Periodic Labs, a new startup from Liam Fedus, one of OpenAI’s most respected researchers, and his former Google Brain colleague Ekin Dogus Cubuk, emerged from stealth last month with a hefty $300 million seed round. It was led by Felicis and included Angel and other top VC figures.
The startup began about seven months ago when Fedus had a conversation with Cubuk (known to his friends as “Doge”). Cubuk was one of Google Brain’s leading machine learning and materials science researchers. After endlessly experimenting with how generative AI could fundamentally change scientific discovery, Silicon Valley has decided that the ingredients are finally in place to make this a reality. At least to find a startup that has tried it.
“There were some things that were happening in the LLM field, in experimental science, in simulation, and this was the right time,” Cubuk told TechCrunch.
As an example, he said, a robotic arm that can handle powder synthesis, the process of mixing and creating new materials, recently proved its reliability. Second, machine learning simulations have become more efficient and accurate in modeling the complex physical systems needed to develop new materials.
And third, LLM now has powerful inference capabilities thanks to the work of Fedus and his team at OpenAI. Fedus was one of the small teams that created ChatGPT in the first place, and ran OpenAI’s all-important post-training team that refined the models after initial development.
Once pieced together, the picture became clear. Simulations can theoretically discover new compounds, robots can mix materials, and LLM can analyze the results and suggest course corrections. Ready to build AI-powered materials science.
In fact, Cubuk was one of the researchers who published a groundbreaking paper in 2023 that was the precursor to the Google Research Project. The team built a fully automated, robot-powered laboratory to create 41 novel compounds from recipes suggested by the language model.
tech crunch event
san francisco
|
October 27-29, 2025
Equally important, the founders realized that even failed experiments are valuable to new startups because data is the lifeblood of AI. AI science has provided an entirely new source for real-world training and post-training data. The founders believe this has the potential to upend the existing scientific motivation system, which seeks success rather than exploration that is rewarded through publications and grants.
“Getting in touch with reality and bringing experimentation into the (AI) loop. We feel this is the next frontier,” Fedas told TechCrunch.
Felicis wins the deal. OpenAI does not invest
After talking with Cubuk, Fedus went to the powers that be at OpenAI and shared his resignation and plans. He then tweeted to the world, gleefully announcing that he was leaving with OpenAI’s blessing and what appeared to be an investment.
However, that investment never materialized. OpenAI is not a backer of Periodic, the founders confirmed to TechCrunch. And although Fedus declined to say why, it didn’t actually need OpenAI’s funding.
Fedus’ tweet sparked a frenzy among VCs courting the company. “It almost felt like a contrarian move. An investor actually wrote a love letter to Periodic Labs,” Fedus said with a laugh, explaining that he and Kubek “didn’t know what to make of it.” Some sent multi-page documents promoting themselves.
But the first call they actually received was from Peter Deng, a former OpenAI colleague who went on to become an investor in leading seed company Felicis. (Deng left OpenAI in early 2025 and moved to Felicis.)
“Liam is a very important figure within OpenAI, a very well-loved, very influential researcher,” Deng told TechCrunch. “As soon as I heard he left, I emailed him.”
Mr. Deng met Mr. Fedas for coffee in San Francisco’s Noe Valley neighborhood. Fueled by caffeine and enthusiasm, Mr. Fedus invited Mr. Deng to finish their conversation with a stroll through the region’s famous hills. Pitch walks may be a Silicon Valley tradition, but they happen in real life.
The chilly days are now getting hotter. Deng Xiaoping, wearing a sweater, told TechCrunch that he hustled to keep up with the sweaty, fit and friendly Fedus until the founder uttered the words, “It literally stopped me in my tracks.” He told Deng Xiaoping, “Everyone talks about doing science, but in order to do science, you have to actually do science,” Deng recalls.
In other words, we needed to provide the AI with a fully equipped wet lab to test out ideas in a real-world, controlled environment.
“The truth about these models is that everything they know is within a normal distribution. We get a lot of data, but the data just spits out what it knows,” Deng said.
To discover something new, you need to test your hypothesis.
“And I promised to write the check right then and there, in the middle of the hills in Noe Valley,” Deng says.
Fedus also remembers the moment when Deng asked how he could get involved, and Fedus told him the startup needed cash to buy a laptop and a temporary office. “And he said, ‘Great, I’ll give you the money now.’ And that was just a huge vote of confidence.”
However, Deng Xiaoping did not actually take out his checkbook on the street. He returns to his office, elated with the deal, but is met by Felicis’ lawyer, who points out that the company is not yet incorporated and cannot close the deal right away. I didn’t even have a name, much less a bank account to transfer funds. “That’s how fast we were,” Deng Xiaoping said with a grin.
Right off the bat, they had all that stuff and all the term sheets they could handle. Cubuk and Fedus used their $300 million war chest to hire over 20 of the most prominent AI and science talents, like Alexandre Passos (creator of o1 and o3). Eric Tobeler, a materials scientist who has already made important superconductor discoveries. Matt Horton is the creator of two of Microsoft’s GenAI materials science tools. And the list goes on.
All team members are experts in different fields, from AI to physics, so each week one of us gives a graduate-level lecture to the others. “We feel that tight coupling is very important,” Cubuk said. He wants everyone to be able to understand every part of what they’re building.
Periodic Labs has also already set up a lab and is working on testing experimental data, simulations, and some predictions. The main initial mission is to find new superconducting materials, which could lead to the discovery of a gold mine. Improved superconductors could power the next era of powerful yet low-energy-consuming technologies.
But the last part, the robot, is not yet operational. “It’s going to take some time to train them,” Kubek said.
Of course, all of this is a big change for the fence. Scientific discoveries, whether powered by AI or not, are typically not quick, easy, or predictable. While there are some indications that this team of experts may find what they’re looking for, or make other discoveries along the way (or simply generate valuable data about failures), there are no guarantees.
And we know that the modelers themselves are inching their way to more AI science. Last month, OpenAI Vice President Kevin Weil said the company was launching an OpenAI for Science division to build “the next great scientific instrument: an AI-powered platform that accelerates scientific discovery.”
As for the investor he wrote a love letter to, he didn’t get the deal (although Fedas said the letter was “very flattering”). Other seed investors include Andreessen Horowitz, DST, NVentures, NVIDIA’s venture capital arm, Accel, and angel backers including Jeff Bezos, Elad Gil, Eric Schmidt, and Jeff Dean.
Elad Gil will be speaking at Disrupt in San Francisco on October 29th about how AI is changing the startup landscape.
