Ex-OpenAI, DeepMind pros raise $300M to build AI scientists

Just a heads up, if you buy something through our links, we may get a small share of the sale. It’s one of the ways we keep the lights on here. Click here for more.

A new startup called Periodic Labs emerged from stealth, and it didn’t tiptoe in. It stomped in carrying a $300 million seed round, the kind of check usually reserved for companies that have already changed the world. 

The backers read like a Silicon Valley hall of fame: Andreessen Horowitz, DST, Nvidia, Accel, Elad Gil, Jeff Dean, Eric Schmidt, and even Jeff Bezos. Yes, that Jeff Bezos.

So what’s all the hype about? Periodic Labs wants to build AI scientists, not just software that chats with you, but systems that can actually run experiments, test hypotheses, and iterate like real researchers. 

Imagine a robot chemist in a lab coat, only faster and with less chance of spilling acid on itself.

The brains behind the project are Ekin Dogus Cubuk and Liam Fedus. 

Cubuk previously led materials and chemistry teams at Google Brain and DeepMind, where he helped develop GNoME, an AI that discovered more than 2 million new crystals in 2023, materials researchers say could power futuristic tech. 

Fedus, meanwhile, is a former OpenAI VP of Research who helped create ChatGPT and led the team that trained the world’s first trillion-parameter neural network. 

In short, if you wanted to assemble an AI dream team to reinvent science, these would be your draft picks.

Periodic’s first target: superconductors, the holy grail of materials science. Today’s superconductors work, but they often demand freezing temperatures or tons of energy. (Via: TechCrunch)

Crack that nut, and suddenly you’ve got the building blocks for faster computers, better power grids, and maybe levitating trains that actually feel like the future.

But Periodic isn’t stopping there. The startup plans to build autonomous labs where robots mix, heat, and tweak substances endlessly, generating not just new materials but a steady stream of fresh physical-world data.

 That’s important, because as the company notes, today’s AI models have pretty much “eaten the internet.” If AI needs new fuel to evolve, Periodic wants to be the one cooking it up.

Could Periodic Labs’ AI scientists actually accelerate breakthrough discoveries in materials science, or is this $300M bet on autonomous research labs overhyped given AI’s current limitations in physical experimentation? Do you think AI-driven materials discovery represents the future of scientific research, or will human intuition and creativity remain irreplaceable in making truly revolutionary scientific breakthroughs? Tell us below in the comments, or reach us via our Twitter or Facebook.