Periodic Labs, the latest startup from one of OpenAI’s most respected researchers, Liam Fedus, and his former Google Brain colleague Ekin Dogus Cubuk, emerged from stealth last month with a massive $300 million seed round. It was hosted by Felicis and covered information about angels and other top VCs.
The startup began when Fedus had a conversation with Cubuk (whose friends call him “Doge”) about seven months ago. Cubuk was one of the top machine learning and materials science researchers at Google Brain. After countless Silicon Valley shows how generative artificial intelligence will transform scientific discovery, it was decided that the elements that may make this concept a reality had finally been found. Or at least found a startup that attempted to do so.
“There are a few things that happened in the LLM, experimental sciences and simulation fields that made this the right time,” Cubuk told TechCrunch.
First, he said, robotic arms that may handle powder synthesis – the strategy of mixing and creating latest materials – have recently proven reliable. Second, machine learning simulations have change into powerful and accurate enough to model complex physical systems comparable to those needed to develop latest materials.
And third, LLMs now had powerful reasoning capabilities — thanks in part to the work of Fedus and his team at OpenAI. Fedus was one of the small teams that originally created ChatGPT and led the all-important OpenAI post-training team that refines the models after their initial development.
The combined picture was clear: the simulation could theoretically discover latest compounds, the robot could mix materials, and the LLM could analyze the results and suggest course corrections. Automated materials science using artificial intelligence was able to be built.
In fact, Cubuk was one of the researchers who published a groundbreaking article in 2023 documenting Google’s preceding research project. The team built a fully automated laboratory equipped with robots and created 41 novel chemical compounds based on recipes suggested by language models.
Techcrunch event
San Francisco
|
October 27-29, 2025
Equally vital, the founders realized that even failed experiments could be useful to their latest startup because data is the lifeblood of artificial intelligence. The science of artificial intelligence has provided an entirely latest source of real-world training and post-training data. According to the founders, this may occasionally turn the wrong way up the existing system of scientific motivation, which is focused on success and not on exploration, rewarded with paper publications and grants.
“Making contact with reality, introducing experiments into [AI] loop — we feel like this is the next frontier,” Fedus told TechCrunch.
Felicis wins the deal; OpenAI does not invest
After this conversation with Cubuk, Fedus went to OpenAI authorities to share his resignation and plan. Then happily he tweeted a world that he left with what gave the impression to be a blessing and an investment from OpenAI.
However, this investment didn’t actually materialize. The founders confirmed to TechCrunch that OpenAI does not support Periodic. And although Fedus would not say why, they didn’t really want OpenAI’s money.
Fedus’ tweet sparked a frenzy of VC investors courting the company. “It almost felt like the tables had been turned. One of the investors wrote a love letter to Periodic Labs,” Fedus laughed, explaining that neither he nor Cubuk “knew what to make of it.” Others submitted multi-page documents themselves.
But the first call they really answered was from Peter Deng, a former OpenAI colleague turned investor in leading seed company Felicis. (Deng left OpenAI for Felicis in early 2025)
“Liam is a very important person at OpenAI, a very well-liked and extremely influential researcher,” Deng told TechCrunch. “When I found out he was out, I texted him immediately.”
Deng met Fedus for coffee in the Noe Valley neighborhood of San Francisco. Fueled by caffeine and enthusiasm, Fedus invited Deng to complete the conversation with a walk through the area’s famously hilly terrain. Pitch walks could also be a Silicon Valley trope, but they do occur.
The cold day turned hot. Deng, wearing a sweater, was sweating and attempting to sustain with the athletic and friendly Fedus until the founder said something that “literally stopped me in my tracks,” Deng told TechCrunch. He told Deng that “everyone talks about doing science, but to do it, you actually have to do it,” Deng recalled.
In other words, that they had to supply the AI with a fully equipped wet lab so it could check out its ideas in a controlled, real-world environment.
“The truth about these models is that everything the models know falls within a normal distribution. We take in a lot of data and they can just return what they know,” Deng said.
Discovering something latest must involve testing hypotheses.
“And I committed to writing a check on the spot, in the middle of the hills in the Noe Valley,” Deng says.
Fedus also remembers the moment when Deng asked how he could get involved and Fedus told him the startup needed money for laptops and a temporary office. And he said, “Great, I’ll give you the money in a minute. And that was just a huge vote of confidence.”
But Deng didn’t actually take out his checkbook on the street. He returned to the office blissful with the transaction and met with Felicis’ lawyer, who identified to him that the company couldn’t sign the contract immediately: the company had not yet been registered. It didn’t even have a name, much less a checking account to which funds may very well be transferred. “We were so early.” Deng smiled.
Soon that they had all of it and all the term sheets they may handle. With a $300 million war chest at their disposal, Cubuk and Fedus have hired over twenty of the most prestigious talents in artificial intelligence and science, comparable to Alexandre Passos (creator of o1 and o3); Eric Toberer (materials scientist who has already made key discoveries in superconductors); and Matt Horton, creator of two Microsoft materials science tools, GenAI. AND the list goes on.
Since all team members are experts in various fields, from artificial intelligence to physics, each week one of them gives a master’s-level lecture to a different. “We believe tight connection is extremely important,” Cubuk said. He wants everyone to know all parts of what they are building.
Periodic Labs has also already arrange its own lab and is working on experimental data, simulations and testing some predictions. The predominant initial mission is to search out latest superconducting materials – potentially discovering a gold mine. Improved superconductors could power the next era of powerful but less energy-intensive technology.
But the last part – the robots – has not been launched yet. “It will take them some time to train,” Cubuk said.
This is all a huge swing for fences, in fact. Scientific discoveries, whether AI-based or not, are typically not quick, easy or predictable. While the team of experts has some confidence that it would find what it’s looking for – or make other discoveries along the way (or simply generate useful data on its failures), there are no guarantees.
We also know that modelers themselves are moving toward more AI science. Last month, OpenAI Vice President Kevin Weil said it was launching OpenAI for Science units inside the company to “build the next great scientific instrument: an AI-powered 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 seed investors include Andreessen Horowitz, DST, Nvidia’s enterprise arm NVentures, Accel and angel investors comparable to Jeff Bezos, Elad Gil, Eric Schmidt and Jeff Dean.