Spheroid Labs
CONTACTPhysics governs the universe — from galaxies to atoms, from protein folding to fluid flow. Every phenomenon we understand deeply, we understand through physics. Intelligence is no exception.
Virtually all of physics can be expressed as a single kind of theory. Hamiltonian mechanics, variational principles, energy minimisation — the deep structure is the same. All of physics, at its core, is one theory.
The Wright brothers didn't build a better bird. They built a wind tunnel. They isolated the mechanism of lift and solved in a few years what centuries of bird-copying could not. AI needs its Wright brothers moment — not bigger models, but the right theory.
We are building a Physics based AI engine where computation IS physics. The energy landscape encodes knowledge. The dynamics find answers. The structure of the physics provides guarantees that no black-box model can.
Early results confirm the thesis. Systems that learn from a handful of examples — because physics carries the inductive bias that data used to. Solutions that generalise to sizes never seen in training — because the rules live in the energy, not in memorised patterns.
The scaling law will continue. But to learn world models and Physcial AI, the next generation of AI models must learn the way physics does naturally, where a single equation governs every instance of a phenomenon — not just the ones it was trained on.
The hardware is catching up to the theory. Neuromorphic chips, analog accelerators, optical processors — a new generation of substrates can execute physics-native computation directly. This opens the door to intelligence that is decentralised, embodied, and running on milliwatts instead of megawatts.
We prove the paradigm on structured reasoning — combinatorial puzzles, planning, tasks that demand search and generalisation. Demonstrate that physics-native computation achieves with OOM less parameters what current methods need for.
Medium term: Scale to physical AI and embodied intelligence — robotic manipulation, dexterous control, visual reasoning, multi-modal inference. Build compositional world models where different aspects of knowledge are separate energy terms that combine the way forces combine in nature.
Long term: Intelligence that runs on physics, not despite it. Hierarchical energy landscapes for abstraction at every scale. Machines that think the way nature computes — fast, efficient, always on. Like a brain.
Former Google DeepMind researchers, physicists, and engineers. People who've always suspected that intelligence and physics are the same problem. Physicists who dream in Hamiltonians. Engineers frustrated by the gap between deep theory and current practice — building in Europe, where the scientific tradition from Euler to Hamilton to Noether to Boltzmann produced the mathematics this work requires.
Whether you're a researcher who thinks in energy landscapes, an engineer who wants to build what comes after transformers, or an investor backing fundamental breakthroughs — we want to hear from you.
peter@spheroid.ai