Microsoft Fairwater AI Superfactory | Hyperscale AI & Cloud Infrastructure Hub in Wisconsin

Status: Under Construction / Initial phases operational (announced September 2025 with $4 billion expansion)   |   Location: Mount Pleasant, Wisconsin, USA

Project Overview

The Microsoft AI Superfactory in Mount Pleasant, Wisconsin, exemplifies the radical evolution of data center design driven by the extreme demands of Generative AI and large language models (LLMs) like those developed by OpenAI. This facility is not a traditional cloud data center; it is a purpose-built, single-cluster supercomputer designed to operate as one massive machine, integrating seamlessly with Microsoft’s global Azure infrastructure. The initial commitment to the Wisconsin site alone is $4 billion, signaling its critical role in the new AI economy.

The central challenge addressed by this design is power density. Traditional data center racks consume 5-15 kW, but AI training racks—packed with hundreds of NVIDIA GPUs (such as the Blackwell GB200)—can draw 60 kW to over 150 kW. To manage the resulting thermal load, the facility relies on advanced closed-loop liquid cooling systems. This direct-to-chip cooling, which pipes chilled fluid directly to the processors, ensures optimal performance while being engineered for near-zero water consumption during operation, a crucial sustainability measure.

Connectivity is arguably the most innovative feature. The facility is interconnected with other dedicated Microsoft AI sites (e.g., in Atlanta, Georgia) via the AI Wide Area Network (AI WAN). This dedicated, high-capacity, low-latency fiber network allows the geographically dispersed data centers to function as a single, cohesive AI Superfactory. This allows for massive, distributed model training, enabling breakthroughs to be achieved in weeks instead of months. The internal network uses high-performance interconnects like NVIDIA’s Spectrum-X Ethernet AI fabric to ensure that all chips—millions of them across the campus—can communicate at high speed with minimal latency.

The project leverages innovative construction techniques, including partnerships with engineering giants like Bechtel to use modular construction. This "data center Lego blocks" approach involves manufacturing electrical, mechanical, and plumbing systems off-site for rapid assembly on-site, drastically accelerating the time-to-market. The goal is to deliver what amounts to Gigawatt-scale compute capacity with the speed and efficiency required to keep pace with the accelerating global AI race.

Quick Facts Selected Details

City Profile