Periodic Labs Raises $300M: Top AI Researchers Launch Startup

Periodic Labs: Pioneering AI-Driven Material Science

Periodic Labs, a recently launched startup founded by prominent researchers Liam Fedus, formerly of OpenAI, and Ekin Dogus Cubuk, previously with Google Brain, emerged from stealth mode last month. The company secured a substantial $300 million in seed funding.

This significant investment was spearheaded by Andreessen Horowitz, with initial funding provided by Felicis. A diverse group of angel investors and leading venture capitalists also participated in the round.

The Genesis of Periodic Labs

The concept for Periodic Labs originated approximately seven months ago during a discussion between Fedus and Cubuk (often referred to as Dogus by friends). Cubuk held a distinguished position as a machine learning and material science researcher at Google Brain.

Following extensive debate within Silicon Valley regarding the potential of generative AI to revolutionize scientific discovery, they concluded that the necessary components were finally aligned to translate this vision into reality. This led to the establishment of a startup dedicated to pursuing this goal.

“Several key developments converged – advancements in the field of Large Language Models (LLMs), progress in experimental science, and improvements in simulation technologies – creating an opportune moment,” Cubuk explained to TechCrunch.

Key Technological Advancements

Cubuk highlighted that robotic arms capable of handling powder synthesis – the process of combining and creating new materials – had recently demonstrated consistent reliability. This was a crucial enabling factor.

Furthermore, machine learning simulations had reached a level of efficiency and accuracy sufficient to model intricate physical systems, essential for the development of novel materials.

Crucially, LLMs now possess enhanced reasoning capabilities, partly attributable to the work of Fedus and his team at OpenAI. Fedus played a pivotal role in the creation of ChatGPT and led OpenAI’s post-training team, responsible for refining models after their initial development.

A Synergistic Approach

The integration of these technologies presents a compelling prospect: simulations can identify promising new compounds, robots can synthesize the materials, and LLMs can analyze the outcomes and recommend adjustments. This creates a fully AI-automated system for material science.

In 2023, Cubuk contributed to the groundbreaking “GNoME” research paper, detailing a model training methodology for discovering new materials. He also co-authored a study documenting a fully automated, robotic laboratory that successfully created 41 unique compounds based on recipes generated by language models.

The Value of Data and Experimentation

The founders recognized that even unsuccessful experiments would yield valuable data for their startup, as data is fundamental to AI development. This approach offers a novel source of real-world training and post-training data for AI systems.

This could potentially reshape the current scientific incentive structure, which prioritizes successful outcomes – typically measured by publications and grant funding – over exploratory research.

“Establishing a connection with real-world results, incorporating experiments into the AI learning process – we believe this represents the next significant advancement,” Fedus stated to TechCrunch.

Felicis Secures Investment in Periodic Labs; OpenAI Opts Out

Following discussions leading to his departure, Fedus communicated his resignation and future plans to OpenAI leadership. Subsequently, he publicly announced his exit, suggesting OpenAI’s support and potential investment would accompany his move.

However, this anticipated investment did not come to fruition. Founders have confirmed that OpenAI is not currently an investor in Periodic. While Fedus refrained from detailing the reasons, the company ultimately didn’t require OpenAI’s financial backing.

Fedus’ announcement sparked considerable interest from venture capitalists. He described a situation where they felt “reverse pitched,” with one investor even submitting a letter expressing strong enthusiasm for Periodic Labs. Both he and Cubuk admitted they were initially unsure how to respond to such overtures.

The first investor to engage in formal talks was Peter Deng, a former OpenAI colleague who had transitioned to the seed firm Felicis. Deng had joined Felicis earlier in 2025.

Deng emphasized Fedus’ significant contributions and popularity within OpenAI, characterizing him as a highly impactful researcher. He stated he immediately reached out to Fedus upon learning of his departure.

Their initial meeting took place over coffee in San Francisco’s Noe Valley. Energized by caffeine and excitement, Fedus proposed continuing their conversation with a walk through the neighborhood’s challenging hills. Such “pitch walks” are a known Silicon Valley practice.

As the day warmed, Deng, dressed for cooler weather, found himself struggling to keep pace with the energetic Fedus. The conversation took a pivotal turn when Fedus asserted that genuine scientific progress necessitates actively *performing* science, rather than merely discussing it.

This highlighted the need for a fully equipped laboratory environment where AI could test its hypotheses in a controlled, real-world setting.

Deng explained that current AI models primarily operate within the bounds of existing data, essentially replicating known information. Truly novel discoveries require rigorous hypothesis testing.

“I committed on the spot, in the middle of the hills of Noe Valley, to write the check,” Deng revealed.

Fedus also recalled Deng’s immediate offer of funding for laptops and a temporary office space, describing it as a significant demonstration of confidence.

However, Deng didn’t immediately produce a check. Returning to the office, he learned from Felicis’ legal counsel that a contract couldn’t be signed immediately, as the company hadn’t yet been formally incorporated or established a bank account. “That’s how early we were,” Deng remarked.

The necessary legal and financial structures were quickly established, and numerous investment offers followed. The resulting $300 million in funding enabled Cubuk and Fedus to recruit leading AI and scientific experts, including Alexandre Passos, Eric Toberer, and Matt Horton.

To foster collaboration, team members regularly deliver graduate-level lectures to one another, sharing their expertise across disciplines. “We do feel like a tight coupling is extremely important,” Cubuk stated, emphasizing the importance of shared understanding.

Periodic Labs has already established its laboratory and is actively working with experimental data, simulations, and hypothesis testing. A primary focus is the discovery of new superconductor materials – a potentially groundbreaking achievement. Advances in superconductors could enable a new generation of powerful, yet energy-efficient technologies.

The robotic components of the lab are still undergoing training and are not yet fully operational. “They will take a bit to train,” Cubuk noted.

This venture represents a substantial undertaking, as scientific discovery is rarely swift, simple, or guaranteed. While the team possesses promising indicators and anticipates breakthroughs, success is not assured.

Meanwhile, other AI developers are also increasing their focus on scientific applications. OpenAI VP Kevin Weil recently announced the launch of an “OpenAI for Science” unit, aiming to create an AI-powered platform to accelerate scientific discovery.

The investor who sent the initial “love letter” was ultimately unsuccessful in securing a deal, though Fedus acknowledged the gesture as “very flattering.” Other seed investors include DST, Nvidia’s NVentures, Accel, and prominent angel investors such as 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.

Correction: This article has been updated to reflect the lead of the seed round and to acknowledge Cubuk’s significant research paper in materials science.