TMTPOST -- Microsoft Corporation is planning to duplicate   "the   most advanced"   artificial intelligence ( AI ) datacenter   in the world, doubling down   the software   titan ’ s infrastructure buildup.

Credit：Microsoft

Microsoft is in the final phases of building Fairwater, "the world ’ s most powerful AI datacenter"   in Mount Pleasant, Wisconsin, and is "committing an additional $4 billion to be spent in the next three years to build our second datacenter of similar size and scale — bringing our total investment in Wisconsin to more than $7 billion, announced Microsoft   Vice Chair   and President Brad   Smith   on Thursday.

Microsoft is   on track to complete construction and bring the   Fairwater   AI datacenter online in early 2026, fulfilling our initial $3.3 billion investment pledge. We ’ ve already begun hiring full-time employees to support its operation,   according   to   Smith. The datacenter,   designed to help build the world ’ s most advanced models   more efficiently, will house "hundreds of thousands of"   the world ’ s most powerful Nvidia   graphic processing units   ( GPUs ) , operating in seamless clusters connected by enough fiber to wrap the planet four times over,   said Smith.

The Fairwater AI datacenter   covers   315 acres and includes   three buildings with a combined 1.2 million square feet under roofs,   said Scott Guthrie, Microsoft Executive Vice President   of   Cloud AI.   The datacenter   is built to work as one massive AI supercomputer using a single flat networking interconnecting hundreds of thousands of the latest Nvidia   GPUs, and   thus   will enable it to   deliver 10   times   the performance of the world ’ s fastest supercomputer today, enabling AI training and inference workloads at a level never before seen, per Guthrie.

This datacenter runs a single, massive cluster of interconnected Nvidia   GB200 servers and millions of compute cores and exabytes of storage, all engineered for the most demanding AI workloads. Microsoft ’ s   Azure was the first cloud provider to bring online the Nvidia   GB200 server, rack and full datacenter clusters. Each rack packs 72 Nvidia   Blackwell GPUs, tied together in a single NVLink domain that delivers 1.8 terabytes of GPU-to-GPU bandwidth and gives every GPU access to 14 terabytes of pooled memory.

To address latency into the system   due to physical distance between racks   the racks in the Wisconsin AI datacenter are laid out in a two-story datacenter configuration, so in addition to racks networked to adjacent racks, they are networked to additional racks above or below them.   This layered approach sets Azure apart. Microsoft Azure was not just the first cloud to bring GB200 online at rack and datacenter scale, according to Guthrie.

What makes Fairwater   distinctive isn ’ t just its scale or speed, but   the sustainability   Microsoft   has   been   seeking   as a central   to its design, Smith noted. More than 90% of the facility will rely on a state-of-the-art closed-loop liquid cooling system, filled during construction and recirculated continuously, and   the remaining portion of the facility will use outside air for cooling, switching to water only on the hottest days, minimizing environmental impact and maximizing operational efficiency,   said Smith.

The Fairwater   and planned   new datacenter in Wisconsin are   parts of Microsoft ’ s   wave of purpose-built AI   infrastructure   around the world.

Microsoft last   Tuesday announced a   $30   billion   investment in   AI infrastructure and ongoing operations across Britain   from 2025 through 2028.   This marks the largest financial commitment Microsoft   has ever made in the kingdom.   With   this investment,   Microsoft   will partner with   Nscale, a British cloud computing firm,   to build Britain ’ s largest supercomputer, with   more than 23,000 graphics processing units   ( GPUs )   made by   Nvidia Corporation.  

Microsoft a day later   announced plans with nScale and Aker JV to develop a new hyperscale AI datacenter   in the town of Narvik, Norway,   worth an estimated $6.2 billion.   The three companies have signed a five-year agreement that will begin to deliver services in stages, commencing in 2026. The project will deliver advanced AI infrastructure powered by entirely renewable energy in secured grid capacity.