Periodic Labs raises $300M to automate experiments

Former OpenAI and DeepMind researchers have raised $300 million to build Periodic Labs — a network of AI-driven robotic labs that can design and run real experiments. Backed by investors including a16z, Nvidia, Jeff Bezos, and Eric Schmidt, Periodic’s stack couples autonomous research agents with physical robots that synthesize materials, heat and test them, and feed real experimental outcomes back into the models.

The idea addresses an “eval deficit” that OpenAI’s research leadership recently acknowledged: large models excel on benchmarks and web-scale data but struggle with real-world physics and chemistry because the internet and scientific literature don’t provide clean, repeatable training signals. Periodic’s loop is straightforward — humans seed evaluation goals, AI proposes experiments, robots execute them, and nature’s outcomes become the reward signal to improve the models.

Periodic is already applying the system to semiconductor heat-dissipation problems and plans to pursue superconductors, next‑gen chips, and materials research that could extend Moore’s Law. The approach reframes lab automation as continuous scientific feedback: not just simulating discovery but physically closing the loop between hypothesis, experiment, and model refinement.

Source