Microsoft Fairwater Wisconsin
AI supercomputer campus in Mount Pleasant, Wisconsin
This facility is a purpose-built AI data centre in Mount Pleasant, Wisconsin. It forms part of Microsoft's Fairwater series, designed to function as a single massive supercomputer for training large language models using NVIDIA GPUs and liquid cooling.
- Covers 315 acres with three main buildings
- Uses closed-loop liquid cooling to save water
- Functions as a single massive AI supercomputer
- Features a two-story rack layout to reduce latency
- Hosts hundreds of thousands of NVIDIA GPUs
What it is
The Fairwater site in Wisconsin is a specialised infrastructure hub built by Microsoft. It serves as an AI factory rather than a traditional cloud host. The campus covers 315 acres and contains three large buildings with over one million square feet of roofed space. Its primary function is to train frontier artificial intelligence models. It operates as a unified supercomputer, linking hundreds of thousands of graphics processing units into a single cluster. This design allows it to handle complex mathematical computations required for modern AI applications.
How it came about
Microsoft introduced this facility in 2025 as part of a broader investment in AI infrastructure. It represents the first of several identical Fairwater campuses planned across the United States. The company describes it as the largest and most sophisticated AI factory it has constructed to date. Construction involved significant material usage, including millions of pounds of steel and extensive underground cabling. The site was developed to support the growing demand for generative AI services and connects to a global network of similar facilities.
The building and the halls
The campus features a unique two-storey layout for its server halls. This vertical arrangement reduces the physical distance between racks, which helps lower latency in network communications. The structures house dedicated clusters of NVIDIA GB200 servers. Each rack contains up to 72 Blackwell GPUs, linked via high-bandwidth connections. The design prioritises density and speed, allowing the hardware to function as a cohesive unit. The buildings are engineered to support extreme compute loads that exceed those of conventional data centres.
Power and cooling
The facility relies on a closed-loop liquid cooling system for the majority of its capacity. This method circulates cold liquid directly into servers to extract heat efficiently, reusing the same water without evaporation losses. A massive chiller plant supports this process, with hot water piped to external fins where large fans cool it for recirculation. The remaining ten per cent of servers use outdoor air, switching to water only during peak heat. This approach minimises water waste compared to traditional evaporative cooling methods.
What runs inside
This centre powers Microsoft Azure AI services, including Copilot and workloads for OpenAI. Unlike standard centres that host independent applications, this site runs single, massive training jobs across its entire hardware fleet. It processes trillions of tokens to refine AI models. The interconnected GPUs allow for parallel processing at a scale that delivers significantly higher performance than previous supercomputers. It supports both the training of new models and the inference tasks required for live AI applications.
The place
Located in Mount Pleasant, Wisconsin, the site was chosen for its ability to support large-scale infrastructure projects. The region provides the necessary land and grid connectivity for such a substantial installation. As part of a distributed network, it links to other Fairwater sites via dedicated fibre optics. This geographic distribution allows Microsoft to create a virtual supercomputer that spans multiple locations, enhancing resilience and compute availability for global users.
Environment and impact
The project represents a major capital investment, part of a tens-of-billions-dollar global programme. By using closed-loop cooling, the facility aims to reduce its environmental footprint regarding water consumption. The design avoids the high water usage typical of evaporative systems. While specific local economic impacts are not detailed, the scale of construction suggests significant activity in materials and labour during the build phase. The focus on efficiency addresses concerns about the resource intensity of AI training.
Network and links
The internal network uses a fat-tree architecture with InfiniBand and Ethernet fabrics. This setup ensures every GPU can communicate with others at full speed without congestion. Racks are connected via NVLink for high-bandwidth data transfer. Externally, the site connects to other Fairwater campuses through a dedicated AI Wide Area Network. This fibre-linked system allows multiple sites to operate as a single distributed supercomputer, sharing workloads seamlessly across long distances.
Questions people ask about Microsoft Fairwater Wisconsin
What makes Fairwater different from a normal data centre?
Normal centres host many separate apps. Fairwater acts as one giant supercomputer. It links thousands of chips to train single AI models. The hardware works together as a unified cluster rather than independent servers.
How does the cooling system work?
It uses closed-loop liquid cooling. Cold water circulates directly into servers to remove heat. The water is reused continuously without evaporation. Only a small fraction of servers use air cooling during hot weather.
Where is this data centre located?
It is situated in Mount Pleasant, Wisconsin. The campus spans 315 acres. It is the first of several identical Fairwater sites being built across the United States by Microsoft.
What hardware is inside the facility?
The centre uses NVIDIA GB200 servers. Each rack holds up to 72 Blackwell GPUs. These chips are connected via NVLink to form high-performance computing clusters for AI training tasks.
Good reading and sources
There is no English Wikipedia article on Microsoft Fairwater Wisconsin yet, so the facts on this page come from the sources listed above: blogs.microsoft.com/blog/2025/09/18/inside-the-worlds-most-powerful-ai, datacenterdynamics.com/en/news/microsoft-launches-atlanta-fairwater-da. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryDatacenter wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.