Periodic Labs

AI scientists plus autonomous physical labs. Robots run thousands of physics and chemistry experiments, and the experiment data (including negative results) trains models that design the next ones. The first target is higher-temperature superconductors. It also covers semiconductors R&D.

Founded: 2025
Headquarters: San Francisco Bay Area, CA
Founders: Liam Fedus, Ekin Dogus Cubuk
Pushes against: Training science models only on published literature and simulation.
Now: Announced Periodic Neon (2026-09-15), a 1T-parameter model post-trained on its own lab data to analyze X-ray diffraction results and deployed in its Menlo Park labs; it says Neon beats GPT-6 Astra and Claude Fable 5.1 on that task (company claim). No new-material discovery announced in the sources read. Reported (Forbes, 2026-05) to be raising $500M+ at $7.5B, led by AMP, with 20+ researchers hired from Meta, OpenAI and DeepMind. Launched an academic grant program.

Key people

Liam Fedus (co-founder; ex-OpenAI VP research)
Ekin Dogus Cubuk (co-founder; ex-DeepMind, GNoME)

Funding

May 2026 reported
Amount
>=$500M (reported, in progress)
Valuation
$7.5B (reported)
Lead
AMP (Anjney Midha's vehicle), reported
Source forbes.com.au
September 2025
Amount
$300M seed
Valuation
~$1.3B
Lead
a16z, with DST, NVentures, Accel, Felicis
Source forbes.com.au

Every funding round in the atlas

Launches

The Art of Scaling RL Compute (ScaleRL) Meta / UT Austin / UCL / Berkeley / Harvard / Periodic Labs · 15 October 2025
A 400,000-GPU-hour study finds RL performance follows predictable sigmoid compute curves, and publishes ScaleRL, a recipe validated to 100,000 GPU-hours.

People moves

Of the moves the atlas tracks, 1 joined and 0 left.

Liam Fedus September 2025
OpenAI to Periodic Labs
VP of research and post-training, OpenAI

Sources

This profile was checked and corrected against its sources on 6 October 2026. How we check