Nvidia has agreed to deliver approximately 1 million GPUs to Amazon Web Services through the end of 2027, marking one of the largest single cloud-infrastructure orders for AI accelerators. Deployment will span AWS’s global cloud regions and include collaboration on networking and rack architectures designed for agentic AI workflows capable of autonomous reasoning and planning.

The GPUs supplied under this deal are optimized for inference tasks, which now represent roughly two-thirds of total AI compute demand compared to about one-third in 2023. Industry forecasts estimate the inference-focused chip market will surpass $50 billion by 2026, driven by growing live-service deployments and real-time model execution.

Beyond chips, Nvidia is deepening its role as a foundational infrastructure provider by integrating networking technology and rack designs into AWS systems, enabling customers to mix Nvidia accelerators with AWS’s own silicon within the same environment. This flexible approach differentiates AWS from competitors with closed architectures and cements Nvidia’s position beneath major cloud platforms.