AI scientists and autonomous labs — automating physics, chemistry, and materials discovery
HQ: San Francisco, CA, United States | Founded: 2025 | Employees: ~50-100 (est.) | Stage: Early Stage VC ($1.3B seed valuation; ~$7B talks reported) | Website: https://periodic.com
Periodic Labs emerged from stealth on September 30, 2025 with an unusual founding pedigree: Liam Fedus, OpenAI's VP of Research who led post-training for ChatGPT (and the GPT-4 era post-training organization), left OpenAI to co-found the company with Ekin Dogus Cubuk, a Google Brain/DeepMind research scientist who specialized in applying machine learning to materials science and physics. Their thesis: the next frontier of AI isn't more chat — it's science, where models learn from physical experiments rather than internet text, and where breakthroughs (new superconductors, better batteries, novel materials) create trillions in value.
The company's approach pairs large-scale AI with autonomous robotic laboratories: AI systems propose experiments, robotic labs run them physically, and the results — success or failure — feed back into the models. This closes the loop that pure-computational AI-for-science efforts leave open: reality is the ultimate verifier. The founding team recruited scientists and engineers from OpenAI, DeepMind, and Meta's hardware teams.
Investors funded the vision at extraordinary speed: a $300M seed round led by Andreessen Horowitz with Felicis and Accel participating, at a $1.3B valuation — one of the largest seed rounds in venture history — followed within six months by Bloomberg-reported talks to raise hundreds of millions more at around $7 billion. The company's stated first target: discovering new superconductors, a decades-old grand challenge in physics with transformative industrial implications.
Periodic Labs is the purest 'AI for science' bet yet funded: rather than another chatbot, it pairs frontier-scale ML with robotic wet labs and autoclaves, feeding real experimental data back into models — closing the loop between hypothesis and physical experiment. Its $300M seed (led by a16z, Sept 2025, at $1.3B) was among the largest ever, staffed from OpenAI, DeepMind, and Meta hardware teams, and within six months Bloomberg reported deal talks at ~$7B — a fivefold jump. The bull case: science is the ultimate RL environment (physical reality is a perfect verifier), and materials/superconductor discovery is worth trillions. The bear case: it is a brand-new company with no shipped product, competing for a tiny pool of people who can do both frontier ML and physical-science automation, against physics itself — experiments are slow, and AI-for-science has overpromised before.
Category: AI & Machine Learning Startups
Browse: Home | Blog | About | All Startups A-Z | AI & Machine Learning Startups | View full profile on StartupWiki