Mt. Pleasant, Wisconsin, September 19, 2025- Microsoft announced the opening of Fairwater, a next-generation AI data center in Mt. Pleasant, Wisconsin, that the company describes as the world’s most powerful facility of its kind. Fairwater anchors Microsoft’s multi-billion-dollar global investment in purpose-built AI infrastructure and will serve as a cornerstone for training and deploying large-scale AI models across OpenAI, Microsoft Copilot, and enterprise workloads.
Spanning 315 acres with three buildings totaling 1.2 million square feet. At its core, Fairwater operates as a unified flat-networked cluster of NVIDIA GB200 servers, achieving performance levels more than 10 times faster than the world’s leading supercomputer today. Each rack processes up to 865,000 tokens per second.
The architecture integrates 72 Blackwell GPUs per rack within NVLink domains, providing 1.8 terabytes of GPU-to-GPU bandwidth and 14 terabytes of pooled memory. An 800-Gbps InfiniBand/Ethernet non-blocking fat-tree interconnect ensures ultra-low latency, reinforced by a two-story rack layout. Storage is handled separately in a five-football-field-long facility, powered by Azure Blob Storage with throughput exceeding two million transactions per second.
Fairwater also underscores Microsoft’s sustainability focus. The site employs closed-loop liquid cooling through what the company calls the world’s second-largest water-cooled chiller plant. With 172 massive fans circulating water, the system operates with zero operational water waste. Over 90% of the campus relies on liquid cooling, with the remainder using minimal-use air cooling systems.
The Wisconsin launch is part of a broader network strategy. Microsoft confirmed that identical campuses are under construction elsewhere in the U.S., alongside new hyperscale AI partnerships in Narvik, Norway, via nScale and Aker JV, and in Loughton, UK, where nScale is building what will be the country’s largest supercomputer. Future sites will incorporate NVIDIA’s GB300 chips for expanded memory and scaling.
“Fairwater represents the future of data centers, transforming them from general-purpose cloud infrastructure into AI factories designed for trillion-parameter models,” said Scott Guthrie, EVP of Cloud + AI at Microsoft. He added that the approach will “democratize access to AI services” while harmonizing hardware and software to maximize performance and efficiency.
By tightly integrating GPUs, interconnects, storage, and WAN networking, Microsoft aims to slash inefficiencies in AI training cycles and keep accelerators fully utilized, an optimization it compares to a football team’s drills. Beyond performance, the project is expected to generate local jobs and economic growth in Wisconsin while setting a benchmark for sustainable, AI-era infrastructure.
