日本語 
AWS and NVIDIA are expanding their partnership to meet growing demand for AI infrastructure, with plans to deploy an additional 2 million NVIDIA GPUs across AWS’s global infrastructure between 2027 and 2028.
The expansion builds on 16 years of collaboration and covers more than GPU computing. The companies are working across CPUs, networking, open models, data processing and robotics to support customers moving AI workloads into production.
The planned infrastructure will include NVIDIA Blackwell Ultra, Rubin and Rubin Ultra GPUs, along with NVIDIA Vera CPU-based systems. AWS and NVIDIA are also extending support for NVIDIA NVLink Fusion and NVHBM to improve memory performance and power efficiency across AI infrastructure.
Security is another focus. The companies plan to build a secure AI factory for the U.S. government with 100,000 NVIDIA GPUs to support federal and national security workloads, including highly sensitive workloads at Impact Level 6 or higher.
Also Read: VALANCE and Riken Sangyo Partner to Bring AI to Hiroshima
The partnership also brings NVIDIA Nemotron open models to Amazon Bedrock and Amazon SageMaker. AWS and NVIDIA are working to accelerate data processing and vector indexing through NVIDIA CUDA-X libraries, with GPU-based processing available through Amazon EMR and Amazon OpenSearch.
Robotics is also part of the expansion. Amazon Robotics will use NVIDIA’s physical AI technologies, including Jetson, Omniverse and Isaac, for simulation, robot training, route optimization and real-world validation.
The broader push reflects how AI demand is moving beyond model training into agent-based AI, enterprise automation, scientific research and physical AI. AWS is responding by expanding infrastructure and integrating more NVIDIA technologies across its cloud platform.
English