Microsoft is developing a global 'supercluster' network that connects multiple data centers across vast distances to power next-generation artificial intelligence models requiring hundreds of trillions of parameters. The initiative, first launched in October, links facilities such as those in Mount Pleasant, Wisconsin, and Atlanta, Georgia, forming a unified high-performance computing system. Though Microsoft has not disclosed the exact technology behind the interconnection, it aligns with recent advances like Cisco’s 51.2 Tbps routers and Broadcom’s new high-bandwidth hardware. The company’s Fairwater data centers employ direct-to-chip liquid cooling to minimize water use. According to Azure CTO Mark Russinovich, the goal is to distribute AI workloads across multiple centers, enabling efficient training of massive models. Analysts say this approach will give Microsoft flexibility in choosing new data center locations with optimal climate, land cost, and energy availability.