06-reference

stratechery amazon earnings capex commodity ai

Mon Feb 09 2026 19:00:00 GMT-0500 (Eastern Standard Time) ·reference ·source: Stratechery ·by Ben Thompson
amazonawsearningscapexcommodity-aitrainiumcloud

Amazon Earnings, CapEx Concerns, Commodity AI

Thompson is less bullish on Amazon’s $200B CapEx plan than Google’s, despite similar demand-exceeds-supply narratives. AWS grew 24% on a $142B annualized run rate with backlog of $244B (up 40% YoY), and margins expanded to 35%.

Three Amazon-specific concerns: (1) Amazon operates at much lower margins than other hyperscalers — AWS is the high-margin business, unlike Microsoft/Google where cloud is funded by higher-margin software/ads. (2) Amazon’s projected CapEx exceeds its operating cash flow ($200B vs. ~$178B estimated), meaning it will burn cash — a Rubicon crossing for the industry where the “funded by free cash flow” defense no longer holds. (3) Amazon’s advertising business ($21.3B, up 22%) depends on customers starting shopping journeys on Amazon, which AI shopping agents could disrupt.

The bull case centers on Amazon’s Trainium chip strategy: if AI compute commoditizes, lowest-cost-structure wins, and Trainium under Bedrock gives Amazon both lower customer pricing and better economics. Additionally, as a purer cloud play than Microsoft or Google (who prioritize internal workloads), AWS can more credibly claim third-party customers come first.

Thompson notes the broader theme: Big Tech’s collective $700B+ CapEx for 2026 — approaching Japan’s national budget — is forcing companies into debt issuances, a meaningful shift from the self-funded era.

RDCO Mapping

The commodity AI thesis is relevant for infrastructure planning — if inference costs normalize, the economics of AI-powered services improve significantly for everyone, including RDCO.