Periodic Labs builds AI systems designed to conduct scientific discovery in the physical sciences. The company combines large language models with autonomous laboratory automation, creating a closed loop between hypothesis generation and physical experimentation. Rather than relying solely on internet text, Periodic Labs uses reinforcement learning trained on real experimental data - generated by its own automated laboratories - to teach AI agents how to reason about physics and chemistry.
The company's approach addresses a specific gap: frontier AI models need more than passive learning from text to advance physical sciences. Periodic Labs runs weekly teaching sessions where physicists instruct large language models in quantum mechanics whilst machine learning researchers absorb physics fundamentals, enabling cross-disciplinary collaboration. The autonomous laboratories generate gigabytes of experimental data that exists nowhere else, providing the real-world feedback necessary to train these AI scientists.
Work spans materials science, quantum mechanics, semiconductor engineering and superconductivity research. Periodic Labs engages customers in the space, defense and semiconductor sectors, including work on heat dissipation problems for semiconductor manufacturers and involvement in discovering new superconductors. The team comprises physicists, chemists and machine learning researchers operating with minimal bureaucracy and a focus on identifying and solving problems regardless of disciplinary boundaries.