Periodic Labs, a new startup from one of OpenAI’s most respected researchers, Liam Fedus, and his former colleague at Google Brain, Ekin Dogus Cubuk, came out of stealth last month with a massive $300 million seed round. It was chaired by Felicis and featured a who’s who of angels and other top VCs.
The startup started when Fedus had a conversation with Cubuk (whose friends call him “Doge”) about seven months ago. Cubuk was one of Google Brain’s top machine learning and materials science researchers. After endless Silicon Valley takes on how genetic AI would revolutionize scientific discovery, they decided the pieces were finally in place to make it a reality. Or at least found a startup that attempted it.
“There are a couple of things that happened in the LLM field, in experimental science, and in simulations that made this moment the right time,” Cubuk told TechCrunch.
For one, he said, robotic arms that could handle powder compounding — the process of mixing and creating new materials — had recently proven reliable. On the other hand, machine learning simulations had become efficient and accurate enough to model complex physical systems, such as those needed to develop new materials.
And, third, LLMs now had strong reasoning capabilities — in part through the work of Fedus and his team at OpenAI. Fedus was one of the small team that built ChatGPT to begin with, and he leads OpenAI’s all-important post-training team, which refines the models after their initial development.
Combined, the picture was clear: A simulation could theoretically discover new compounds, a robot could mix the materials, and an LLM could analyze the results and suggest course corrections. Automated materials science with AI was ready to be built.
In fact, Cubuk was one of the researchers who published a groundbreaking paper in 2023 documentation of a precursor Google research project. The team built a fully automated, robotic laboratory and created 41 new compounds from recipes suggested by language models.
Techcrunch event
San Francisco
|
27-29 October 2025
Just as importantly, the founders realized that even failed experiments would be valuable to their new company because data is the lifeblood of AI. The science of artificial intelligence offered an entirely new source for real-world training and post-training data. This could, the founders believe, overturn the existing system of scientific incentives, which pursue success, not exploration, rewarded through publication and grants.
“Making contact with reality, bringing experiments to [AI] loop — we feel like that’s the next frontier,” Fedus told TechCrunch.
Felicity wins the deal. OpenAI does not invest
After this discussion with Cubuk, Fedus went to the OpenAI powers to share his resignation and his plan. Then happy he tweeted to the world that was walking away with what appeared to be OpenAI’s blessing and investment.
However, this investment did not actually materialize. OpenAI is not a supporter of Periodic, the founders confirmed to TechCrunch. And while Fedus declined to say why, they didn’t actually need OpenAI’s money.
Fedus’ tweet sparked a frenzy of VCs courting the company. “There was almost like a sense of reversal. An investor actually wrote a love letter to Periodic Labs,” Fedus laughed, explaining that neither he nor Cubuk “knew what to do with it.” Others sent multi-page documents introducing themselves.
But the first call they made was from Peter Deng, a former OpenAI colleague turned investor for leading seed company Felicis. (Deng left OpenAI for Felicity in early 2025.)
“Liam is a very big deal at OpenAI, very well-liked and an extremely impressive researcher,” Deng told TechCrunch. “When I heard he left, I texted him right away.”
Deng met Fedus for coffee in San Francisco’s Noe Valley neighborhood. Fueled by caffeine and excitement, Fedus invited Deng to complete their conversation on a ride through the area’s famously hilly terrain. Field trips may be a Silicon Valley trope, but they also really happen.
The cold day had turned hot. Deng, wearing a sweater, was sweating and shuffling to keep up with the kind and friendly Fedus until the founder said something that “literally stopped me,” Deng told TechCrunch. He told Deng that “everyone talks about science, but to do science, you have to actually do science,” Deng recalled.
In other words, they had to give the AI a fully equipped wet lab to test its ideas in a real, controlled environment.
“The truth about these models is that everything the models know is within the normal distribution. We get a bunch of data and they can just recall what it knows,” Deng said.
Discovering something new must involve hypothesis testing.
“And I committed on the spot, in the middle of the hills of Noe Valley, to write the check,” says Deng.
Fedus also remembers when Deng asked how he could get involved and Fedus told him the startup needed cash for laptops and a temporary office. And “it’s, great, I’ll give you money right away. And it was just this huge vote of confidence.”
But Deng didn’t actually take his checkbook out on the street. He returned to the office excited about the deal only to meet Felicis’ lawyer, who pointed out that the company couldn’t sign a contract right away: The company hadn’t been incorporated yet. He didn’t even have a name, let alone a bank account to pay funds. “That’s how early we were.” Deng smiled.
Soon they had all these things and all the term sheets they could handle. With their $300 million war chest, Cubuk and Fedus hired over two dozen of the most famous AI and science talent, such as Alexandre Passos (creator of o1 and o3). Eric Toberer (a materials scientist who has already made important superconductor discoveries). and Matt Horton, creator of two of Microsoft’s GenAI materials science tools. And the list goes on.
Because the team members are all experts in different fields, from artificial intelligence to physics, each week one of them gives a grade-level lecture to the others. “We feel that a tight coupling is extremely important,” Cubuk said. He wants everyone to understand all the parts of what they are building.
Periodic Labs has already set up its lab, and is working with experimental data, simulations and testing some predictions. The main initial mission is to find new superconducting materials – possibly a gold mine discovery. Improved superconductors could power the next era of powerful, but lower-power technology.
But the last part – the robots – isn’t up and running yet. “They’re going to need some training,” Cubuk said.
All of this is, of course, a lot of swinging for the fences. With artificial intelligence or not, scientific discovery is not usually quick, easy or predictable. While this team of experts has some indication that they’ll find what they’re looking for — or make other discoveries along the way (or simply generate valuable data about their failures), there are no guarantees.
And we know that the model makers themselves are marching towards the science of artificial intelligence. Last month, OpenAI vice president Kevin Weil said he was launching an OpenAI for Science unit at the company to “build the next great scientific instrument: an artificial intelligence platform that accelerates scientific discovery.”
As for the investor who wrote the love letter, he didn’t win the deal (although Fedus said the letter was “very flattering.”) Other investors include Andreessen Horowitz, DST, Nvidia’s venture capital arm NVentures, Accel and angel backers such as Jeff Bezos, Elad Gildan, Eric Shffmi.
Elad Gil will talk about how artificial intelligence has changed the startup landscape at Disrupt in San Francisco on October 29.
