AWS will receive two million additional Nvidia GPUs in 2027 and 2028, a move that materially expands the cloud provider's capacity for training and running large AI models.
The expanded agreement, announced during Nvidia's quarterly earnings call, covers Nvidia's newest accelerators, including Blackwell Ultra, Rubin, and Rubin Ultra. It follows a separate commitment five months ago in which Amazon agreed to deploy more than 1 million Nvidia GPUs across AWS starting this year. Nvidia said that since then, "demand has exceeded those expectations."
Neither firm disclosed financial terms. Based on typical GPU unit pricing, the added orders are likely worth tens of billions of dollars, though the companies declined to spell out exact returns for Nvidia.
The deal stretches beyond chips. Nvidia said its networking hardware, open models, CPUs, data processing software, and robotics platforms will be integrated across AWS. That includes plans to make Nvidia's Nemotron family of open models available on Amazon Bedrock and SageMaker, and to supply technology to power Amazon's warehouse robots.
Nvidia CFO Colette Kress said the company will also ship Vera CPUs, "some integrated with Rubin, others standalone," and she expects Vera to be deployed by "every major hyperscaler, neocloud, AI lab, and system OEM, with shipments already underway to our lead partners," naming Oracle and SpaceXAI as examples.
The announcement arrives while Amazon continues to build its own processors. AWS has developed Trainium chips as an alternative for deep learning workloads and sells Arm-based Graviton CPUs as a server option. Amazon said its custom-chip business crossed a $25 billion annualized revenue run rate, supported by $225 billion in total commitments from AI labs such as Anthropic and OpenAI.
Nvidia reported $96.2 billion in sales for the quarter, with data center revenue at $89 billion, up 117% year over year. The company expects roughly $108 billion in revenue for the third quarter, with Rubin shipments already in production this quarter. To secure capacity, Nvidia has committed $279 billion to supply and manufacturing projects, up from $119 billion last quarter, including $92 billion planned for the remainder of the fiscal year and $87 billion for fiscal 2028.
Jensen Huang framed the moment as a turning point for the industry, saying, "The thing that matters for the industry is that AI is now doing productive and useful work," and, "AI is generating profitable tokens... If we had more compute, we could generate more profitable tokens, which results in more profit for all of the services. This is the exact phase where we're at, which is the reason why everybody's leaning in."
The near-term test will be commercial: whether extra compute capacity accelerates revenue for cloud customers and chip suppliers alike. Investors and customers will watch Rubin and Vera shipments, AWS deployments of Nvidia's software stack, and how Amazon balances those investments with its own Trainium and Graviton chips as the AI arms race continues.
