AWS Expands Nvidia GPU Fleet to Two Million Additional Units

Amazon Web Services (AWS) has signed a deal to deploy an additional two million Nvidia GPUs across its global infrastructure. The agreement also includes plans to deepen collaboration on various technologies, including AI factories, CPUs, networking, open models, data processing, and robotics.

The deployment of the GPUs will be completed by 2027 and 2028, with AWS bringing online more than one million Nvidia GPUs in 2026 as previously announced. The two million additional GPUs will include Blackwell Ultra, Rubin, and Rubin Ultra GPUs.

AWS will also expand its capacity for Blackwell servers using the Nvidia RTX Pro 4500 Blackwell Server Edition GPUs for its EC2 G7 instances. This upgrade offers improved AI inference performance by a factor of 4.6 and graphics performance by a factor of 2.1 compared to previous-generation G6 instances.

Nvidia CEO Jensen Huang said that the company has built 'one of the great growth engines of the AI era' with AWS, and that demand is running ahead of every forecast. He added that the expansion of their partnership will make agentic and physical AI real at an unprecedented pace and scale.