Microsoft builds world's most powerful AI datacenter in Wisconsin

Microsoft announced on September 18, 2025, the nearing completion of Fairwater, described as the world's most powerful artificial intelligence datacenter, located in Mount Pleasant, Wisconsin. According to Microsoft CEO Satya Nadella's announcement on social media, the facility houses hundreds of thousands of NVIDIA GB200 graphics processing units connected by enough fiber optic cable to circle Earth 4.5 times.

The Wisconsin facility represents Microsoft's single largest AI infrastructure investment, featuring a seamless cluster designed to deliver 10 times the performance of today's fastest supercomputers. According to the company, Fairwater enables AI training and inference workloads "at a level never before seen."

The announcement came as part of Microsoft's broader global AI infrastructure expansion, which includes multiple similar datacenters under construction across the United States and international partnerships. According to Brad Smith, Microsoft's Vice Chair and President, the company completed construction and will bring the AI datacenter online in early 2026, fulfilling its initial $3.3 billion investment pledge.

Microsoft's global infrastructure spending has reached unprecedented levels. According to reports from The Motley Fool dated January 11, 2025, Microsoft announced plans to spend $80 billion during 2025 on AI datacenter construction worldwide, with more than half allocated to United States facilities. The investment exceeds the gross domestic product of numerous countries, including Croatia and Lithuania.

Technical specifications reveal advanced architecture

Fairwater spans 315 acres and houses three massive buildings with a combined 1.2 million square feet under roof. According to Scott Guthrie, Microsoft's Executive Vice President for Cloud + AI, construction required 46.6 miles of deep foundation piles, 26.5 million pounds of structural steel, 120 miles of medium-voltage underground cable, and 72.6 miles of mechanical piping.

The datacenter operates as a single massive AI supercomputer using flat networking to interconnect hundreds of thousands of NVIDIA GPUs. According to Guthrie's technical documentation, each rack contains 72 NVIDIA Blackwell GPUs tied together in a single NVLink domain that delivers 1.8 terabytes of GPU-to-GPU bandwidth. Individual racks provide access to 14 terabytes of pooled memory and can process 865,000 tokens per second.

Microsoft Azure was the first cloud provider to bring online NVIDIA GB200 server, rack, and full datacenter clusters. According to the technical specifications, the facility uses both InfiniBand and Ethernet fabrics delivering 800 Gbps in a full fat tree non-blocking architecture. This design ensures every GPU can communicate with every other GPU at full line rate without congestion.

The datacenter layout employs a two-story configuration to minimize latency. According to Microsoft's engineering documentation, racks are networked not only to adjacent units but also to additional racks positioned above or below them, reducing physical distance and associated communication delays.

Cooling system minimizes water consumption

Microsoft implemented advanced liquid cooling technology to address the thermal demands of high-density AI hardware. According to company documentation, more than 90 percent of Fairwater relies on a closed-loop liquid cooling system that requires zero operational water after initial construction filling.

The facility features the second largest water-cooled chiller plant on the planet. According to the technical specifications, the system continuously circulates water through the closed loop while 172 twenty-foot fans chill and recirculate the water back to the datacenter. This design maintains efficiency even at peak computational loads.

The remaining 10 percent of traditional servers use outdoor air for cooling, switching to water only during the hottest days. According to Microsoft's sustainability documentation, this approach dramatically reduces water usage compared to traditional datacenters. Annual water consumption equals roughly what a typical restaurant uses annually or an 18-hole golf course consumes weekly during peak summer.

Microsoft pre-paid for energy and electrical infrastructure to ensure the datacenter does not increase costs for local residents. According to the company, it will match every kilowatt hour consumed from fossil fuel sources one-for-one with carbon-free energy returned to the grid. This includes a new 250-megawatt solar project under construction in Portage County.

Global expansion includes international partnerships

Microsoft simultaneously announced international AI datacenter projects. According to a September 17, 2025 announcement, the company partnered with Nscale and Aker for a $6.2 billion facility in Narvik, Norway, located more than 200 kilometers north of the Arctic Circle.

The Norway location was chosen for abundant hydropower, low local electricity demand, affordable energy prices, cool climate, and established industrial infrastructure. According to Jon Tinter, president of business development and ventures for Microsoft, the five-year agreement beginning service delivery in 2026 will provide advanced AI infrastructure powered entirely by renewable energy.

