--- title: "Amazon's AI Turnaround: Selling Claude Tokens Propels AWS Profit Margins Past Microsoft and Google" type: "News" locale: "en" url: "https://longbridge.com/en/news/287883051.md" description: "SemiAnalysis believes that AWS is quietly widening its lead in the cloud computing profit war through a unique structural logic. The Bedrock platform accounts for only 4% of AWS's total revenue but contributes 30% of the year-over-year increase in gross profit, with an EBIT margin as high as 55%. The core logic behind this is a leap from the IaaS model of selling computing power to the TaaS model of securing model distribution rights. Anthropic's quarterly net new ARR reached $21 billion, and its API revenue surged 13-fold year-over-year, becoming the core fuel for this leveraged structure" datetime: "2026-05-28T08:31:17.000Z" locales: - [zh-CN](https://longbridge.com/zh-CN/news/287883051.md) - [en](https://longbridge.com/en/news/287883051.md) - [zh-HK](https://longbridge.com/zh-HK/news/287883051.md) --- # Amazon's AI Turnaround: Selling Claude Tokens Propels AWS Profit Margins Past Microsoft and Google In an era of broad-based growth in AI cloud services, profit margins have become the true dividing line. By leveraging a unique structure that converts demand for Claude tokens into operating leverage, AWS is pulling ahead of Azure and Google Cloud. The latest data shows that AWS's EBIT margin increased by 213 basis points quarter-over-quarter in the first quarter of 2026, while Azure's margin weakened during the same period, and Google Cloud's margin saw limited improvement with differences in accounting standards. **Amazon is the only cloud service provider (CSP) that has made Token-as-a-Service (TokenaaS) a major component of its artificial intelligence business.** Jeremie Eliahou Ontiveros, an analyst at research firm SemiAnalysis, attributes this divergence to three factors: "**AWS's larger share of third-party model API spending, the Anthropic/Bedrock transaction structure, and Anthropic's better-than-expected ARR in the first quarter of 2026.**" This is not a simple logic of "strong AI demand leads to good profits," but a structural shift from being a computing power lessor to becoming a model distribution platform. The key signal is that Bedrock, currently on a run-rate scale of approximately $5.5 billion, accounts for only about 4% of AWS's total revenue, yet it contributes 30% of AWS's year-over-year increase in gross profit. As long as demand from Anthropic continues to explode, this leverage effect will continue to amplify. ## Structural Differences: Lower AI Proportion, Higher Profit Margins Cloud vendors are all benefiting from AI demand, but the divergence is occurring in profit margins, not revenue growth rates. In terms of the proportion of AI revenue to total revenue, AWS is not leading. Estimates show that AWS's AI revenue share rose from 2% in the first quarter of 2024 to 10% in the first quarter of 2026, while GCP and Azure reached 36% and 27%, respectively, during the same period. However, a high AI proportion has not automatically translated into high profit margins. The AI businesses of Azure and GCP are still dominated by AI IaaS, accounting for more than 80% of their respective AI business portfolios. **AWS's structure is changing: Bedrock's share of AWS's AI revenue rose from 9% in the first quarter of 2025 to 37% in the first quarter of 2026.** This explains a superficial contradiction—AWS's AI revenue share is far lower than that of its competitors, yet its profit margins are outperforming. The issue is not "how much AI there is," but "what kind of AI revenue it is." ## The Bedrock Model: From Selling Computing Power to Securing Distribution Rights **Bedrock is AWS's model invocation platform, allowing customers to access cutting-edge large models like Claude through a unified billing and security compliance framework.** Its competitors include Microsoft Foundry, Google Gemini Enterprise Agent Platform, and platforms focused more on open-source models such as TogetherAI and Fireworks. The core difference among these platforms lies not in the number of models or latency metrics, but in the ability to access cutting-edge models. Cutting-edge large models contribute the majority of revenue in the AI API industry, and this is where AWS, Microsoft, and Google hold an advantage over other endpoint platforms. However, accessing models is just the first step. The true significance of Bedrock to AWS's profit margins lies in its transaction structure. Under the arrangement where AWS distributes Claude tokens via Bedrock, Anthropic, as the seller of record, recognizes the full token sales revenue; customers are invoiced by AWS, and the models are deployed on AWS infrastructure; AWS then receives two types of revenue: infrastructure fees similar to EC2/IaaS, and distribution or revenue sharing. **Compared to five-year take-or-pay IaaS contracts, this Token-as-a-Service (TaaS) business has lower revenue lock-in but thicker profit margins.** Estimates of the Anthropic/Bedrock arrangement show that the combination of fixed IaaS fees, revenue sharing, and excess performance thresholds enabled Bedrock to achieve an EBIT margin of approximately 55% in the first quarter of 2026. The cost is also clear: if Claude token consumption declines, AWS bears higher demand risk than in traditional IaaS. Currently, the TaaS businesses of Amazon, Microsoft, and Google have all reached the multi-billion dollar ARR scale, while Oracle and neoclouds have almost no scale at this layer—this is the key factor widening the gap between hyperscale cloud vendors and other AI computing providers. ## Anthropic's Explosion: The Core Fuel for AWS's Profit Leverage Bedrock is highly tied to Anthropic's demand. **Estimates show that more than 80% to 90% of Bedrock customers use Anthropic models, meaning Bedrock is essentially a business driven by Claude demand.