Physical Superintelligence secures $58m to build AI physics platform

Physical Superintelligence is betting that the next big frontier for AI is not chatting or coding, but doing physics. With a $58 million seed round and an early flagship customer in the Fermi Explorer Mission, PSI is positioning its Get Physics Done system as an “AI physicist” that can scope problems, choose simulations and discover novel trajectories that human teams missed. The fact that this stack produced a viable perihelion‑pump route for an 80,000‑year Alpha Centauri mission is less important than the pattern: high‑stakes physical design tasks are being handed to agentic systems.

For the race to AGI, this sits in the same family as DeepMind’s AlphaFold and AI‑for‑science initiatives at other labs, but with a more explicit commercial focus. If PSI can consistently turn physics questions into optimised designs for data centres, energy systems or space infrastructure, it creates a template for domain‑specific superintelligence, where deeply specialised agents outclass generalist models on consequential problems. That sort of capability can feed back into better chips, cheaper compute and more efficient infrastructure for frontier models themselves.

The Fermi Explorer tie‑in is also a clever narrative move. A mission that exists partly to test an AI‑discovered trajectory generates public legitimacy for using agentic systems in scientific planning, even when timelines stretch beyond human civilisation. The more comfortable institutions get with that pattern, the easier it becomes to give similar systems a role in sensitive domains closer to home.