On September 18, 2025, Microsoft confirmed it will pour an additional $4 billion into building a second AI datacenter in Mount Pleasant, Wisconsin—more than doubling its investment in the site to over $7 billion. The move transforms the project, called Fairwater, into what the company claims will be “the world’s most powerful AI datacenter,” capable of roughly ten times the performance of today’s fastest supercomputers. The news lands on land once earmarked for a Foxconn factory that famously failed to deliver on grand promises, igniting both excitement and hard-nosed skepticism among locals and industry watchers alike.

The Hardware Inside Fairwater

Microsoft isn’t shy about the scale. According to company executives, Fairwater will house “hundreds of thousands” of NVIDIA GB200 GPUs—the same Grace-Blackwell superchips that NVIDIA designed for trillion-parameter AI models. These aren’t just racks of standard servers; they are liquid-cooled, NVLink-connected systems built into NVIDIA’s NVL72 rack architecture, where each rack packs 36 Grace CPUs and 72 Blackwell GPUs into a shared memory domain.

The numbers Microsoft throws around are staggering. The internal fiber network, they say, could wrap the earth 4.5 times—roughly 180,000 kilometers of cabling. Cooling will rely on what Microsoft calls the second-largest water-cooled chiller plant on the planet, running a closed-loop system that recirculates water instead of evaporating it. The company claims annual water consumption for thermal management will be comparable to a single U.S. restaurant. To offset the enormous power draw, Microsoft has signed a 250-megawatt solar power purchase agreement with National Grid Renewables, and the company says it will match every fossil-fuel kilowatt-hour with renewable energy added to the grid.

While the hardware specs align with NVIDIA’s published documentation, the 10x performance figure lacks independent verification. Microsoft’s claim depends heavily on which benchmark you choose—traditional LINPACK floating-point operations won’t capture AI workload speeds, and no third-party results are yet public. For now, the 10x figure is best read as a corporate target rather than a proven fact.

Who Wins, Who Worries

For developers and enterprises, Fairwater signals a coming surge in Azure AI capacity. If the hardware is integrated into Azure’s public cloud, expect new virtual machine series that can train massive models in days instead of weeks. IT managers planning large-scale AI projects should keep an eye on Azure’s roadmap—these GPUs will likely appear in high-end, reserved-capacity tiers first. But until Microsoft publishes real-world throughput metrics, the performance leap is theoretical.

For Wisconsin residents, the picture is mixed. Microsoft promises 3,000 construction jobs at peak and around 800 permanent operations roles once both datacenters come online. The company also pledges local broadband upgrades and a Datacenter Academy for workforce training. These are tangible near-term benefits. However, the electric load is a genuine risk. Citing a Clean Wisconsin analysis, environmental groups warn that the combined demand from Fairwater and another planned Vantage datacenter in Port Washington could exceed the power use of all Wisconsin households. Without careful grid planning, ratepayers could see higher electricity bills or reliability headaches. Microsoft says it will pre-pay for grid upgrades, but the details aren’t public yet.

For environmental advocates, the closed-loop cooling is a genuine improvement over evaporative towers that suck up millions of gallons of freshwater. Yet the overall electricity consumption—and the carbon implications of scaling up renewable matching—deserves scrutiny. The 250 MW solar farm will produce clean energy, but it won’t cover the entire load at all times; the project will still draw from a grid that relies on natural gas and coal. Groups like Clean Wisconsin are pushing for fully audited water and energy reports, not just marketing comparisons.

Shaking Off Foxconn’s Ghost

To understand the local mix of hope and doubt, you have to know the site’s history. In 2017, Foxconn promised a $10 billion LCD manufacturing campus that would bring up to 13,000 jobs to Mount Pleasant. Wisconsin offered over $3 billion in subsidies. The project withered: Foxconn delivered just 281 jobs and never built the promised factory. It left a hole in the local economy and a deep distrust of grandiose tech pledges.

Microsoft entered the picture in 2024, announcing a $3.3 billion AI datacenter on part of the vacated land. But even that hit a speed bump: in January 2025, the company said it would “evaluate scope and recent changes in technology,” prompting fears of another pullout. The September 2025 announcement of a second, even larger datacenter—and an expanded total investment—is Microsoft’s attempt to prove its commitment. Brad Smith, Microsoft’s president and a Wisconsin native, has framed the project as personal, recalling his childhood in Mount Pleasant. Still, local officials and residents are now well-rehearsed in asking for binding guarantees, not handshake deals.

Your Action Plan

If you live in Wisconsin or care about how hyperscale AI reshapes local communities, here’s what you can do now:

  • Stay informed. Watch for public meetings held by Wisconsin’s Department of Natural Resources (DNR) and the Public Service Commission. They will review water permits, environmental impact statements, and grid interconnection studies. These documents will contain real numbers, not corporate rhetoric.
  • Ask for audits. Community and advocacy groups should push for independent verification of Microsoft’s water-use figures, actual GPU counts, and the progress of renewable energy projects. The more transparent the process, the lower the risk of repeating Foxconn-style disappointments.
  • For developers and IT buyers: Microsoft is unlikely to open Fairwater for general cloud usage for at least a year—construction alone takes time. But you can start now by testing smaller-scale GB200 instances on Azure when they appear, building prototypes that can scale later. Keep an eye on NVIDIA’s own DGX cloud offerings as a comparison point.
  • For utility professionals and regulators: The Midwest grid operator, MISO, must model the aggregated load carefully. Push for binding interconnection studies and cost-allocation rules that protect residential customers. Microsoft’s pre-payment offers are welcome, but they shouldn’t replace rigorous regulatory oversight.

What to Watch Next

Fairwater’s real test will come in the next 12 to 24 months. Here are the signposts that will tell you if the project is delivering on its promises:

  • Benchmarks, not boasts. Microsoft must publish transparent, reproducible performance data for defined AI workloads. Until then, the “10x” figure is marketing.
  • Grid integration reports. MISO and We Energies will release detailed studies that show whether the grid can handle the load without raising rates or causing brownouts. Follow those dockets.
  • Water audits. The DNR will issue permits and require monitoring reports. Look for actual water consumption numbers, not restaurant analogies.
  • Community deliverables. Track enrollment and job placement numbers from the Datacenter Academy, and watch for local supplier contracts. These are concrete signs that economic benefits are staying in the region.

Microsoft’s Fairwater campus is an audacious bet on AI’s future and a test case for how America builds the physical backbone of the AI economy. It holds genuine promise: more compute could mean better AI tools for healthcare, science, and productivity—tools that eventually filter down to everyday Windows users through services like Copilot. But that promise runs on a high-wire act of engineering, environmental stewardship, and community trust. For now, cautious optimism beats blind faith.