Amazon Web Services delivered its strongest quarterly expansion in five years during the June period, propelled by surging demand for artificial intelligence infrastructure and custom silicon. Revenue reached $42.23 billion, exceeding analyst forecasts by nearly $1.7 billion and accelerating from the prior quarter’s 28 percent pace. The unit’s AI services and proprietary chips each crossed a $25 billion annualized run-rate threshold, more than doubling year-over-year.
This performance underscores a structural shift: cloud providers are no longer merely hosting workloads but are becoming the primary factories for generative AI. The scale of capital commitments and the breadth of new technical capabilities released alongside the earnings reveal how deeply AWS is embedding itself in the AI stack.
Record Growth Driven by AI Workloads and Custom Silicon
AWS operating income climbed to $16.62 billion, producing a 36.8 percent margin and accounting for 61 percent of Amazon’s total operating profit. Chief Executive Andy Jassy highlighted that the company added $4.6 billion in sequential revenue—an increase roughly 80 percent larger than any prior quarter—while the backlog expanded to $496 billion at triple-digit growth. These figures indicate that demand for both training and inference capacity continues to outpace supply, with 2027 capacity already largely committed into 2028.
The custom-chip business, centered on Graviton and Trainium/Inferentia instances, is expanding at triple-digit rates. A three-year agreement with Meta to deploy hundreds of thousands of Graviton processors illustrates how even hyperscale customers are turning to AWS silicon to reduce costs and improve performance at massive scale. The combination of general-purpose cloud growth and specialized AI accelerators has created a self-reinforcing cycle in which higher utilization funds further silicon development.
Capital Spending and the Infrastructure Arms Race
Second-quarter capital expenditures jumped 68 percent to $54.21 billion, surpassing consensus estimates. Amazon raised its full-year guidance to approximately $220 billion, signaling that the company intends to sustain elevated investment levels. Wall Street has largely accepted the increased spend because AWS margins are expanding even as capacity grows, a sign that utilization and pricing power remain strong.
The spending is concentrated on GPU-dense clusters and high-bandwidth networking required for frontier models. Jassy noted that customers increasingly want AI inference to run close to their existing applications and data, favoring providers with both breadth of services and geographic reach. This preference helps explain why AWS continues to hold a larger overall cloud revenue base—$148.4 billion annualized—than either Microsoft Azure or Google Cloud despite posting slower percentage growth than its two main rivals.
Hosting Frontier-Scale Open Models on Managed Infrastructure
AWS published detailed guidance for running Moonshot AI’s Kimi K3, a 2.8-trillion-parameter Mixture-of-Experts model released in late July 2026. The architecture activates only 104 billion parameters per token across 896 experts, delivering a 2.5-times improvement in scaling efficiency over its predecessor. Customers can deploy the model on SageMaker HyperPod or Amazon EKS using vLLM containers optimized for the MXFP4 weight format.
The availability of such guidance on day one of a frontier open-weight release demonstrates how AWS is positioning its managed services as the default production environment for the largest models. Organizations gain access to enterprise-grade orchestration, security controls, and cost-optimization features without having to build bespoke GPU clusters from scratch. This lowers the barrier for enterprises that want frontier intelligence but cannot justify the capital or operational overhead of self-managed supercomputing.
Security Controls for Agentic Development and Operations
Alongside infrastructure offerings, AWS released frameworks addressing the security implications of AI coding agents and autonomous operations tools. The control framework separates author-time guardrails—such as least-privilege access for agents exposed to untrusted content—from build-time verification gates that prevent risky code from reaching production. A parallel integration between AWS DevOps Agent and Wiz security graph allows on-call engineers to receive vulnerability context during incident investigations, distinguishing operational anomalies from potential security events in real time.
These controls reflect recognition that AI agents expand the attack surface faster than traditional application-security processes can accommodate. By embedding policy enforcement and security telemetry directly into the agent workflow, AWS aims to preserve developer velocity while maintaining enterprise risk standards.
Expanding Use Cases Across Advertising, Retention, and Cost Optimization
Yahoo has integrated Amazon Bedrock into its Demand-Side Platform to improve search retargeting, replacing an older Word2Vec-plus-LSH pipeline with more semantically aware embeddings. The change addresses stale vocabulary and limited phrase-level understanding that previously constrained audience expansion. Meanwhile, new capabilities in Amazon Quick and EMR Advanced Managed Scaling give customers finer control over both customer-retention automation and big-data cluster economics, allowing them to prioritize cost or performance through a single utilization slider.
These releases illustrate how AWS is translating its AI infrastructure advantage into vertical solutions that generate additional high-margin workloads. Each new capability increases customer stickiness while widening the moat around the core cloud platform.
The convergence of accelerating revenue, sustained capital intensity, and rapid product innovation positions AWS at the center of enterprise AI adoption. As capacity constraints persist into 2028 and frontier models continue to grow in size, the providers that can reliably supply both raw compute and production-grade governance tooling will capture disproportionate value. The coming quarters will test whether AWS can maintain margin expansion while meeting the unprecedented infrastructure demand its own success has helped create.
