Stratechery Mythos, Muse, and the Opportunity Cost of Compute
274 snips
Apr 13, 2026 Discussion of aggregation theory and how compute shifts economics from marginal cost to opportunity cost. Examination of real-world GPU allocation decisions at major cloud providers. Exploration of competing narratives around advanced AI systems, access limits, and risks from model distillation. Contrast between companies prioritizing consumers versus controlling compute supply.
AI Snips
Chapters
Transcript
Episode notes
Why Aggregation Theory Depended On Zero Marginal Costs
- Aggregation theory relied on near-zero marginal costs enabled by digital goods and fixed infrastructure scale.
- Ben Thompson explains marginal vs fixed costs using a widget factory then shows digital outputs make electricity and humans effectively fixed costs in tech.
Microsoft Admitted GPUs Were Reserved For Its Own Products
- Microsoft chose to allocate newly installed GPUs to M365 and GitHub Copilot rather than Azure revenue growth.
- CFO Amy Hood described guidance as an "allocated capacity guide" and said internal use came before serving more Azure customers.
AI Turns Marginal Costs Into Opportunity Costs
- AI flips the problem from marginal cost to opportunity cost because compute used for one workload can't serve another.
- Citing Microsoft's Azure comments, Ben Thompson shows GPUs allocated to first-party apps reduce capacity to sell to customers.