Periodic Labs is making one of the clearest arguments yet for what AI-driven science could look like in practice: not just models producing suggestions, but a real laboratory system that can test those ideas and feed the results back into the loop.
The startup, founded by Liam Fedus and Ekin Dogus Cubuk, came out of stealth last month with a $300 million seed round. The round was led by Andreessen Horowitz, with Felicis cutting the first check in, and included DST, Nvidia's venture capital arm NVentures, Accel, and angel backers like Jeff Bezos, Elad Gil, Eric Schmidt, and Jeff Dean.
A large bet on AI science
The company began after a conversation between Fedus and Cubuk about seven months ago. Fedus had been one of OpenAI's most respected researchers and was part of the small team that created ChatGPT. He also ran OpenAI's post-training team, which refines models after their initial development.
Cubuk, a former Google Brain colleague of Fedus, was one of Google Brain's foremost machine learning and material science researchers. In 2023, he was one of the researchers behind the groundbreaking "GNoME" paper, a model training method for discovering new materials. He was also one of the researchers who published a paper that same year documenting a fully automated, robotic-powered lab that created 41 novel compounds from recipes suggested by language models.
The founders saw several technical pieces coming together at once. Robotic arms that can handle powder synthesis had recently proved reliable. Machine learning simulations had become efficient and accurate enough to model complex physical systems needed for new materials. LLMs had also gained stronger reasoning capabilities.
Put together, those pieces suggest a different kind of research system. A simulation could propose new compounds, a robot could mix materials in a lab, and an LLM could evaluate results and recommend what to try next.
Why the lab matters
The central idea behind Periodic Labs is that AI should not remain separated from experimental reality. Fedus described the premise to TechCrunch this way: "Making contact with reality, bringing experiments into the [AI] loop — we feel like this is the next frontier."
That distinction matters because models can only go so far by working from existing data. Peter Deng, a former OpenAI colleague of Fedus who became an investor at Felicis, told TechCrunch that models know what is within their normal distribution and can regurgitate what they already know. In his telling, the search for something genuinely new requires testing hypotheses.
This is where Periodic Labs sees an opening. The startup is not only chasing successful discoveries. It also wants the data produced by experiments that do not work. The founders believe that failed attempts can still be valuable because AI systems depend on data for training and post-training.
That could challenge the usual incentives in scientific work. Traditional research rewards success through publications and grants. Periodic Labs is betting that exploration itself, including the dead ends, can become a useful asset when tied to AI systems that learn from real-world outcomes.
How the funding came together
Fedus told OpenAI he was leaving and posted publicly that he hoped to work with the company as a partner going forward. Although the post appeared to suggest OpenAI's blessing and investment, the founders confirmed to TechCrunch that OpenAI is not a backer of Periodic Labs.
The lack of OpenAI funding did not slow investor interest. Fedus said VCs began courting the company intensely, with one investor even writing a love letter to Periodic Labs. Others sent multi-page documents pitching themselves.
The first call the founders took was from Deng at Felicis. Deng met Fedus for coffee in the Noe Valley neighborhood of San Francisco, then continued the conversation during a walk through the area's hilly terrain. Deng later said he committed on the spot to write the check after Fedus made the case that anyone serious about AI science needed to actually do science in a real lab.
Even that first commitment came before the startup had the usual basics in place. Felicis' lawyer pointed out that the firm could not promptly sign a contract because the company was not incorporated yet. It did not yet have a name or a bank account.
The team and the first target
With the $300 million seed round, Periodic Labs has hired over two dozen AI and science specialists. The team includes Alexandre Passos, a creator of o1 and o3; Eric Toberer, a materials scientist who has already made key superconductor discoveries; and Matt Horton, a creator of two of Microsoft's GenAI materials science tools.
The company is also trying to make its different areas of expertise work together closely. Because team members come from AI, physics, and other technical backgrounds, one person gives a grad-level lecture to the others each week. Cubuk told TechCrunch that the team believes tight coupling is extremely important.
Periodic Labs has already set up its lab and is working with experimental data, simulations, and tests of some predictions. Its main initial mission is to find new superconductor materials. The source article frames that as a potentially major discovery because improved superconductors could support more powerful technology with lower energy consumption.
The robotic portion of the system is not fully running yet. Cubuk said the robots "will take a bit to train." That leaves Periodic Labs in an early stage despite the scale of its funding and the profile of its team.
A high-risk path
The startup's thesis is ambitious, but the source article is clear that scientific discovery is not usually fast, easy, or predictable. AI may help organize and accelerate parts of the process, but it does not remove uncertainty from experimental science.
Periodic Labs has indications that it may find what it is looking for, make other discoveries, or at least generate useful data from failed attempts. None of those outcomes is guaranteed.
The broader field is also moving. OpenAI VP Kevin Weil said last month that he was launching an OpenAI for Science unit to "build the next great scientific instrument: an AI-powered platform that accelerates scientific discovery." That context makes Periodic Labs part of a larger push to connect model development with scientific work.
For now, the startup stands out because of the size of its seed round, the backgrounds of its founders, and the practical shape of its plan. It is not just promising AI for science. It is building around the idea that models, simulations, lab experiments, and failure data all need to operate together.