Microsoft Azure Launches First NVIDIA GB300 NVL72 Supercomputing Cluster for OpenAI

Microsoft Azure has introduced the NDv6 GB300 VM series, the world’s first production-scale cluster built on NVIDIA GB300 NVL72 systems, according to a Microsoft announcement. The new cluster, designed for NVIDIA’s Blackwell Ultra GPUs, is optimized for OpenAI’s most demanding inference workloads.

The Azure system integrates more than 4,600 NVIDIA Blackwell Ultra GPUs connected through the NVIDIA Quantum-X800 InfiniBand platform, providing 800 Gb/s bandwidth per GPU. Each rack-scale unit combines 72 Blackwell Ultra GPUs with 36 Grace CPUs, delivering up to 1.44 exaflops of FP4 Tensor Core performance and 37 terabytes of fast memory per VM. The NVLink Switch fabric enables 130 TB/s of bandwidth within each rack for high-speed intra-rack communication.

Microsoft Azure engineered the cluster with custom cooling, power distribution, and reworked software for orchestration and storage to support large-scale AI workloads. NVIDIA noted in its blog that the system is designed for reasoning models, agentic AI, and multimodal generative AI, providing unified memory and compute capacity for frontier model development.

The GB300 NVL72 deployment marks the next phase of Microsoft’s and NVIDIA’s collaboration to scale AI infrastructure globally. Microsoft plans to expand the rollout to hundreds of thousands of Blackwell Ultra GPUs across its data centers to support next-generation AI training and inference at supercomputing scale.