Periodic Labs, a brand new startup by considered one of OpenAI’s most revered researchers, Liam Fedus, and his former Google Mind colleague, Ekin Dogus Cubuk, got here out of stealth final month with an unlimited $300 million seed spherical. It was led by Felicis and included a who’s who of angels and different high VCs. 

The startup started when Fedus had a dialog with Cubuk (whose pals name him “Doge”) about seven months in the past. Cubuk was considered one of Google Mind’s foremost machine studying and materials science researchers. After infinite Silicon Valley takes on how GenAI would transform scientific discovery, they determined that the items have been lastly in place to make this a actuality. Or at the least to discovered a startup that tried it.

“There are some things that occurred within the LLM discipline, in experimental science and in simulations that sort of made this the proper time,” Cubuk informed TechCrunch.

For one, he stated, robotic arms that would deal with powder synthesis — the method of blending and creating new supplies — had not too long ago proved themselves dependable. For an additional, machine studying simulations had turn into environment friendly and correct sufficient to mannequin complicated bodily programs akin to these wanted to develop new supplies. 

And, third, LLMs now had highly effective reasoning capabilities — partially by means of the work of Fedus and his group at OpenAI. Fedus was one of many small group who created ChatGPT to start with and was operating OpenAI’s uber essential post-training group, which refines fashions after their preliminary growth.

Stitched collectively, the image was clear: a simulation might theoretically uncover new compounds, a robotic might combine the supplies, and an LLM might analyze the outcomes and counsel course corrections. AI-automated materials science was able to be constructed.

Actually, Cubuk was one of many researchers who revealed a groundbreaking paper in 2023 documenting a precursor Google analysis mission. The group constructed a completely automated, robotic-powered lab and created 41 novel compounds from recipes instructed by language fashions. 

Equally importantly, the founders realized that even failed experiments can be helpful for his or her new startup as a result of knowledge is the lifeblood of AI. AI science provided a wholly new supply for real-world coaching and post-training knowledge. This might, the founders consider, flip the present scientific motivation system on its head, which seeks success, not exploration, rewarded through paper publication and grants.

“Making contact with actuality, bringing experiments into the [AI] loop — we really feel like that is the following frontier,” Fedus informed TechCrunch. 

Felicis wins the deal; OpenAI doesn’t make investments 

After that dialogue with Cubuk, Fedus went to the powers that be at OpenAI to share his resignation and his plan. He then gleefully tweeted to the world that he was leaving with what gave the impression to be OpenAI’s blessing and funding.  

That funding didn’t really materialize, nonetheless. OpenAI just isn’t a backer of Periodic, the founders confirmed to TechCrunch. And whereas Fedus declined to say why, they really didn’t want OpenAI’s cash.  

Fedus’s tweet set off a frenzy of VCs courting the corporate. “There was nearly like a sense of being reverse pitched. One investor really wrote a love letter to Periodic Labs,” Fedus laughed, explaining that neither he nor Cubuk “knew what to make of it.” Others despatched multi-page paperwork pitching themselves.

However the first name that they really took was from Peter Deng, a former OpenAI colleague turned investor for top-tier seed agency Felicis. (Deng left OpenAI for Felicis at first of 2025.)  

“Liam is a really massive deal inside OpenAI, very properly liked and a particularly impactful researcher,” Deng informed TechCrunch. “After I heard he left, I texted him instantly.” 

Deng met Fedus for espresso within the Noe Valley neighborhood of San Francisco. Hyped on caffeine and enthusiasm, Fedus invited Deng to complete their dialog on a stroll over the world’s famously hilly terrain. Pitch walks could also be a Silicon Valley trope, however in addition they actually occur. 

The chilly day had turned scorching. Deng, sporting a sweater, sweated and scrambled to maintain up with the match and pleasant Fedus till the founder stated one thing that “actually stopped me in my tracks,” Deng informed TechCrunch. He informed Deng that “everybody talks about doing science, however with the intention to do science, you really need to do science,” Deng recollects. 

In different phrases, they wanted to provide AI a completely outfitted moist lab to strive its concepts in an actual world, managed setting.

“The reality about these fashions is that every thing that the fashions know is inside regular distribution. We take a bunch of information, and it may simply regurgitate what it is aware of,” Deng stated.  

Discovering one thing new has to contain testing hypotheses.  

“And I dedicated on the spot, in the midst of the hills of Noe Valley, to put in writing the verify,” Deng says.  

Fedus additionally remembers the second Deng requested how he may very well be concerned, and Fedus informed him the startup wanted money for laptops and a short lived workplace. And “he’s like, nice, I’ll provide you with cash proper now. And it was simply this enormous vote of confidence.”  

However Deng didn’t really whip out his checkbook on the road. He went again to the workplace elated over the deal solely to come across Felicis’ lawyer, who identified that the agency couldn’t promptly signal a contract: the corporate wasn’t integrated but. It didn’t actually have a identify, a lot much less a checking account to wire funds. “That’s how early we have been,” Deng grins. 

Quickly they’d all of these issues and all of the time period sheets they might deal with. With the $300 million struggle chest, Cubuk and Fedus employed over two dozen of essentially the most prestigious AI and scientific expertise like Alexandre Passos (a creator of o1 and o3); Eric Toberer (a supplies scientist who has already made key superconductor discoveries) and Matt Horton, a creator of two of Microsoft’s GenAI supplies science instruments. And the checklist goes on. 

As a result of the group members are all consultants in numerous areas from AI to physics, every week considered one of them provides a grad-level lecture to the others. “We do really feel like a decent coupling is extraordinarily essential,” Cubuk stated. He needs everybody to know all components of what they’re constructing. 

Periodic Labs has already arrange its lab, too, and is working with experimental knowledge, simulations and testing some predictions. The primary preliminary mission is to search out new superconductor supplies – probably a gold mine discovery. Improved superconductors might energy the following period of potent, however decrease energy-consuming tech. 

However the final half – the robots – should not but up and operating. “They’ll take a bit to coach,” Cubuk stated. 

All of that is, after all, a giant swing for the fences. AI powered or not, scientific discovery just isn’t sometimes quick, simple or predictable. Whereas this group of consultants has some indications that they may discover what they’re in search of — or make different discoveries alongside the way in which (or just generate helpful knowledge on their failures), there’s no ensures.

And we all know that mannequin makers themselves are inching their manner towards extra AI science. Final month, OpenAI VP Kevin Weil stated he was launching an OpenAI for Science unit on the firm to “construct the following nice 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 (though Fedus did say that the letter was “very flattering.”) The opposite seed buyers embody Andreessen Horowitz, DST, NVIDIA’s enterprise capital arm NVentures, Accel, and angel backers like Jeff Bezos, Elad Gil, Eric Schmidt, and Jeff Dean. 

Elad Gil will probably be talking about how AI has modified the startup panorama at Disrupt in San Francisco on October 29.