one year on
Periodic Labs exits stealth with $300M seed to build AI scientists that run their own labs
Two of the researchers behind ChatGPT and GNoME say the internet has been exhausted as a training source, and the next corpus has to be generated by robots mixing chemicals.
Originally published Published on this site
Periodic Labs came out of stealth on Tuesday with a $300 million seed round, backed by a tech industry who’s who: Andreessen Horowitz, DST, Nvidia, Accel, Elad Gil, Jeff Dean, Eric Schmidt and Jeff Bezos.
The founders are Liam Fedus, a former VP of research at OpenAI who worked on ChatGPT and led the team behind the first trillion-parameter neural network, and Ekin Dogus Cubuk, who ran the materials and chemistry team at Google Brain and DeepMind and whose GNoME project surfaced more than 2 million candidate crystals in 2023. The team around them is drawn from OpenAI’s Operator and Microsoft’s MatterGen, among other projects.
The pitch is not another model. Periodic says it is building AI scientists and the robotic laboratories they operate: powder synthesis rigs where machines mix precursors and apply heat, characterize what comes out, and feed the results back. The lab’s first goal is to invent new superconductors, and it also hopes to find other new materials such as magnets and heat shields. The company argues the internet has been exhausted as a training source, and that experimental data generated in-house is the next corpus. It says it is already working with customers in space, defense and semiconductors.
It is not the only group chasing automated chemistry — academic labs, nonprofits and smaller startups have been at it since at least 2023.
The record
The a16z venture partner who led the round writes that frontier models are objectively poor at scientific analysis, and in his own measurements performed worse than human investigators; he argues the missing ingredient is real-world data, so that nature itself becomes the reinforcement learning environment.
In an introductory blog post, the company says AI advances in science have until now come from models trained on the internet, that LLMs have exhausted that source, and that it is building AI scientists and the autonomous laboratories for them to operate.
One year later — open only if you can handle spoilers
Periodic became one of the flagship bets in a wave of 'AI scientist' startups alongside Lila Sciences. By December 2025, MIT Technology Review framed the test facing the whole field as moving AI-predicted materials out of simulation and into real, synthesized and measured compounds.
Filed
Compiled and edited by Adam Allcock.
The Weekly Replay · free by email
This week, one year ago — every Sunday.
One email each Sunday: the week's replayed AI news, with the one-year-later annotations included. Written like it's breaking — dated like it isn't.
Free · double opt-in · unsubscribe anytime · privacy