Discovery is the bottleneck — AI that finds new materials faster will unlock the value.

Why the market believes it

GNoME (2.2M structures), MatterGen MIT-licensed, ARPA-E CATALCHEM-E + DARPA Crystal Palace all presume more candidates = upstream value.

Breaking conditions

Customer voice: Cubuk: "we won't reason our way into a much better superconductor."
Data: 80%+ disorder in AI-predicted; A-Lab Nature correction Jan 2026; 83K GNoME entries silently removed.
Track record: zero commercial AI-discovered materials at volume from Periodic, Lila, CuspAI, Radical AI.
Open-source erosion: MatterGen MIT-licensed Dec 2025.

Winners & losers if it breaks

Winners

Matter Loop's category — applied-AI reformulation shops
Contract-lab networks
Atinary, Cubuk's contract-R&D pivot

Losers

$1B-priced discovery startups (Periodic, Lila, CuspAI at Series-C)
Anyone whose pitch is 'we generated N novel structures'

What happens when it breaks

Discovery-platform valuations compress 60–80%. Capital rotates out of generative-model labs and into integrators with synthesis + cert relationships.

Trigger

2026–2027 capital-call cycle for the Periodic / Lila / CuspAI cohort. Series-C reality without commercial volume forces a public re-rating. Matter Loop must be in position as "destination of distressed capital + distressed IP" by then.

All assumptions →