Ex-Amazon insider: AWS earns higher margins on open-weight tokens than Claude

RihardJarc · x · 2026-11-04

An ex-AWS employee (via AlphaSense) explains why existing workloads are hyperscalers' biggest AI advantage:

Started behind: AWS was seen as infrastructure-only and caught flat-footed by OpenAI and Microsoft; before adding Anthropic models it competed on compliance, reliability, scalability and GPU availability. 2024-25 were pilot years with revenue mostly from coding assistance; 2026 is the first year agentic workloads are scaling into production with meaningful revenue.
Model layer commoditized: customer evaluations shifted from "whose model is best" to "where do my data and workloads live" — a workload already on AWS is a major edge over Microsoft or Google.
Data security sells: AWS hosts Claude on its own GPUs on Anthropic's behalf, so customer data never leaves AWS — a key reason customers buy Claude via Bedrock rather than directly.
Capacity and cost: GPU demand outstrips supply, so installed capacity means no throttling; falling inference prices make open-weight models increasingly attractive.
Margin twist: the expert says AWS earns higher margins on open-weight tokens than Anthropic tokens, but doesn't pitch it that way — AWS plays the long game, using Claude on Bedrock as the entry point to win the workload.

More from Venture