Stratechery Amazon's Durability
114 snips
May 5, 2026 Amazon’s latest logistics move sparks a bigger look at how it turns internal tools into massive businesses. The conversation explores cloud cost advantages, why AI training once favored rivals, and how inference may swing momentum back. It also touches on Bedrock, Anthropic, satellites, drones, and long-term infrastructure bets.
AI Snips
Chapters
Transcript
Episode notes
Amazon Keeps Reusing The AWS Playbook
- Amazon repeatedly builds infrastructure for itself first, then sells it to others to deepen scale advantages.
- Ben Thompson links AWS and logistics through the same playbook: turn marginal costs into capital costs, then monetize excess capacity over a decade.
AWS Built Its Moat Through Structural Cost
- AWS won cloud by treating it as a commodity market where durable profits come from structurally lower costs.
- Nitro raised server utilization, and Graviton first paid off inside managed services like RDS before becoming competitive as a direct compute offering.
Why AWS Looked Weak In Early AI
- AWS looked vulnerable in 2023 because AI demand centered on training, which favored tightly networked NVIDIA systems.
- SemiAnalysis argued Amazon's Nitro, EFA, and in-house chips could reduce NVIDIA priority and leave AWS short of the best GPU clusters.