Periodic Labs, founded by ex-OpenAI and DeepMind researchers, has raised $300 million in seed funding to transform scientific discovery by building AI scientists and autonomous laboratories. Founded by Ekin Dogus Cubuk, former leader of Google Brain and DeepMind’s materials and chemistry team, and Liam Fedus, former OpenAI VP of Research and one of the architects behind ChatGPT, Periodic Labs is backed by a list of impressive investors, including Andreessen Horowitz, Nvidia, Jeff Dean, Eric Schmidt, Jeff Bezos, and more.
Cubuk’s prior work included GNoME, an AI model that discovered over two million new crystal structures in 2023, revealing new materials with potential for next-gen technologies. Fedus contributed substantially to OpenAI’s breakthroughs, including leading development of the first trillion-parameter neural network. Their team also features key contributors from OpenAI’s Operator agent and Microsoft’s MatterGen, an LLM focused on materials discovery.
The Vision
Periodic Labs is on a mission to automate the process of scientific discovery by creating AI scientists capable of running autonomous labs. These labs use robots to physically conduct experiments, be it mixing, heating, or analyzing materials, then collect data, learn from it, improve, and iterate continuously.
Unlike traditional AI which relies heavily on internet-trained models, Periodic Labs generates entirely novel, real-world datasets through hands-on experiments, enabling a new frontier of AI learning and discovery.
At the core of their current efforts is the quest to invent new superconductors that could surpass existing ones in performance and energy efficiency. Superconductors hold promise for transforming energy grids, transport, and computing.
Periodic Labs also aims to uncover a broad range of novel materials that could impact industries ranging from aerospace to semiconductors, pushing scientific boundaries further with its automated approach.
Data as the New Fuel
Beyond discovering new materials, Periodic Labs seeks to establish a vast reservoir of experimental data that other AI models can consume and evolve from. This contrasts with the current generation of AI which largely mines data from static sources like the internet, a resource now considered “exhausted” for groundbreaking discoveries.
“AI models have exhausted the internet’s finite troves of text and code. The next breakthrough requires giving AI the means to generate new knowledge itself—by conjecturing, experimenting, and learning, as human scientists do,” said Fedus.
By continuously generating high-quality experimental data, Periodic Labs hopes to fuel sustained AI development and scientific breakthroughs.
Though Periodic Labs has one of the most well-funded and experienced teams in the space, it is part of a wider movement. Other startups like Tetsuwan Scientific, as well as nonprofits such as Future House and academic consortiums including the University of Toronto’s Acceleration Consortium, are actively exploring similar AI-driven scientific automation. However, Periodic Labs’ deep expertise and substantial capital put it in a strong position to lead this effort.
While pioneering projects like DeepMind’s GNoME and Berkeley’s A-Lab have demonstrated the promise of AI-suggested materials, translating these into reliable commercial successes is an ongoing battle. Periodic Labs is betting that its closed-loop robotic labs will avoid hype cycles by delivering consistent, verified results of industrial relevance.
