How AWS and Anthropic Are Quietly Rewriting Cloud Profit Economics
Benzinga reported Wednesday that a structural quirk inside Amazon‘s cloud-AI partnership with Anthropic is quietly compounding AWS Bedrock margins in ways rivals have struggled to replicate.
The analysis, drawing on research from chip and infrastructure firm Semianalysis, argues the gap between AWS and competing cloud providers is widening. The driver is not raw server capacity. It is the commercial architecture underpinning how Claude tokens are sold.
How the Bedrock Deal Structure Works
Under the current arrangement, Anthropic acts as the seller of record for Claude tokens moving through AWS Bedrock. That means Amazon collects an infrastructure fee on the compute side and a separate distribution revenue share on the model side simultaneously.
The result is a layered revenue stream where Amazon earns twice per transaction without bearing any model development cost. Semianalysis describes it as functioning like a high-margin royalty channel stacked on top of existing compute revenue.
Few commercial AI distribution arrangements in enterprise technology history have replicated that kind of structural asymmetry.
Volume Is Now Large Enough to Matter
The tokens flowing through this arrangement have scaled rapidly. Anthropic’s annualised revenue surpassed $30 billion as of April 2026, up from roughly $9 billion at the close of last year. More than 100,000 enterprise customers now run Claude through Amazon Bedrock, per Anthropic’s own disclosure tied to an expanded AWS compute agreement.
Critically, Anthropic’s inference gross margins have climbed into the mid-60s, sharply higher than the 38% recorded in 2025 and well above the negative readings posted in 2024. Healthier inference margins mean more profitable tokens cycling through the Bedrock pipeline, which feeds directly into Amazon’s revenue share upside.
Background: Cloud as an AI Distribution Business
Cloud computing began as a commodity infrastructure play. The prevailing logic was that scale and price reductions would define the competitive order over time.
That dynamic has shifted materially. The platforms controlling distribution of frontier AI models are accruing margin advantages that traditional infrastructure economics do not fully explain. AWS posted $37.6 billion in Q1 2026 revenue, up 28% year over year. Its EBIT margin expanded by 213 basis points in a single quarter, a move the Semianalysis report links directly to the Claude arrangement.
What Investors Are Watching
Amazon is compounding two advantages at once. The Bedrock distribution deal provides a high-margin revenue layer. Separately, Amazon’s custom silicon development lowers the underlying cost of delivering AI inference at scale.
Neither edge is straightforward for Microsoft Azure or Google Cloud to neutralise quickly. The competitive moat, Benzinga noted, is structural rather than cyclical.
Read Next: AI Infrastructure Spending Hits Record as Hyperscalers Race to Deploy Capacity
