AWS and NVIDIA deepen strategic collaboration to accelerate AI from pilot to production
AWS will deploy over 1 million NVIDIA GPUs starting in 2026, including new Blackwell chips.
At NVIDIA GTC 2026, AWS and NVIDIA announced a significant expansion of their strategic collaboration to accelerate enterprise AI adoption from pilot to production. The centerpiece is AWS's commitment to deploy more than 1 million NVIDIA GPUs across its global regions starting in 2026, incorporating both the new Blackwell and upcoming Rubin GPU architectures. This massive compute expansion is designed to meet the growing demand for running reliable, scalable AI systems that meet strict security and compliance requirements.
A key technical announcement is AWS becoming the first major cloud provider to offer Amazon EC2 instances powered by NVIDIA's new RTX PRO 4500 Blackwell Server Edition GPUs. These instances, built on the AWS Nitro System for enhanced security and performance, target production workloads like data analytics, conversational AI, and content generation. The partnership also introduces interconnect acceleration for disaggregated LLM inference by integrating NVIDIA's Inference Xfer Library (NIXL) with AWS Elastic Fabric Adapter (EFA). This integration aims to reduce communication bottlenecks between GPUs and AWS Trainium chips, enabling high-throughput, low-latency KV-cache data movement critical for scaling modern AI inference clusters.
- AWS will deploy over 1 million NVIDIA GPUs (Blackwell & Rubin) across its regions starting in 2026.
- AWS is first to announce EC2 instances with NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs, built on the Nitro System.
- New NVIDIA NIXL integration with AWS EFA accelerates disaggregated LLM inference, reducing bottlenecks for large models.
Why It Matters
This provides enterprises with the massive, production-ready infrastructure needed to scale complex agentic AI systems from pilot to business impact.