AI drug discovery hits its HBM moment as the bottleneck shifts to the wet lab

QuanquanGu · x · 2026-11-11

An analysis of the AI drug discovery landscape: like chips moving from GPUs to HBM to power, the bottleneck is shifting from design to the wet lab.

Designing candidates is nearly free: Claude designed 1,320 protein binders, Adaptyv Bio tested all, 354 bound; the 966 failures became training data
The loop: model → design → DNA/protein → assay → better model. Fastest loop wins
Money is moving: Isomorphic in talks at up to $50B; Biohub, Meta, Google DeepMind and US agencies committed $1.8B to the data layer; Insilico's AI-designed drug is in Phase III
Quanquan Gu left ByteDance Seed (where he built SeedFold) in June; four months later his lab Geodesic Intelligence shipped NovaAtom, all-atom structures with calibrated confidence, live as an API

More from Venture