Periodic Labs, backed by Andreessen Horowitz (a16z), Nvidia, Jeff Bezos, and Eric Schmidt, aims to solve AI’s dirty secret: models excel at math and code but struggle with real-world physics and chemistry.
Why? They’ve exhausted the internet’s ~10 trillion training tokens, and the scientific literature is too noisy to learn from.
The timing is perfect
OpenAI’s Chief Scientist recently admitted they have an “eval deficit” and need “actual discovery” on real problems. Its new Applied Evals team is testing usefulness beyond benchmarks to create “vibe researchers,” and it is working on an automated research agent.
But vibe researchers need something to evaluate. That’s where Periodic Labs comes in.
The company is building robotic labs where AI can synthesize materials, heat them up, measure their properties, and learn from what actually happens. Real robots. Real chemicals. Real physics. As Periodic puts it: “Unless you have an experiment in the loop, we’re just thinking. Until you try it, you’re no further along.”
Connecting the dots
This looks like: Human evaluators initiate → AI agent designs experiments → Robotic labs execute them → “Nature” itself provides a reward signal (did the experiment work?) → Data improves models.
Periodic’s is already helping semiconductor manufacturers solve heat dissipation problems. They’re starting with that focus but are eventually targeting superconductors, next-gen chips, and materials that could restart Moore’s Law. This is comparable to the anti-infinite slop machine (i.e., Sora 2)… an infinite SCIENCE machine.
And as Periodic says, “you never finish science.”
Editor’s note: This content originally ran in today’s newsletter send from our sister publication, The Neuron. To read more from The Neuron, sign up for its newsletter here.