Amazon and NVIDIA are deepening their AI partnership in a major expansion that will bring up to three million NVIDIA AI GPUs to Amazon Web Services infrastructure worldwide. The move is designed to support the growing demand for advanced AI workloads, including agentic AI, automation, physical AI, and large-scale enterprise AI applications.
The announcement follows NVIDIA’s latest fiscal second-quarter earnings and builds on a previous agreement revealed earlier this year, when the companies outlined plans for one million NVIDIA AI GPUs on AWS. With this expanded collaboration, Amazon is adding another two million GPUs, significantly increasing the scale of NVIDIA-powered AI computing available through its cloud platform.
Amazon says the goal is to give AWS customers one of the broadest possible selections of compute options for artificial intelligence. Alongside NVIDIA’s GPUs, AWS will continue offering its own custom AI silicon, including Trainium chips. This approach allows businesses, developers, and government customers to choose the hardware that best fits their performance, efficiency, and cost requirements.
A key part of the expanded partnership is Amazon’s plan to bring NVIDIA’s Vera CPUs to AWS. CPUs are becoming increasingly important in modern AI infrastructure, especially as agentic AI systems require more complex coordination, data handling, and real-time decision-making. By adding Vera CPUs to its cloud offerings, AWS aims to strengthen its full-stack AI computing capabilities.
Amazon is also working with NVIDIA to incorporate NVHBM technology into its AI infrastructure strategy. NVHBM is designed to deliver more efficient high-bandwidth memory for AI workloads. NVIDIA says the technology offers 30% more bandwidth and 15% better power efficiency compared with HBM4E memory. This could help improve the performance and energy efficiency of AI systems running on AWS, including those powered by Amazon’s Trainium chips.
The three-million-GPU deployment will also include a major government-focused component. Amazon plans to allocate 100,000 NVIDIA GPUs to dedicated AI factories built for the United States government. These facilities are expected to support Impact Level 6 security classifications, a standard used for highly sensitive government workloads.
NVIDIA CEO Jensen Huang described the long-running partnership with AWS as one of the major growth engines of the AI era. He said demand for NVIDIA-powered cloud computing continues to exceed expectations, and the expanded collaboration will now cover GPUs, CPUs, networking, open models, and software.
The deal highlights how quickly demand for AI infrastructure is rising across industries. As companies race to build smarter applications, automate workflows, and deploy advanced AI models, cloud providers are under pressure to deliver more powerful and flexible computing platforms. For AWS and NVIDIA, this expanded partnership positions both companies at the center of the next phase of AI adoption.
