AWS and NVIDIA commit to 2 million more GPUs for 2027-2028, on top of last year's 1 million-GPU deal

AWS and NVIDIA announced on August 26, 2026 an expansion of their GPU supply agreement, committing to deploy 2 million additional NVIDIA GPUs across AWS infrastructure in 2027 and 2028 — on top of the more than 1 million GPUs already committed for deployment starting this year.

What's new

The new GPUs span three NVIDIA architectures: Blackwell Ultra, Rubin, and the still-unreleased Rubin Ultra. NVIDIA CEO Jensen Huang said the scale reflects demand outpacing projections: "NVIDIA and AWS have built one of the great growth engines of the AI era, and demand is running ahead of every forecast." AWS CEO Matt Garman framed the expansion around customer flexibility: "Customers want the freedom to choose the best tools for their AI workloads, and they want confidence that everything works seamlessly together."

Beyond raw GPU count, the two companies detailed a broader technical push:

• CPUs and networking — deeper integration of NVIDIA Vera CPUs for agentic AI workloads and NVIDIA Spectrum networking tuned for AWS's infrastructure.
• Open models — NVIDIA's Nemotron open models made available through Amazon Bedrock and SageMaker.
• Data processing — GPU acceleration for Amazon EMR via the cuDF library, and for Amazon OpenSearch vector indexing via cuVS.
• Robotics and physical AI — Amazon Robotics building on NVIDIA Jetson, Omniverse, and Isaac platforms.
• Federal workloads — a dedicated commitment of 100,000 GPUs for U.S. government infrastructure meeting IL6+ security requirements.

Context

This builds directly on the GPU commitment the two companies announced at NVIDIA's GTC conference earlier in 2026, when AWS agreed to deploy more than 1 million NVIDIA GPUs starting this year. That original deal already ranked among the largest GPU supply agreements disclosed by any cloud provider. The new 2-million-GPU tranche, targeting next-generation Rubin-family silicon not yet shipping, signals both companies are locking in supply years ahead of deployment rather than negotiating architecture-by-architecture.

The announcement came during NVIDIA's quarterly earnings call, in which the company reported revenue nearly doubling year-over-year — context that underscores just how much of NVIDIA's growth is now tied to multi-year hyperscaler commitments like this one rather than one-off purchases.

Why it matters

Committing to 2 million GPUs across two future NVIDIA architectures, years before they ship, is a bet on where AI compute demand is headed rather than a response to current load — and it locks AWS into NVIDIA's roadmap through 2028 at a moment when AMD, Google's TPUs, and custom silicon from AWS itself (Trainium) are all vying for a share of that same buildout. The scope of the deal — spanning GPUs, CPUs, networking, open models, and even robotics and federal workloads — shows the AWS-NVIDIA relationship has moved well past chip sales into a full-stack infrastructure partnership, raising the bar for what "cloud AI partnership" means for competitors trying to match it.