Microsoft links 700-mile-apart data centers for first AI superfactory

Microsoft has launched its first AI "superfactory" by linking two large data centers located 700 miles apart—in Wisconsin and Atlanta—into a unified system called Fairwater. This network of interconnected data centers is designed to accelerate the training of massive AI models from months down to weeks by operating as a distributed supercomputer. Unlike traditional data centers that run many separate applications, Fairwater sites collaborate on a single, complex AI training task, enabling the handling of models with hundreds of trillions of parameters, a scale beyond the capability of any single facility. The Fairwater data centers utilize advanced NVIDIA GB200 NVL72 rack-scale systems, supporting hundreds of thousands of NVIDIA Blackwell GPUs, combined with innovative chip and rack architectures to maximize throughput per rack. The facilities feature a two-story design for increased GPU density and employ advanced liquid cooling with minimal water usage. Microsoft emphasizes the importance of building robust infrastructure that integrates GPUs into a cohesive system rather than simply increasing GPU count. Future expansions will connect more Fairwater sites