In the United Kingdom, Microsoft announced a $30 billion investment over four years from 2025 through 2028. According to Brad Smith's September 16, 2025 announcement, this includes $15 billion in capital expenditures to build the country's largest supercomputer featuring more than 23,000 NVIDIA GPUs in partnership with Nscale.

The UK investment will expand Microsoft's datacenter footprint to meet growing AI demand from customers including Barclays, the NHS, London Stock Exchange Group, Premier League, Vodafone, UK Met Office, Unilever, and Wayve. According to the company documentation, Vodafone expanded Microsoft Copilot to 68,000 employees worldwide after seeing productivity gains of four hours per week during pilot testing.

Industry implications and market position

Meta's $64-72 billion 2025 AI investment positions it competitively against Microsoft's $80 billion infrastructure spending. According to analysis, this represents a 46% increase from 2024's $223 billion combined spending across major tech companies. Meta's Prometheus cluster targets over 1 gigawatt capacity from New Albany, Ohio, while Hyperion in Louisiana will reach 5 gigawatts across several years, spanning 4 million square feet across 2,250 acres.

Microsoft maintains the largest customer relationship with NVIDIA. According to The Motley Fool analysis, Microsoft purchased 485,000 NVIDIA GPUs in 2024, more than twice the volume of the second-largest customer, Meta Platforms. NVIDIA holds approximately 90 percent market share in the GPU space.

The announcement has significant implications for the marketing technology industry. AI infrastructure investments drive increasing demand for compute-intensive advertising algorithms, creating both opportunities and challenges for digital marketing platforms. Microsoft's datacenter capacity expansion enables more sophisticated AI-powered advertising tools and customer targeting capabilities.

Microsoft's investment scale reflects broader industry trends toward AI infrastructure development. Other technology companies have emphasized the importance of not underinvesting in AI capabilities. Meta announced on July 14, 2025, plans to invest "hundreds of billions of dollars" in AI infrastructure, including Prometheus and Hyperion data center clusters consuming energy equivalent to 4-5 million American homes annually. Meta's massive AI infrastructure expansion creates a direct competitive dynamic with Microsoft's $80 billion 2025 spending commitment.

The competitive pressure has led to partnerships between tech companies and energy providers, including Microsoft's agreement to restart the Three Mile Island nuclear plant. Meta similarly announced a 20-year nuclear energy agreement with Constellation for the Clinton Clean Energy Center in June 2025.

The facilities will support Microsoft's AI services including Azure OpenAI, Microsoft Copilot capabilities, and various customer workloads across multiple industries. According to Azure documentation, the company has deployed AI infrastructure across more than 100 datacenters worldwide, powering model training, test-time compute, reinforcement learning tuning, and real-time inference at global scale.

Microsoft's approach contrasts with traditional cloud datacenters optimized for smaller, independent workloads such as hosting websites, email, or business applications. The new AI datacenters function as integrated systems designed specifically for training frontier AI models and running large-scale artificial intelligence applications.

Environmental considerations mirror industry challenges

Microsoft's sustainability approach for Fairwater addresses similar challenges facing other tech giants pursuing AI infrastructure expansion. The Wisconsin facility matches every kilowatt hour consumed from fossil fuel sources one-for-one with carbon-free energy returned to the grid, including a new 250-megawatt solar project under construction in Portage County.

The closed-loop cooling system requiring zero operational water after construction contrasts with industry trends toward liquid cooling. Meta's sustainability tensions amid AI expansion illustrate broader conflicts between environmental commitments and AI infrastructure demands. Meta's Air-Assisted Liquid Cooling architecture requires hundreds of thousands of gallons daily per facility, despite achieving industry-leading 0.20 L/kWh water usage effectiveness.

Both companies face similar environmental pressures. Meta committed to net zero emissions and water positive status by 2030 while announcing data centers consuming energy equivalent to 4-5 million American homes annually. Microsoft's approach of pre-paying for energy infrastructure and implementing closed-loop cooling represents one strategy for addressing these tensions.

Employment and economic impact

The Fairwater facility will employ approximately 500 full-time workers once the first datacenter becomes operational, expanding to around 800 employees after completion of the second datacenter. According to Microsoft documentation, construction employed more than 3,000 workers during peak activity, including electricians, plumbers, pipefitters, carpenters, structural iron and steel workers, concrete workers, and earth movers.

