Periodic Labs

Periodic Labs is a San Francisco-based artificial intelligence research company founded in 2025 that is building 'AI scientists' together with the autonomous laboratories in which those systems operate. Its stated premise is that internet-scale text data, estimated at roughly 10 trillion tokens, has been largely exhausted by frontier models, and that further progress in scientific AI requires generating new experimental data. Autonomous labs are central to the strategy because each experiment can produce gigabytes of data, including negative results that are rarely published, and because they give the models the ability to act rather than only predict.

The company is starting in the physical sciences, arguing that experiments there have high signal-to-noise ratios, are relatively fast, are well modeled by simulation, and are verifiable — framing nature as a reinforcement-learning environment. A near-term research goal is discovering superconductors that operate at higher temperatures than existing materials, with broader ambitions in semiconductors, energy grids, transportation, space travel, and nuclear fusion. In parallel, the company is deploying applied work with industry; it cites training custom agents for a semiconductor manufacturer struggling with chip heat dissipation.

The founders describe a convergence of three enabling developments as the reason for founding the company in 2025: robotic arms that had become reliable at powder synthesis, machine-learning simulations accurate and efficient enough to model complex physical systems, and large language models with strong reasoning capabilities. In this design, simulation proposes candidate compounds, robots synthesize them, and an LLM analyzes results and suggests course corrections.

Founding story

Periodic Labs began roughly seven months before its October 2025 press coverage, when Liam Fedus, then VP of Research at OpenAI and a co-creator of ChatGPT, discussed with former Google Brain colleague Ekin Dogus Cubuk whether generative AI could finally be applied to scientific discovery. Cubuk had led materials and chemistry research at Google Brain and DeepMind, contributed to the 2023 GNoME paper — the tool credited with discovering over two million new crystals — and co-authored 2023 work on a fully automated robotic lab that produced 41 novel compounds from language-model-suggested recipes. Fedus announced his OpenAI departure publicly on March 17, 2025, initially indicating an OpenAI partnership and investment; the founders later confirmed OpenAI did not invest. His announcement triggered heavy investor interest, and the first call taken was from Peter Deng of Felicis, who committed during a walk in San Francisco's Noe Valley before the company was even incorporated.

Business model

Periodic Labs builds full-stack systems combining AI models ('AI scientists') with autonomous laboratories that run physical experiments. Alongside internal materials-discovery research, the company deploys solutions with industrial partners, including training custom agents for a semiconductor manufacturer's engineers and researchers to analyze experimental data. Public sources do not describe pricing or contract structure; one aggregator labels the business model 'B2B SaaS' without supporting evidence.

Not disclosed in the sources. One aggregator lists a revenue estimate, but explicitly labels it a modeled estimate based on comparable companies rather than reported figures.

Traction

Traction to date consists primarily of capital and talent rather than disclosed commercial metrics: a $300 million seed round, a team of more than two dozen senior researchers listed publicly, at least one industrial deployment with a semiconductor manufacturer, and a scientific advisory board of five university professors. No revenue, customer count, or experimental throughput figures are disclosed in primary sources.

Latest developments

Following its September 30, 2025 launch and $300 million seed round, the company hired over two dozen senior AI and scientific researchers and stated that investor capital will fund team growth, lab scaling, and development of its first generation of AI scientists. It also announced an academic grant program. Internally, team members deliver weekly graduate-level lectures to one another to bridge AI and physical science expertise.