Jonathan Cohen

Biology computing leader at NVIDIA. Presented at Session II on the compute stack coming for biology.

Key Arguments

Biology is accelerated computing’s next killer app
Each layer — instruments, predictive models, agents, robotic labs — multiplies (not adds to) the compute load
The future isn’t a generic AI scientist — it’s specialized sub-agents wired to tools, orchestrated by a generic reasoning agent
Software vendors will ship agents that know how to drive their tools — interface shifts from human-to-software to human-to-agent-to-software
Predictive biology needs pre-competitive consortia — the Protein Data Bank is the precedent
Models interpolate well, extrapolate poorly — diversity of training data is the constraint

Parabricks — 30× whole-genome assembly: ~half a day on CPU → ~10 minutes on GPU
CodonFM — mRNA foundation model tokenized at the codon level (~130M sequences, ~5B parameters)

“Skeptics about agents in biology should talk to any software engineer.”