AWS plans 2 million more Nvidia GPUs as AI demand explodes
AWS is adding 2 million Nvidia GPUs in 2027 and 2028 as the AI boom pushes demand for computing power to new levels.
The AI infrastructure race is getting seriously crowded. Amazon Web Services and Nvidia are expanding their long-running partnership, with AWS planning to deploy 2 million additional Nvidia GPUs across its global infrastructure in 2027 and 2028.
AWS had already announced plans to add more than 1 million Nvidia GPUs from 2026, but the companies say demand has gone beyond those expectations.
Why AWS needs so many GPUs
AI needs a lot of computing power, especially when companies are training large models or running AI tools for millions of users. The new capacity will include Nvidia’s Blackwell Ultra, Rubin and Rubin Ultra GPUs.
These chips will support workloads ranging from AI assistants and model training to scientific research and automation. The partnership is also targeting agentic AI, where systems can plan and carry out multi-step tasks with less human input.
Physical AI is another focus, covering systems such as robots and autonomous machines that interact with the real world.
It is about more than just GPUs
AWS and Nvidia are expanding their work across the wider AI stack. The companies plan to bring Nvidia Vera CPU-based infrastructure to AWS, integrate Nvidia technology more deeply with AWS Nitro System and Elastic Fabric Adapter, and expand access to Nvidia Nemotron open models through Amazon Bedrock and Amazon SageMaker.
They also plan to build AI factories for the US government, including 100,000 GPUs on secure AWS infrastructure for federal and national security workloads. For data-heavy applications, AWS and Nvidia are working on GPU-accelerated processing for Amazon EMR using Nvidia cuDF, along with faster vector indexing for Amazon OpenSearch using Nvidia cuVS.
In short, the goal is not simply to throw more chips at AI. It is to make the entire infrastructure around those chips faster and more useful.
Google brings Intrinsic under its fold to scale robotics and physical AI
Nvidia is also getting deeper into robotics
Amazon Robotics will use Nvidia’s physical AI platform, including Jetson, Omniverse and Isaac technologies.
These tools can support robot simulation, training, route optimisation and safety testing, giving Nvidia another route into the growing robotics market.
The bigger picture is simple. AWS gets more capacity to meet its customers’ AI appetite, while Nvidia gets another huge deployment of its chips and software. With 2 million more GPUs planned, the cloud AI race clearly isn't slowing down. It is just getting a lot more expensive.