Nvidia to supply AWS chips
Nvidia struck a multi‑year deal to sell 1 million AI chips to Amazon’s AWS by end of 2027 — a major capacity push that will accelerate hyperscaler GPU availability for enterprise GenAI workloads. That scale shift tightens competition for high‑performance inference/training capacity and reinforces the need for hybrid GPU strategies. ( )
AWS said it will deploy more than 1 million NVIDIA GPUs across AWS Regions starting in 2026, explicitly naming the Blackwell and Rubin GPU architectures for that rollout. (aws.amazon.com) Amazon announced new EC2 offerings that will support NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs (AWS said it is the first major cloud provider to do so) and EC2 G7e instances powered by NVIDIA RTX PRO 6000 Blackwell Server Edition for analytics and generative workloads. (aws.amazon.com) The collaboration includes interconnect acceleration for disaggregated LLM inference via NVIDIA NIXL on the AWS Elastic Fabric Adapter and a claim of 3x faster Apache Spark performance on Amazon EMR running on EKS with EC2 G7e instances. (aws.amazon.com) Nvidia’s Ian Buck told Reuters the deliveries will begin in 2026 and continue through 2027, and he said the order covers a broad mix beyond GPUs including NVIDIA Spectrum networking silicon and new Groq-based chips. (investing.com) Nvidia’s December 2025 licensing/asset deal with Groq—reported at roughly $20 billion—gave Nvidia rights to Groq’s inference LPU technology and led to Groq executives joining Nvidia, a move companies cited while describing Groq LPUs being integrated into new rack platforms. (bloomberg.com) At GTC 2026, CEO Jensen Huang said the Blackwell and Rubin families represent at least a $1 trillion revenue opportunity through the end of 2027, a figure Nvidia used when outlining its expanded multi‑chip and platform strategy. (bloomberg.com) AWS also announced expanded NVIDIA Nemotron model support on Amazon Bedrock and tighter model-serving integrations on EC2, linking the hardware scale commitment to concrete platform-level model fine-tuning and inference capabilities. (aws.amazon.com)