Microsoft partnered with Gateway Technical College to launch Wisconsin's first Datacenter Academy, training more than 1,000 students over five years for high-demand datacenter roles. According to the company, Microsoft and more than 40 partners including United Way, University of Wisconsin, and Wisconsin Technical College System worked with Gener8tor to train 114,000 people in AI skills, including 1,400 residents of Racine County.

The company sponsored the nation's first manufacturing-focused AI Co-Innovation Lab at University of Wisconsin-Milwaukee. The facility resulted from partnerships with UW-Milwaukee, Wisconsin Economic Development Corporation, Gateway Technical College, and TitletownTech. According to Microsoft documentation, the lab helped 23 Wisconsin companies including RegalRexnord, Renaissant, and BW Converting implement AI solutions.

Microsoft expanded broadband access to more than 9,300 rural residents and delivered next-generation internet service to 1,200 homes and businesses in Sturtevant. The company joined forces with Root-Pike Watershed Initiative Network to restore prairie and wetland habitats in Racine and Kenosha counties, funding 20 ecological restoration projects.

Storage infrastructure supports massive data requirements

The AI datacenter includes separate storage and compute infrastructure spanning five football fields in length. According to Azure engineering documentation, Microsoft reengineered storage systems for demanding AI workloads across massive datacenter deployments. Each Azure Blob Storage account sustains over 2 million read/write transactions per second.

The storage foundation aggregates capacity and bandwidth across thousands of storage nodes and hundreds of thousands of drives. This enables exabyte-scale storage while eliminating manual sharding requirements and simplifying operations for large AI and analytics workloads.

Key innovations include BlobFuse2 technology delivering high-throughput, low-latency access for GPU node-local training. According to Microsoft documentation, this ensures compute resources remain active and massive AI training datasets stay available when needed. Automatic scaling dynamically allocates resources as demand grows.

Future expansion and industry implications

Microsoft's Wisconsin commitment extends beyond the initial facility. According to Brad Smith's announcement, the company committed an additional $4 billion over the next three years to build a second datacenter of similar size and scale, bringing total Wisconsin investment to more than $7 billion.

The company's finance leases indicate additional expansion plans. According to The Motley Fool analysis, Microsoft maintained $108.7 billion in finance leases for datacenter construction scheduled to commence between fiscal 2025 and 2030, with lease terms ranging from one to 20 years.

Microsoft positioned the Wisconsin facility as part of a distributed AI supercomputer concept. According to company documentation, these datacenters connect via Wide Area Network infrastructure, creating a resilient and scalable system operating as a single powerful AI machine. This distributed approach provides greater resiliency, scalability, and flexibility compared to single-facility limitations.

The announcement reflects industry-wide infrastructure requirements for frontier AI development. According to CEO Satya Nadella's statement, "For AI training workloads, you need compute at exponential scale." Microsoft designed the datacenter, GPU fleet, and network together as one integrated system enabling single jobs to run from day one at exponential scale across thousands of GPUs.

Timeline

• September 2024: Microsoft announces partnership with Constellation to restart Three Mile Island nuclear plant for AI power
• January 11, 2025: Microsoft announces $80 billion global AI infrastructure spending plan for 2025
• September 16, 2025: Microsoft announces $30 billion UK AI investment over four years
• September 17, 2025: Microsoft partners with Nscale and Aker for $6.2 billion Norway AI datacenter
• September 18, 2025: Microsoft announces completion of Fairwater AI datacenter in Wisconsin
• Early 2026: Target completion date for Fairwater datacenter operations
• 2025-2028: Timeline for additional $4 billion Wisconsin investment in second datacenter

Summary

Who: Microsoft Corporation, led by CEO Satya Nadella and Vice Chair Brad Smith, in partnership with NVIDIA for GPU hardware and various local organizations for workforce development.

What: Construction and announcement of Fairwater, described as the world's most powerful AI datacenter, featuring hundreds of thousands of NVIDIA GB200 GPUs with 10x the performance of today's fastest supercomputers.

When: Announced September 18, 2025, with operational completion targeted for early 2026 as part of Microsoft's broader $80 billion global AI infrastructure investment planned for 2025.

Where: Mount Pleasant, Wisconsin, spanning 315 acres with three buildings totaling 1.2 million square feet, as part of Microsoft's global network of AI datacenters including planned facilities in Norway and the UK.

Why: To meet exponential demand for AI training and inference capabilities, enabling frontier AI model development while supporting Microsoft's Azure cloud services, Copilot products, and customer AI workloads across multiple industries.