Summary
- • AWS will deploy over 1 million NVIDIA GPUs across global regions starting in 2026
- • AWS is the first major cloud provider to announce NVIDIA RTX PRO 4500 Blackwell support
- • NVIDIA NIXL on AWS EFA enables interconnect acceleration for disaggregated LLM inference
- • NVIDIA Nemotron model support expanded on Amazon Bedrock
Details
AWS to deploy 1M+ NVIDIA GPUs globally starting 2026
The deployment covers both Blackwell and next-generation Rubin GPU architectures across AWS global cloud regions, reinforcing AWS's claim to the broadest NVIDIA GPU-based instance portfolio among major cloud providers.
AWS first major cloud to support NVIDIA RTX PRO 4500 Blackwell Server Edition
Instances built on the AWS Nitro System, suited for data analytics, conversational AI, content generation, recommender systems, video streaming, and graphics workloads. Availability described as 'coming soon.'
NVIDIA NIXL on AWS EFA accelerates disaggregated LLM inference
NIXL provides interconnect acceleration specifically for disaggregated large language model inference workloads running on AWS Elastic Fabric Adapter, targeting latency and throughput bottlenecks in production AI serving.
3x faster Apache Spark performance with NVIDIA RTX PRO 6000 Blackwell GPUs
Achieved using Amazon EMR on Amazon EKS with EC2 G7e instances. The improvement targets data engineering and analytics pipelines, broadening GPU utility in enterprise workflows beyond pure model training or inference.
Expanded NVIDIA Nemotron model support added to Amazon Bedrock
Nemotron is NVIDIA's family of enterprise-focused LLMs. Expanding availability on Bedrock gives AWS customers access to these models through the managed model API service alongside existing foundation model offerings.
AWS and NVIDIA extend collaboration to Spectrum networking
Added to over 15 years of joint innovation, the Spectrum networking collaboration extends the partnership beyond compute into high-speed interconnect infrastructure, relevant for large-scale distributed AI training and inference.
Infrastructure = compute/networking buildout, Product Launch = new service or instance availability, New Tech = novel technical capability, Stat = quantified performance claim, Partnership = joint business/technical agreement
What This Means
This expanded AWS-NVIDIA partnership represents a significant escalation in cloud GPU commitments, with over one million GPUs — spanning current Blackwell and upcoming Rubin architectures — set to land across AWS regions within the year. For enterprises moving AI from pilot to production, this signals more availability and variety of high-performance GPU instances, including first-mover access to RTX PRO 4500 Blackwell hardware. The additions to interconnect technology and Spark performance also indicate that the collaboration is maturing beyond model training into the full AI infrastructure stack, including inference serving and data pipelines.