AWS and NVIDIA: 2 Million Additional GPUs for Agentic and Physical AI in 2027-2028

AWS and NVIDIA expand a 16-year partnership to meet accelerating demand for agentic and physical AI infrastructure at global scale.

Enterprises, frontier labs and governments are moving AI workloads from pilot to production faster than infrastructure can keep pace. Teams need broader model choice, higher-throughput data pipelines and new capabilities for emerging use cases such as physical AI, without compromising security or reliability. The joint expansion between AWS and NVIDIA directly addresses this gap by combining accelerated compute, custom silicon, networking and open models into co-engineered stacks that customers can consume as managed cloud services.

What was announced

On August 26, 2026, Amazon Web Services and NVIDIA announced a major expansion of their strategic collaboration. AWS plans to deploy 2 million additional NVIDIA GPUs across its global infrastructure in 2027-2028, including NVIDIA Blackwell Ultra, Rubin and Rubin Ultra generations. The companies will also bring NVIDIA Vera CPU-based infrastructure to AWS, extend NVLink Fusion with custom NVIDIA high-bandwidth memory (NVHBM), and build AI factories for the U.S. Government with 100,000 GPUs on secure AWS infrastructure for federal and national-security workloads at Impact Level 6 (IL6) and above.

The scope covers the full stack: GPUs, CPUs, networking, open models and software. NVIDIA Nemotron open models remain available on Amazon Bedrock as serverless models and on Amazon SageMaker for custom deployment. NVIDIA cuDF and cuVS CUDA-X libraries accelerate data processing on Amazon EMR and vector indexing on Amazon OpenSearch. Amazon Robotics adopts NVIDIA's physical AI platform, including Jetson, Omniverse and Isaac, for warehouse automation and next-generation robots.

NVIDIA and AWS have built one of the great growth engines of the AI era, and demand is running ahead of every forecast. For 16 years, we have scaled NVIDIA computing in the cloud together. Now we are expanding our partnership across the full stack to make agentic and physical AI real at an unprecedented pace and scale that only AWS and NVIDIA can deliver.

Why this matters

For CIOs, CISOs and IT directors, this expansion translates into predictable capacity for training and inference workloads that would otherwise face allocation constraints. Procurement leaders gain a broader choice of compute options ranging from AWS custom silicon (Trainium) to NVIDIA GPUs and Vera CPUs on a single cloud, reducing vendor lock-in inside a single platform. The AWS Nitro System and Elastic Fabric Adapter (EFA) underpin all NVIDIA GPU-based and Trainium-based EC2 instances, delivering the security isolation and network performance required for production AI.

For government agencies, dedicated AI factories with 100,000 GPUs on IL6-certified infrastructure remove a critical blocker to deploying generative AI on classified workloads. For enterprise data teams, GPU-accelerated processing on Amazon EMR delivers up to 3.7x faster processing and 30% better price-performance versus CPU configurations, while GPU-accelerated vector indexing on Amazon OpenSearch reaches up to 9x faster indexing at a quarter of the cost.

Technical details

• GPU expansion: 2 million additional NVIDIA Blackwell Ultra, Rubin and Rubin Ultra GPUs in 2027-2028 across AWS Global Infrastructure and AI factories.
• Amazon EC2 G7 instances: Powered by NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs, delivering 4.6x AI inference and 2.1x graphics performance versus G6.
• NVIDIA Vera CPUs on AWS: High-performance CPU compute purpose-built for the next generation of AI workloads.
• NVLink Fusion with NVHBM: Custom high-bandwidth memory integrated with Trainium for higher performance and power efficiency in a common rack-scale architecture.
• Federal AI factories: 100,000 GPUs on secure AWS infrastructure for U.S. Government workloads at IL6 and above.
• Nitro System and EFA: Security isolation and low-latency scale-out networking for all GPU and Trainium-based EC2 instances.
• Nemotron open models: Available on Amazon Bedrock (serverless) and Amazon SageMaker (self-managed).
• Data and vector acceleration: NVIDIA cuDF and cuVS CUDA-X libraries on Amazon EMR and Amazon OpenSearch for faster analytics and RAG pipelines.
• Physical AI: Amazon Robotics integrates NVIDIA Jetson, Omniverse and Isaac for simulation, synthetic data generation and robot training on GPU-accelerated EC2.

Softprom and Amazon Web Services

Softprom is the official partner of Amazon Web Services. Our team helps enterprises, government organizations and startups architect, migrate and scale workloads on AWS, including AI/ML platforms based on NVIDIA-accelerated instances, Amazon Bedrock, Amazon SageMaker and Amazon EMR.

This content was prepared as part of the Softprom DistriFlow project — an automated system for monitoring and adapting vendor news. Original source: original article.