MOUNT PLEASANT, WISCONSIN – Microsoft has announced the development of its newest U.S. AI datacenter, dubbed Fairwater, in Mount Pleasant, Wisconsin, which is engineered to function as a singular, massive AI supercomputer. This facility, part of a multi-billion dollar global infrastructure investment, is poised to deliver an unprecedented 10 times the performance of today’s fastest supercomputers, revolutionizing AI training and inference workloads.
The Fairwater datacenter, expected to be fully operational in early 2026, represents a significant leap in AI infrastructure, moving beyond traditional cloud datacenters optimized for numerous smaller tasks. Instead, it is purpose-built for large-scale artificial intelligence, operating as a cohesive, interconnected system.
The Architecture of an AI Supercomputer
Unlike conventional datacenters that manage independent workloads, Fairwater’s unique design integrates hundreds of thousands of the latest NVIDIA GPUs via a single flat networking interconnect. This architecture allows the entire facility to behave as one colossal AI supercomputer. Microsoft CEO Satya Nadella highlighted that these NVIDIA GB200 and GB300 GPUs are “connected by enough fiber to circle the Earth 4.5 times.”
Each rack within the datacenter is designed to function as a single, giant accelerator, packing 72 NVIDIA Blackwell GPUs. These are tied together in a single NVLink domain, providing 1.8 terabytes of GPU-to-GPU bandwidth and offering every GPU access to 14 terabytes of pooled memory. This setup enables each rack to process an astonishing 865,000 tokens per second, claiming the highest throughput of any cloud platform currently available.
Scale and Investment
The Fairwater complex spans 315 acres, housing three massive buildings with a combined 1.2 million square feet of space. Microsoft’s initial investment in Fairwater was $3.3 billion, with an additional $4 billion committed for a second, equivalent datacenter in the same state, bringing the total investment in Wisconsin to over $7.3 billion.
This Wisconsin project is part of a broader, estimated $80 billion global investment by Microsoft in AI infrastructure through 2028, aiming to significantly expand its AI capabilities worldwide.
Environmental Considerations and Global Expansion
Recognizing the immense energy demands of AI datacenters—which require 2-3 times more electricity than traditional ones—Microsoft has implemented advanced cooling and energy strategies. Fairwater utilizes a closed-loop liquid cooling system for over 90% of its compute capacity, which recirculates water continuously, thus minimizing water waste. The facility also aims to match all energy consumed with renewable sources, including a 250-megawatt solar project in Portage County. Microsoft is also actively exploring nuclear energy as a reliable, carbon-free power source for its expanding AI operations.
Beyond Wisconsin, Microsoft is constructing identical Fairwater-class datacenters in other U.S. locations and making substantial investments internationally. These include a $6.2 billion deal in Norway with Aker-Nscale to deploy up to 100,000 NVIDIA GPUs by late 2026, leveraging the region’s hydropower. Additionally, a $30 billion investment in the UK through 2028 will establish the UK’s largest supercomputer in partnership with Nscale and Nvidia.
Impact on AI Development
This new generation of datacenters is crucial for accelerating the development of frontier AI models, powering services like OpenAI, Microsoft AI, and Copilot capabilities. By creating a distributed network of these AI datacenters, interconnected via a robust Wide Area Network (WAN), Microsoft aims to democratize access to powerful AI services globally, offering greater resiliency, scalability, and flexibility for customers.
As Microsoft President Brad Smith stated, Fairwater is where “the most advanced AI models are going to be built for the world,” signifying its importance not just for the community and state, but for the global advancement of AI.
