The infrastructure arms race for AI continues, and AWS just made a major move. The cloud giant has announced the general availability of its P6e-GB200 UltraServers, powered by Nvidia Grace Blackwell Superchips. These new systems are engineered to handle the most demanding AI workloads on the planet—from trillion-parameter model training to real-time reasoning systems and next-gen agentic AI.
“The scale of the AI systems that our customers are building today—across drug discovery, enterprise search, software development, and more—is truly remarkable,” said David Brown, VP of compute and machine learning at AWS. “With the new Blackwell systems, we’re providing secure, reliable GPU infrastructure at a massive scale, so that customers can confidently push the boundaries of AI.”
Extreme performance meets efficient cooling
Each UltraServer is a powerhouse of up to 72 Blackwell GPUs linked by Nvidia’s latest NVLink interconnect, acting as a single compute unit. The result is 360 petaflops of FP8 compute, 13.4 TB of high-bandwidth memory, and 28.8 Tbps of Elastic Fabric Adapter (EFAv4) networking. This is enough to support training and inference at the very edge of what’s possible.
These UltraServers aren’t just a leap in processing power; they’re also AWS’s first large-scale deployment of liquid cooling, featuring a custom-built In-Row Heat Exchanger (IRHX). According to Brown, existing commercial cooling solutions “either consumed too much floor space or couldn’t scale to our needs.” So AWS built its own.
Two Blackwell options, built for different scales
The P6e-GB200 joins the previously launched P6-B200 instance family, which features 8 Blackwell GPUs and is designed for more flexible, medium-to-large scale AI and high-performance computing workloads. Together, they form the core of AWS’s updated AI infrastructure strategy.
“P6-B200 instances are a great fit for customers looking to migrate existing workloads with minimal code changes,” added Brown. “For those training frontier models at the trillion-parameter scale, the P6e-GB200 unlocks performance and efficiency that was simply out of reach before.”
Tightly integrated with AWS services and security
AWS has also embedded these new systems into its broader managed ecosystem, including Amazon SageMaker HyperPod, which simplifies provisioning and management of large GPU clusters, and Amazon EKS, now optimized for GB200 NVL72 topology and GPU-aware scheduling. P6e-GB200 UltraServers will also be available through NVIDIA DGX Cloud, offering access to Nvidia’s full AI software stack and expert support.
Beyond raw performance, AWS emphasized security, reliability, and scale as core differentiators. Both P6e and P6-B200 systems run on the Nitro System, AWS’s secure compute foundation that enables live patching, fine-grained isolation, and continuous optimization without downtime.
In a market where AI workloads grow more complex by the day, AWS is clearly betting that infrastructure matters, and it’s building for the long game. As AI workloads become more complex, AWS is betting that robust, efficient, and secure infrastructure will help partners stay ahead.
With innovations spanning compute, networking, sustainability, and managed services, the Blackwell-powered UltraServers give MSPs a robust foundation to expand their AI services portfolios and support clients at every stage of AI adoption.
“This launch announcement is an important milestone, and it’s just the beginning,” said Brown. “As AI capabilities evolve rapidly, you need infrastructure built not just for today’s demands but for all the possibilities that lie ahead. With innovations across compute, networking, operations, and managed services, P6e-GB200 UltraServers and P6-B200 instances are ready to enable these possibilities. We can’t wait to see what you will build with them.”
Recently, AWS announced new initiatives to change how partners engage with AWS and provide enhancements that focus on helping partners grow their relevance, reach, and relationships with their customers.