Periodic Labs Powers Up for ‘Scientific AI Advances’

Periodic Labs has emerged with a goal to create an AI-powered scientist capable of running its own experiments and generating new knowledge.

The company’s vision centers on rethinking how science is done. Traditional AI systems have been trained on the internet’s vast — but finite — supply of text data, estimated at around 10 trillion tokens. Leading models have already exhausted this resource. Periodic’s founders argue that true scientific breakthroughs require more than intelligence: they demand interaction with reality.

“Autonomous labs are central to our strategy. They provide huge amounts of high-quality data (each experiment can produce GBs of data!) that exists nowhere else. They generate valuable negative results which are seldom published. But most importantly, they give our AI scientists the tools to act,” William Fedus, Co-Founder of Periodic Labs, said in an X post.

Fedus also explains that the startup is backed by investors Andreessen Horowitz (a16z), who led its $300 million funding round, as well as Felicis, DST Global, NVentures (Nvidia’s venture capital arm), Accel, and individuals including Jeff Bezos, Elad Gil, Eric Schmidt, and Jeff Dean.

Starting with physics and materials

By embedding AI directly into the scientific process, the company believes its systems can accelerate discovery across the physical sciences. The latter is where the startup is beginning its work. It reckons experiments are relatively fast, data-rich, and verifiable. Physics provides a natural testing ground for reinforcement learning approaches that have driven AI progress in fields like math and code.

One key target is developing new superconducting materials that can operate at higher temperatures. Success could pave the way for next-generation transportation, ultra-efficient power grids, and breakthroughs in materials science. More broadly, the company sees potential to accelerate Moore’s Law, nuclear fusion, and even space travel.

Periodic is already collaborating with industry partners. In one case, it is working with a semiconductor manufacturer struggling with heat dissipation in chips. The company is building custom AI agents to help engineers interpret experimental data and iterate faster on solutions.

Founders with proven track records

Periodic Labs is led by a team with deep roots in AI and scientific discovery. Its founders have co-created ChatGPT, DeepMind’s GNoME, OpenAI’s Operator (now Agent), the neural attention mechanism, and MatterGen. They have also scaled autonomous physics labs and contributed to some of the most significant materials discoveries of the past decade.

With this pedigree, Periodic aims to position itself at the forefront of a new era where AI not only analyzes data, but also generates the experiments that drive discovery forward.