Periodic Labs raised $300 million in seed funding to build an A.I. scientist that works with real lab experiments. The San Francisco startup will start in the physical sciences and says it will pair models with hands-on research. According to Observer, the company does not plan to make consumer products or chase AGI.
Backers, founders, and the research plan
The round was led by Andreessen Horowitz and included Nvidia, Felicis, Accel, Elad Gil, Jeff Dean, Jeff Bezos, and Eric Schmidt. Co-founder Liam Fedus said the goal is an A.I. scientist that learns beyond the internet. He wrote that the web is vast but finite, so labs must supply new data.
Fedus co-created ChatGPT and was vice president of research at OpenAI. He launched the company with Ekin Dogus Cubuk, a longtime Google DeepMind scientist who led its materials science and chemistry team. The roughly 20-person founding team includes veterans from OpenAI, Google, Meta, and Microsoft projects.
Peter Deng of Felicis said the systems should form hypotheses and run simulations in autonomous labs. He decided to invest after hearing that to do science, A.I. must do real science. Periodic Labs will train agents to streamline research and engineering tasks as well.
Focus areas and early work
Physical sciences first, with materials design
The company will focus on fields with clear data and testable results. It aims to automate parts of materials design that can support faster progress in space, nuclear fusion, and Moore’s Law. The team says that real experiments can push A.I. past text-only learning.
Periodic Labs has started work with semiconductor makers on chip heat dissipation. It is also building agents for research workflows and counts space and defense companies as customers. The company’s approach blends simulations, lab runs, and model training.
Periodic Labs is not alone in this push. OpenAI and Google are pursuing similar goals, said Observer. DeepMind’s AlphaFold helped transform protein structure prediction and earned two researchers a Nobel Prize last year. In San Francisco, the nonprofit FutureHouse is also working on an autonomous A.I. scientist.
The startup also highlights culture. Each staffer picks a periodic table element for a custom desk plate. The team says it wants to bring rigor from science into fast A.I. development, then ship tools that researchers can use in real labs.
