Inference demand growth → TaaS mix shift → AWS paid twice (infra + rev share) → +213bps EBIT expansion → AI capex ROI improves

As AI inference demand scales, AWS structurally gains over Azure/GCP because Anthropic (via Bedrock) generates Token-as-a-Service (TaaS) revenue that pays AWS both an infrastructure fee AND a revenue share — earning the platform twice versus IaaS models. Bedrock grew 170% Q/Q in Q1 2026, driving +213bps EBIT margin expansion for AWS Q/Q, with Trainium custom silicon widening the cost moat. This mechanism explains why AI capex returns are improving rather than declining, directly supporting the sustainability of the AI capex cycle.

• Step 2: Anthropic renegotiates Bedrock deal to remove AWS revenue share, or migrates significant inference volume to Azure/GCP — TaaS double-margin disappears.
• Step 3: Microsoft/Google develop equivalent custom silicon cost advantages at scale — Trainium moat narrows and EBIT convergence follows.
• Step 4: AWS EBIT margins stop expanding in Q2/Q3 2026 despite Bedrock growth — indicates the +213bps was non-recurring or offset by other cost increases.

• Azure/GCP have the same theoretical ability to adopt TaaS models (Azure OpenAI Service, GCP Vertex AI) — the mechanism assumes AWS maintains structural advantage via Anthropic exclusivity and Trainium. If Microsoft deepens OpenAI exclusivity on TaaS terms, the structural gap narrows.
• Anthropic's token share on Bedrock is captive by contract, but Anthropic is building relationships with Azure too. Long-term, Anthropic may distribute across clouds, diluting AWS's sole-beneficiary position.
• The +213bps figure is from a high-growth base (Q1 2025 was very early in Bedrock scaling). Future quarters face tougher comps; the bps expansion rate will slow even if absolute dollars continue growing.

• **AMZN (AWS) is the primary tradeable:** TaaS margin mix at 55% EBIT vs. traditional cloud margins creates a durable EBIT margin expansion tailwind. If Bedrock reaches 20%+ of AWS AI revenue (from 37% of a 10% slice today), the mix impact compounds.
• **AI capex ROI justification:** The mechanism directly addresses the bear case that hyperscaler AI capex has no visible ROI. Bedrock at 55% EBIT margin provides a concrete return signal — the capex cycle is generating high-margin revenue, not just displacement of lower-margin workloads.
• **Trainium moat durability:** Custom silicon cost advantage widens as Anthropic's inference volume scales (inference economics improve at scale) — the margin gap vs. Azure/GCP may widen, not close.
• **Second-order: Anthropic (private)** — if Anthropic's inference margins are mid-60s% and growing, the private mark matters for AI-infrastructure investors. Anthropic's $30B ARR on a high-margin inference basis strengthens the case for continued AI infrastructure investment by AWS.
• **Caution:** Source is a paid SemiAnalysis preview; full implications section paywalled. The +213bps EBIT figure is sourced but AWS doesn't formally break out Bedrock. Treat as an analyst-computed estimate, not a directly disclosed metric.