Microsoft has unveiled a new generation of data centers designed specifically for artificial intelligence. At the heart of this initiative stands Fairwater, a vast complex in Wisconsin described by the company as an “AI factory”—the largest and most technologically advanced site in Microsoft’s portfolio. It is the first in a series of next-generation facilities the corporation is constructing both in the United States and abroad.
Fairwater spans 315 acres in the Mount Pleasant area and comprises three buildings with a combined floor space of roughly 1.2 million square feet. Its construction required 46.6 miles of foundation piles, 26.5 million pounds of steel, and 120 miles of underground cabling. Unlike traditional data centers that handle isolated workloads such as email or web hosting, Fairwater functions as a single unified supercomputer powered by hundreds of thousands of NVIDIA GPUs. According to Microsoft, this cluster delivers computational performance ten times greater than the world’s fastest existing supercomputer.
At the core of its architecture lie NVIDIA GB200 servers, densely packed into clusters. Each rack houses 72 GPUs interconnected via NVLink, providing massive bandwidth and shared memory. This design achieves processing speeds of up to 865,000 tokens per second. Future facilities in Norway and the United Kingdom are slated to deploy the next-generation GB300 processors.
The building itself employs a two-level configuration, with racks positioned in close proximity to minimize network latency. Engineers envision scaling this model into not just a collection of machines, but a globally distributed supercomputer spanning multiple regions.
Given the extreme density of compute, cooling is handled by a closed-loop liquid system that recirculates fluid without ongoing water loss. Fairwater is equipped with one of the largest water-cooled refrigeration plants in the world.
Data storage has also undergone a fundamental overhaul. Azure Blob Storage has been reengineered to support over two million transactions per second per account, eliminating the need for manual sharding and enabling operations at exabyte scale.
In a post on X, Microsoft CEO Satya Nadella stressed that building AI infrastructure requires resources on a scale comparable to energy production: in the past year alone, Microsoft deployed over 2 gigawatts of new capacity—the equivalent of two nuclear power plants. He noted that Fairwater interlinks hundreds of thousands of GB200 GPUs with fiber optic cabling long enough to wrap around the Earth 4.5 times.
Sustainability is also central to the project. Fairwater runs on renewable energy and is integrated into local communities. Microsoft claims to be developing closed energy systems capable of meeting real computational demands without depleting natural resources.
Parallel facilities are now under construction in Norway and the United Kingdom. The Norwegian project is being developed with nScale and Aker JV, while the UK facility—also in partnership with nScale—is intended to become the nation’s largest supercomputer. In total, Microsoft is investing tens of billions of dollars and deploying hundreds of thousands of specialized chips across a global network of more than 400 data centers in 70 regions. The Wisconsin Fairwater complex is set to serve as the blueprint for this expansive AI infrastructure.
