is heading: out of the chat window and into the physical world.

PSI says the seed funding will fund hiring in Boston and San Francisco, construction of the Emmy platform’s models, simulation infrastructure, and verification systems, and directed research across the energy-to-compute stack — from novel sensing to novel compute substrates.