News
Microsoft Signals Azure Datacenter Readiness for NVIDIA Rubin Platform
Microsoft said Monday that its Azure datacenters are engineered to support NVIDIA's newly launched Rubin platform, as cloud providers move to deploy the next generation of AI infrastructure.
The announcement followed NVIDIA's CES 2026 introduction of the Vera Rubin NVL72 rack-scale systems. Microsoft said its Fairwater AI superfactory sites in Wisconsin and Atlanta were designed with the power, cooling and networking capacity needed to accommodate Rubin without significant infrastructure changes.
"Azure's AI datacenters are engineered for the future of accelerated computing," Rani Borkar, president of Azure Hardware Systems and Infrastructure, wrote in a blog post. She said the design approach allows Rubin systems to be deployed across Azure's largest AI-focused facilities, including both current and planned sites.
The Rubin platform delivers a substantial increase in AI computing performance. According to NVIDIA, each Vera Rubin NVL72 rack provides up to 3.6 exaflops of performance, roughly five times that of the company's GB200-based systems.
Supporting that performance requires tighter thermal tolerances and higher rack densities than earlier generations. Microsoft said its datacenter designs incorporate liquid cooling, updated power distribution and rack geometries to support HBM4 memory and higher-speed networking.
Azure's infrastructure already supports the 1.6-terabit-per-second ConnectX-9 adapters used by Rubin for scale-out connectivity. Each rack combines 72 GPUs and 36 CPUs connected through sixth-generation NVLink, which NVIDIA said provides 260 terabytes per second of bandwidth.
Microsoft said its readiness reflects a multi-year co-design partnership with NVIDIA and experience deploying earlier architectures at scale. The company was among the first cloud providers to bring NVL72 platforms online for large AI training workloads.
Unlike some competitors that emphasize single, centralized AI campuses, Microsoft has focused on building regional AI superfactory sites, an approach the company said supports more predictable deployment of new hardware.
Microsoft is expected to offer Rubin-based instances in 2026, alongside AWS, Google Cloud and Oracle Cloud Infrastructure. NVIDIA said Rubin systems are in production, with broader deployments planned for the second half of the year.
The company said the platform is designed to reduce inference costs while improving performance for advanced reasoning and agent-based models, as competition among cloud providers continues to intensify.