Microsoft Azure launches world’s first GB300 supercomputer for OpenAI

Microsoft Azure has unveiled the world’s first production-scale NVIDIA GB300 NVL72 supercomputing cluster, specifically designed for OpenAI’s most demanding AI workloads. This milestone represents a significant leap in AI infrastructure capabilities, featuring over 4,600 NVIDIA Blackwell Ultra GPUs that will power next-generation reasoning models and agentic AI systems while reinforcing American leadership in artificial intelligence.

What you should know: The new NDv6 GB300 VM series delivers unprecedented computational power through liquid-cooled, rack-scale systems that integrate cutting-edge hardware into a unified supercomputer.

• Each rack contains 72 NVIDIA Blackwell Ultra GPUs and 36 NVIDIA Grace CPUs, providing 37 terabytes of fast memory and 1.44 exaflops of FP4 Tensor Core performance per VM.
• The cluster connects over 4,600 GPUs using NVIDIA’s Quantum-X800 InfiniBand networking platform, creating a massive unified memory space essential for complex AI model training and inference.
• Microsoft applied “radical engineering to memory and networking” to achieve the massive scale of compute required for high-performance reasoning models.

Performance breakthrough: Recent benchmarks demonstrate the system’s exceptional capabilities, particularly for large language models and reasoning AI.

• In MLPerf Inference v5.1 benchmarks, the GB300 NVL72 systems delivered up to 5x higher throughput per GPU on the 671-billion-parameter DeepSeek-R1 reasoning model compared to NVIDIA’s previous Hopper architecture.
• The platform achieved leadership performance across all newly introduced benchmarks, including the Llama 3.1 405B model.
• NVIDIA Blackwell Ultra supports new formats like NVFP4 for breakthrough training performance and compiler technologies like NVIDIA Dynamo for optimal inference performance.

The networking architecture: A sophisticated two-tiered approach enables both scale-up performance within racks and scale-out performance across the entire cluster.

• Within each GB300 NVL72 rack, fifth-generation NVIDIA NVLink Switch fabric provides 130 TB/s of direct, all-to-all bandwidth between the 72 Blackwell Ultra GPUs.
• The NVIDIA Quantum-X800 InfiniBand platform delivers 800 Gb/s of bandwidth per GPU, featuring ConnectX-8 SuperNICs and advanced adaptive routing capabilities.
• NVIDIA Scalable Hierarchical Aggregation and Reduction Protocol (SHARP) v4 accelerates operations to significantly boost efficiency of large-scale training and inference.

Why this matters: This deployment marks a crucial step toward building the infrastructure needed for frontier AI development and positions the United States at the forefront of AI innovation.

• The achievement required reimagining every layer of Microsoft’s data center infrastructure, from custom liquid cooling and power distribution to reengineered software stacks.
• As Azure scales toward deploying hundreds of thousands of NVIDIA Blackwell Ultra GPUs, this milestone enables customers like OpenAI to access unprecedented computational resources for developing next-generation AI systems.

What they’re saying: Industry leaders emphasize the collaborative engineering effort required to achieve this scale of AI infrastructure.

• “Delivering the industry’s first at-scale NVIDIA GB300 NVL72 production cluster for frontier AI is an achievement that goes beyond powerful silicon — it reflects Microsoft Azure and NVIDIA’s shared commitment to optimize all parts of the modern AI data center,” said Nidhi Chappell, corporate vice president of Microsoft Azure AI Infrastructure.
• “Our collaboration helps ensure customers like OpenAI can deploy next-generation infrastructure at unprecedented scale and speed.”