The brief
- One of the biggest seed rounds ever: $300M at about $1.3B, led by Andreessen Horowitz with NVIDIA, Accel, Felicis and DST.
- Reported in March 2026 to be in talks to raise ~$500M at ~$7B; not confirmed as of mid-2026.
- Its bet: science AI needs its own lab data, so it builds labs to make that data instead of only reading papers.
Technical approach
Loop
AI proposes, robots test
AI suggests materials and recipes, robot labs make and measure them, and the results train the AI for the next round.
Data
Making new data
Published data is sparse and mostly shows successes. Its labs record everything, including failures, which helps the AI learn.
Targets
Superconductors first
An early target, because one clear number (the temperature at which they start working) makes progress easy to check.
Programs
Periodic Labs, one program
Updated 25d agoResearch · Proto · Pilot · Deploy · Scaling
Open technical problems
- Making it is harder than predicting itActive research
AI can predict millions of materials, but making them pure and right is slow and often fails; automating that is the hard part.
- Proving discoveries are realOpen
Past AI material claims were criticised as not new or badly measured; independent checks are essential.
Peers
Company & funding
- Valuation
- $1.3Bprivate
- Raised
- $300Mdisclosed private rounds
- Last round
- SeedSep 2025
- HQ
- US
- Founded
- 2025
| Date | Round | Amount | Post-money | Lead |
|---|---|---|---|---|
| Sep 2025 | Seed | $300M | $1.3B | Andreessen Horowitz |
Partners & customers
- Investor
- Investor
People
- Liam Fedus (opens en.wikipedia.org)Co-founder
- Ekin Dogus CubukCo-founder

