Periodic Labs has officially emerged from its quiet development phase, announcing a monumental $300 million seed funding round. This substantial investment was spearheaded by a roster of technology industry heavyweights, including Andreessen Horowitz, DST, Nvidia, Accel, Elad Gil, Jeff Dean, Eric Schmidt, and Jeff Bezos, signaling strong confidence in the startup's ambitious vision.
The company was co-founded by two distinguished figures in the AI and materials science domains: Ekin Dogus Cubuk and Liam Fedus. Cubuk previously headed the materials and chemistry division at Google Brain and DeepMind, where his groundbreaking work included the AI tool GNoME. In 2023 alone, GNoME was responsible for unearthing over two million new crystalline structures, holding immense potential for future technological advancements. Fedus, a former VP of Research at OpenAI, played a pivotal role in the creation of ChatGPT and led the development of the first neural network to achieve a trillion parameters.
Periodic Labs' modest yet elite team is composed of researchers who have contributed to other significant AI and materials science projects, such as building OpenAI’s agent Operator and working on Microsoft’s MatterGen, an LLM specifically designed for materials discovery. Their collective mission is nothing less than revolutionizing scientific inquiry by automating the discovery process itself, envisioning a future where "AI scientists" lead the charge. This involves establishing sophisticated autonomous laboratories where robotic systems conduct physical experiments, meticulously gather data, and continually refine their approaches through iterative learning cycles.
The startup's initial scientific objective is to engineer novel superconductors, aiming for materials that offer superior performance and consume less energy than current options. Beyond this primary focus, Periodic Labs is committed to identifying and developing an array of other innovative materials. Crucially, a parallel goal involves systematically collecting the vast amounts of physical world data generated by their AI scientists as they manipulate and process various raw ingredients in their quest for new discoveries.
Periodic Labs argues that the traditional approach of training scientific AI models on existing internet data has reached its saturation point. As stated in their introductory blog post, large language models (LLMs) have "exhausted" the internet as a viable data source. By contrast, Periodic Labs is "building AI scientists and the autonomous laboratories for them to operate," thereby producing invaluable, fresh experimental data that will be essential for the continued evolution and advancement of AI models.
While Periodic Labs represents an exceptionally strong convergence of talent and capital dedicated to this cause, it is not alone in pursuing the automation of science through AI. The concept of AI-driven chemical discovery has been a growing area of academic research since at least 2023, attracting interest from agile startups like Tetsuwan Scientific, non-profit organizations such as Future House, and university initiatives like the University of Toronto’s Acceleration Consortium.
---
Originally published at: https://techcrunch.com/2025/09/30/former-openai-and-deepmind-researchers-raise-whopping-300m-seed-to-automate-science/
