Microsoft Azure has officially launched the deployment of the world’s inaugural production-ready NVIDIA GB300 NVL72 supercomputing cluster. This cluster is meticulously crafted to handle OpenAI's demanding artificial intelligence workloads. It incorporates a staggering array of over 4,600 NVIDIA Blackwell Ultra GPUs, all interconnected through cutting-edge Quantum-X 800 InfiniBand networking technology. Each rack within the cluster is outfitted with 72 GPUs and 36 Grace CPUs, boasting an impressive intra-rack NVLink bandwidth of up to 130TB/s. Additionally, it delivers a maximum FP4 tensor core performance of 1,440 PFlops. The system employs a state-of-the-art liquid cooling solution and a full-stack optimized architecture, significantly slashing the training duration for trillion-parameter models from several months down to mere weeks. Co-developed by Microsoft and NVIDIA, this cluster is poised for future expansion, with plans to incorporate hundreds of thousands of Blackwell GPUs across global data centers. This expansion will provide pivotal computational support for OpenAI's research and development endeavors in multimodal models and intelligent agent systems.