Satya Nadella, CEO of Microsoft, confirmed on X that the company has brought up an NVIDIA Vera Rubin NVL72 system for validation inside its cloud infrastructure. Nadella wrote that Azure is the first cloud to bring up the system for validation, describing the milestone as another step toward building next-generation AI infrastructure in collaboration with NVIDIA.
The Rubin platform represents the architectural successor to NVIDIA’s Blackwell generation. The NVL72 configuration integrates 72 Rubin GPUs into a rack-scale AI system connected through next-generation NVLink fabric and large pools of high-bandwidth memory. The system operates as a single logical accelerator designed to support the training and inference of frontier-scale AI models. Compared with the current GB200 NVL72 generation, Rubin is expected to significantly increase compute density, memory bandwidth, and interconnect performance.
Microsoft’s validation suggests hyperscalers are already preparing infrastructure for Rubin-class AI clusters. These platforms target multi-rack AI systems scaling to tens or hundreds of thousands of GPUs and require new approaches to optical interconnects, power delivery, and liquid cooling. Azure has historically served as an early deployment partner for NVIDIA GPU generations including H100 and Blackwell.
• Rubin follows NVIDIA’s Blackwell generation and targets large-scale AI training and inference
• NVL72 architecture integrates 72 GPUs into a rack-scale AI platform
• Systems operate as a single logical accelerator using NVLink fabric
• Designed for hyperscale AI clusters supporting frontier model training
• Azure validation signals early hyperscaler readiness for Rubin infrastructure
The confirmation arrives during a busy week for the AI infrastructure ecosystem. NVIDIA GTC is taking place in San Jose while the optical networking industry gathers in Los Angeles for the Optical Fiber Communication Conference (OFC). The parallel timing highlights the growing interdependence between AI compute platforms and the high-speed networking technologies required to scale them.
“We’re the first cloud to bring up an NVIDIA Vera Rubin NVL72 system for validation, another big step in building the next generation of AI infrastructure with NVIDIA,” Nadella wrote on X.
🌐 Analysis:
Hyperscalers increasingly validate next-generation GPU systems months ahead of commercial deployment in order to redesign data center fabrics, power systems, and optical interconnect architectures around them. Microsoft’s early Rubin validation suggests the industry transition beyond Blackwell is already underway. Developments being discussed this week across GTC and OFC—including scale-up optical interconnects, co-packaged optics, and 224G networking—directly support the infrastructure required to interconnect Rubin-class AI clusters.
