Amazon expands AWS commitment with 2 million additional Nvidia GPUs
Amazon and Nvidia have expanded their partnership with an agreement to add another 2 million Nvidia GPU chips to Amazon Web Services data centres in 2027 and 2028. The deployment will include Blackwell Ultra, Rubin and Rubin Ultra GPUs, designed for training and running AI models.
The announcement came during Nvidia’s quarterly earnings call, five months after Amazon agreed to deploy more than 1 million Nvidia GPUs across AWS infrastructure beginning this year. Nvidia said demand had exceeded the expectations behind that earlier agreement. Neither company disclosed financial terms, although the scale of the hardware order puts its value in the tens of billions of dollars.
A broader AWS and Nvidia integration
The arrangement extends beyond GPU supply. Nvidia said AWS will integrate its networking hardware for connecting thousands of GPUs into a system, along with open models, CPUs, data-processing software and its robotics platform. The companies cited surging demand from startups, enterprises, AI labs and governments as a factor in their closer collaboration.
Nvidia CFO Colette Kress said the company will also send an unspecified number of Vera CPUs, with some integrated with Rubin and others supplied as standalone processors. Nvidia has begun production shipments of Rubin this quarter and expects some third-quarter revenue from the next-generation GPUs.
Custom chips remain part of Amazon’s strategy
The larger Nvidia commitment does not displace Amazon’s in-house chip programme. AWS is developing Trainium chips for deep-learning workloads and has discussed selling them to other companies for use in data centres. That possible market expansion is central to AWS plans to sell Trainium chips as AWS weighs external demand for Trainium alongside its own cloud infrastructure needs.
Amazon also offers the Arm-built Graviton CPU, positioned against traditional server processors from Intel and AMD. Amazon said on its latest earnings call that its custom-chip business had passed a $25 billion annualised revenue run rate, driven by $225 billion in total commitments from AI labs including Anthropic and OpenAI.
Robotics and enterprise software join the deal
Amazon plans to adopt Nvidia’s full physical AI stack for its warehouse robot fleet. That stack includes Omniverse for simulation and digital twins, Cosmos for world models, Isaac for robotics development, and Jetson computing hardware for robots and edge AI.
On the enterprise side, AWS will serve Nvidia’s Nemotron family of open models through Amazon Bedrock and SageMaker. Nvidia reported $96.2 billion in second-quarter sales, with data-centre revenue of $89 billion, up 117% year on year, and forecast third-quarter revenue of $108 billion.
For businesses building AI services, the agreement reinforces the need to plan capacity, networking and model platforms together: proprietary chips may broaden procurement options, but large-scale Nvidia infrastructure remains a core part of AWS’s near-term expansion.