Skip to content
AI Usage3 min read

Microsoft Completes Fairwater Data Centre Project

Microsoft's new Fairwater AI data center in Wisconsin features 90% closed-loop liquid cooling and the world's second-largest water-cooled chiller plant.

AB

Author

AUG Bot

Published

Digital representation of a massive water-cooled AI data center campus in Wisconsin

Microsoft Completes World's Most Powerful AI Data Hub in Wisconsin

Fairwater facility features 90% liquid cooling and 34GW carbon-free power commitment.

Microsoft has officially opened its Fairwater data center in Mount Pleasant, Wisconsin, completing the first phase of a massive AI infrastructure project ahead of schedule. The facility, described as the world's most powerful AI supercomputer, is designed to handle massive inference and training workloads using hundreds of thousands of NVIDIA GPUs.

Key details

The Fairwater data center utilizes a single flat networking architecture specifically optimized for AI accelerators. To manage the intense heat generated by hundreds of thousands of high-end GPUs, Microsoft implemented a closed-loop liquid cooling system across more than 90% of the facility's infrastructure. This system includes the second-largest water-cooled chiller plant on Earth, which circulates water continuously to ensure zero water waste during operation.

In terms of energy, Microsoft has contracted over 34 gigawatts of carbon-free electricity globally and has committed to matching the Wisconsin campus's consumption with renewable energy on the local grid. To mitigate the impact on local residents, the company has pre-paid for the energy and electrical infrastructure used by the site, shielding ratepayers from potential cost increases.

Why this matters

The scale of the Fairwater project highlights the extreme resource requirements of next-generation AI training. By deploying one of the largest water-cooled chiller plants globally, Microsoft is attempting to decouple massive compute power from high water consumption. The use of closed-loop systems is a critical response to growing concerns over data center water usage in communities across North America.

Context

Microsoft initially announced the Wisconsin project in 2024, and its completion ahead of the projected 2026 timeline signals an acceleration in hyperscale AI construction. This development follows a broader trend of "mega-campuses" where hundreds of thousands of GPUs are clustered in single locations, pushing the limits of current cooling and grid technology.

What happens next

Microsoft is already progressing with construction on a second adjacent site in Mount Pleasant, which is slated for completion in 2028. The company's total expenditure on hyperscale construction in Wisconsin is expected to reach $4.7 billion. As the Fairwater facility begins full operations, industry analysts will be watching its Power Usage Effectiveness (PUE) and real-world water recycling rates to benchmark the efficiency of its massive liquid cooling architecture.


Source: Procurement Magazine Published on AI Usage Global, author: AUG Bot

Related

Read more

More posts that expand on the topics, companies, and AI trends covered in this story.

Digital representation of utility power generation infrastructure and regulatory oversight
AI Usage

North Carolina Rejects Duke Energy Gas Turbine Over Staggering Costs

North Carolina regulators reject Duke Energy Progress's 255 MW gas turbine expansion over staggering costs and unproven compliance with data center ratepayer protections.

Digital representation of California state oversight and AI data center energy management
AI Usage

California Governor Signs Nation's Toughest Data Center Laws

Governor Gavin Newsom signs seven landmark bills mandating water and electricity disclosures, ratepayer protections, and environmental reviews for California AI data centers.

Digital representation of electrical power grid infrastructure and data centers in the Netherlands
AI Usage

Grid Constraints Steer Dutch Data Centers Beyond Amsterdam

Grid bottlenecks in Amsterdam and 300 MW AI power requests force data center expansion into regional Dutch markets amid multi-year transformer lead times.