** **** Anthropic's own growth data is extremely outstanding. Its net new ARR in the first quarter of 2026 was $21 billion, bringing total ARR to $30 billion; API revenue grew approximately 13-fold year-over-year, and year-end ARR could far exceed $100 billion. The rapid rollout of Claude Code among enterprise customers is a major driver, and consumer-side traffic is also beginning to migrate to Claude. Profit margins have also improved significantly. Anthropic's inference gross margin has risen to the mid-60% range, representing a substantial recovery from 38% in 2025 and -94% in 2024. The faster Anthropic grows, the greater the consumption of Claude tokens on Bedrock, and the more infrastructure fees and distribution shares AWS collects. This symbiotic relationship is already reflected in financial data. The path assumptions for the second quarter of 2026 are more aggressive: Bedrock's share of AWS's AI revenue is expected to rise to 53%, contributing an additional 9 percentage points to AWS's total revenue growth. ## Capacity Layout: Locking in Power Early to Meet TaaS Demand For TaaS to scale, the prerequisite is having sufficient inference computing power delivered on time. AWS is more aggressive on this front than most peers. **Data center models show that AWS continues to lead in new capacity additions from 2025 to 2027; Microsoft's pace was close from 2024 to 2026, but it falls significantly behind by 2027. More importantly, Microsoft's internal AI projects consume more computing power than Amazon's, and a large amount of AI computing power is locked into long-term contracts with OpenAI—the backlog related to OpenAI alone is 2.5 times Azure's annual revenue.** AWS treated electricity and capacity as a market share issue earlier, signing billions of dollars worth of PPAs with independent power producers such as Talen, Vistra, and NiSource, and advancing construction nearing 2GW in scale in Indiana and Mississippi. Microsoft previously experienced a roughly one-year pause in data center construction, which lowered its 2027 capacity forecasts; progress on its large AI cluster in Wisconsin is also slower than comparable AWS projects. To catch up, Microsoft can only purchase more capacity from neoclouds, which comes at a higher cost and will pressure profit margins. AWS is also promoting new data center designs with higher proportions of modularity and prefabrication. For AI inference business, this directly relates to revenue delivery capability. ## In-House Chips: Lowering Underlying Costs Beyond Distribution Fees **The Bedrock model is naturally friendly to in-house chips—customers are buying tokens and do not care whether the underlying hardware is NVIDIA GPUs or Trainium. This gives AWS an additional cost lever.** Trainium offers good performance/total cost of ownership in workloads sensitive to memory bandwidth, such as high-batch inference and reinforcement learning. AWS CEO Matt Garman disclosed in November 2025 that Trainium already supports more than 50% of Amazon Bedrock token usage. The CPU side is also worth noting. Demand for CPUs in training and inference for cutting-edge large models is rising, especially in reinforcement learning and agentic workloads. AWS's Graviton4 and Graviton5 offer advantages in performance/total cost of ownership and will be integrated as head nodes for Trn3, or used independently for reinforcement learning and agentic tasks. AWS has signed large-scale CPU and Graviton-related collaborations with Anthropic, OpenAI, and Meta; the larger the Bedrock customer base, the lower the friction in upselling Graviton capabilities to them. ## Peers Lagging: Not Insufficient AI Revenue, But Unchanged Structure The core problem for Azure is that its AI revenue remains highly IaaS-oriented, and Microsoft 365 Copilot and GitHub-related businesses have not yet formed a profit margin boost of comparable scale. Google Cloud's Gemini API performance is decent, but it has not replicated Anthropic's dominance in the coding market. More importantly, Google Cloud has differences in accounting standards—DeepMind's training costs are not included in the GCP segment, making its profit margins not fully comparable to AWS. The situation for Oracle and neoclouds is more direct: they mainly compete in the AI IaaS and computing power leasing layers, with almost no scale in TaaS. Once cloud business profits fall below expectations, the fragility of the wholesale computing power model is immediately exposed. AWS's outperformance in this round relies on several lines holding simultaneously: Anthropic's demand explosion provides the revenue base, the Bedrock transaction structure provides profit margins, electricity and data center capacity provide delivery capability, and Trainium and Graviton lower underlying costs. As long as these lines remain connected, AWS's AI business logic is not just "capital expenditure for growth," but "model demand for operating leverage." ### Related Stocks - [AMZN.US](https://longbridge.com/en/quote/AMZN.US.md) - [AMZU.US](https://longbridge.com/en/quote/AMZU.US.md) - [MSFT.US](https://longbridge.com/en/quote/MSFT.US.md) - [GOOGL.US](https://longbridge.com/en/quote/GOOGL.US.md) - [GOOG.US](https://longbridge.com/en/quote/GOOG.US.md) - [ORCL.US](https://longbridge.com/en/quote/ORCL.US.md) - [OpenAI.NA](https://longbridge.com/en/quote/OpenAI.NA.md) - [VST.US](https://longbridge.com/en/quote/VST.US.md) - [NI.US](https://longbridge.com/en/quote/NI.US.md) - [META.US](https://longbridge.com/en/quote/META.US.md) - [ORCL-D.US](https://longbridge.com/en/quote/ORCL-D.US.md) ## Related News & Research - [How Claude On AWS Bedrock Is Winning The Cloud Margin War](https://longbridge.com/en/news/287825920.md) - [AWS Margins Are Leaving Cloud Rivals Behind as Anthropic Demand Surges](https://longbridge.com/en/news/287812649.md) - [Report AWS: adozione dell’AI in crescita del 33%, ma il gap con l'Europa rischia di allargarsi](https://longbridge.com/en/news/287736235.md) - [Inside startups, Claude has already won the AI coding wars](https://longbridge.com/en/news/287425462.md) - [Snowflake jumps over 30% on earnings beat, AWS deal, AI push](https://longbridge.com/en/news/287827994.md)