Periodic Labs, an innovative startup founded by renowned OpenAI researcher Liam Fedus and his former Google Brain colleague Ekin Dogus Cubuk, emerged from stealth mode last month with an impressive $300 million seed funding round. The initiative garnered support from notable VC firm Felicis and a diverse group of angel investors.
The inception of Periodic Labs traces back to a discussion between Fedus and Cubuk—affectionately known as “Doge”—approximately seven months ago. Cubuk, recognized as a leading figure in machine learning and materials science at Google Brain, felt inspired by the current Silicon Valley discourse around how Generative AI could revolutionize scientific discovery. They concluded that the technological landscape was finally ripe for translating this vision into reality, or at the very least, for creating a startup to explore it.
“Several advancements in large language models (LLMs), experimental science, and simulations have aligned to make this an opportune time,” Cubuk remarked in an interview with TechCrunch.
First, reliable robotic arms capable of performing powder synthesis—the process of mixing and creating new materials—have recently emerged. Additionally, machine learning simulations have reached a level of efficiency and accuracy necessary to model complex physical systems, which is crucial for the development of new materials.
Lastly, LLMs have demonstrated enhanced reasoning capabilities, largely credited to the efforts of Fedus and his team at OpenAI. Notably, Fedus was among the small group responsible for creating ChatGPT and led the organization’s essential post-training team, focusing on refining models post-development.
Together, these developments suggest a promising future: a simulation capable of discovering new compounds, a robot to mix materials, and an LLM to analyze results and recommend adjustments. AI-driven material science was on the verge of becoming a reality.
Interestingly, Cubuk played a key role in a landmark 2023 study that documented an automated robotic lab project at Google. In this endeavor, the team successfully created 41 novel compounds based on recipes proposed by language models.
TechCrunch Event
San Francisco
|
October 27-29, 2025
Crucially, the founders recognized that even unsuccessful experiments would yield valuable data, which is essential for training AI. This insight could fundamentally alter the existing scientific motivation system, which traditionally rewards success through publications and grants.
“We believe that integrating experiments with AI represents the next frontier,” Fedus stated in his conversation with TechCrunch.
Felicis Takes the Lead; OpenAI Opts Out
After their pivotal discussion, Fedus approached OpenAI’s leadership to share his resignation and vision for Periodic Labs. Excitedly, he announced his departure on social media, seemingly with OpenAI’s endorsement and potential investment.
However, contrary to expectations, OpenAI did not invest in Periodic Labs, a fact confirmed by the founders during their conversation with TechCrunch. Although Fedus refrained from explaining the decision, it became clear that they did not require OpenAI’s funding.
Fedus’s announcement sparked a fierce interest among venture capitalists, creating a whirlwind of pitches directed at the fledgling company. “It felt as if we were being reverse pitched. One investor even wrote us a love letter,” Fedus shared, laughing at the overwhelming response.
The first firm to engage with them was Felicis, represented by Peter Deng, a former colleague of Fedus at OpenAI, who transitioned to become a venture capitalist at the esteemed seed-stage firm.
“Liam is highly regarded at OpenAI—beloved and immensely impactful,” Deng remarked in his interview with TechCrunch. “I texted him immediately upon hearing about his departure.”
During their first meeting in Noe Valley, San Francisco, energized with caffeine, Fedus invited Deng to continue their conversation on a scenic walk through the region’s iconic hills—a quintessential Silicon Valley experience.
As they walked, a casual remark from Fedus stopped Deng in his tracks. “Everyone talks about doing science, but to truly do science, one must engage directly with it,” Deng recalled.
In essence, they needed to provide AI with a comprehensive wet lab to explore its hypotheses in real-world conditions.
Deng realized that for groundbreaking discoveries, testing hypotheses is essential, as the models are inherently limited to existing knowledge within normal distributions.
“I decided then and there, amidst the hills of Noe Valley, to invest immediately,” Deng professed.
Fedus also reminisced about their conversation, noting how Deng inquired about immediate needs, leading to a swift commitment to provide funds for essential resources like laptops and a temporary office. “It was an incredible vote of confidence,” he reflected.
Though Deng didn’t physically bring out his checkbook during their hike, his enthusiasm led him back to the office, where he learned that Felicis could not immediately sign a contract, as they had yet to establish an official entity or name—demonstrating just how nascent the process was.
Shortly thereafter, Periodic Labs existed in legal terms, and funding began pouring in. With the generous $300 million in hand, Fedus and Cubuk recruited over two dozen experts from various fields, including AI and physics, such as Alexandre Passos (co-creator of o1 and o3), Eric Toberer (a materials scientist with significant superconductor research), and Matt Horton (developer of two GenAI materials science tools at Microsoft). For a comprehensive list of team members, you can visit Periodic Labs’ website.
To ensure a cohesive working environment, the diverse team conducts weekly graduate-level lectures on their respective areas of expertise. “We believe that a close-knit collaboration is crucial,” Cubuk emphasized, fostering a comprehensive understanding among all team members of the objectives they are pursuing.
Periodic Labs has already established its laboratory space, actively working with experimental data, simulations, and testing predictions. Their primary objective is to identify new superconductor materials, which could lead to transformative advancements in technology, making it more energy-efficient and powerful.
However, the robotic systems are still in the training phase. “Training them will take some time,” Cubuk admitted.
This ambitious endeavor comes with inherent risks. Whether AI-powered or not, scientific discovery is rarely swift, simplified, or predictable. The expert team remains optimistic, with hopes of uncovering their target materials or yielding alternative discoveries—even if it means generating valuable data from their failures.
It’s worth noting that industry leaders are also progressing towards integrating AI into scientific discovery. Recently, OpenAI VP Kevin Weil announced the establishment of an OpenAI for Science unit, focused on creating an AI-powered platform to accelerate scientific discovery.
While the investor who penned the love letter did not secure the deal, Periodic Labs attracted additional seed investments from prominent firms, including Andreessen Horowitz, DST, NVIDIA’s venture capital division NVentures, Accel, and notable angel investors like Jeff Bezos, Elad Gil, Eric Schmidt, and Jeff Dean.
Elad Gil will discuss the impact of AI on the startup ecosystem at Disrupt in San Francisco on October 29.
