Stanford's 9-lecture AI economics course condensed: which layer of the stack gets the money?

le_james94 · x · 2026-09-26

James Le recaps all nine lectures of Stanford MS&E 435, 'Economics of the AI Supercycle,' which brings in one practitioner per week from each layer of Jensen Huang's five-layer stack: energy (Constellation, GE Vernova), chips (NVIDIA, Broadcom), infrastructure (AWS, CoreWeave, Crusoe), models (OpenAI, Anthropic), and applications (Cursor, Harvey, Waymo). Asked where he'd put $100, Databricks CEO Ali Ghodsi picked early-stage applications.

Topics span generative AI economics, the GPU economy, gigawatt-scale AI factories, enterprise software, frontier-lab compute strategy, inference as COGS, coding agents, and AI in life sciences — where Chai-2 hits 16% on de novo antibody design (20 designs per target, one round, wet-lab verified) against 30 of 10,000 possible drug targets reaching clinic yearly. The author notes an unresolved tension: week 2 asks who receives AI's wealth, week 5 aims to make humans the bottleneck.

Original post →

More from AGI Musings

AGI Musings channel →