Periodic Labs will start off in the physical sciences. Grigorii Shcheglov for Unsplash+
It is well funded, backed by Jeff Bezos and Eric Schmidt, and staffed with talent from OpenAI, Google and Meta. But unlike many of its Silicon Valley peers, the new A.I. startup Periodic Labs has no plans to build consumer products or pursue artificial general intelligence (AGI).
“Our goal is to create an A.I. scientist,” said Liam Fedus, one of the startup’s co-founders, in a post announcing its launch yesterday (Sep. 30). “Until now, scientific A.I. advances have come from models trained on the internet. But despite its vastness, it’s still finite.”
To accelerate discovery, Periodic Labs plans to merge A.I. with traditional research methods, including running experiments in real labs. That mission has already secured $300 million in seed funding. The round was led by Andreessen Horowitz and included Nvidia, Felicis, Accel, Elad Gil, Jeff Dean, Bezos and Schmidt.
The San Francisco-based startup is also packed with renowned researchers. Fedus, who co-created ChatGPT and served as vice president of research at OpenAI, launched Periodic Labs with Ekin Dogus Cubuk, a longtime Google DeepMind scientist who formerly led its materials science and chemistry team.
The roughly 20-person founding team is a roster of veterans from major tech firms, with experience on projects like OpenAI’s Operator agent and Microsoft’s MatterGen LLM. In a playful nod to its name, each staffer at Periodic Labs picks an element from the periodic table to adorn a custom desk plate.
The company aims to build A.I. systems capable of forming hypotheses and running simulations in autonomous labs, said Peter Deng, a former OpenAI executive and now a general partner at Felicis, in a recent blog post. Deng said he was convinced to invest after Fedus told him, “in order to do science, you have to do real science.” In other words, A.I. can only learn so much from text—it needs real-world experiments to advance.
For now, Periodic Labs will focus on the physical sciences, a field rich with data and verifiable results that can support rapid A.I. progress. If successful in automating materials design, the company says its work could accelerate breakthroughs in space travel, nuclear fusion and even Moore’s Law.
The startup has already begun partnering with semiconductor makers to improve chip heat dissipation and is training agents to streamline research and engineering workflows. Its customer base also includes companies in space and defense.
Periodic Labs is not alone in its mission. Tech giants like OpenAI and Google are pursuing similar goals. DeepMind’s work on AlphaFold, an A.I. system that transformed protein structure prediction, even earned two of its researchers a Nobel Prize last year. Smaller rivals include FutureHouse, a San Francisco nonprofit also working to create an autonomous A.I. scientist.
