Key Takeaways

  • Local officials approved a major expansion of Microsoft’s datacenter footprint in Wisconsin despite ongoing environmental concerns.
  • Internal forecasts show the company’s water use rising sharply by 2030, even after revised conservation projections.
  • Broader questions continue to surface about the sustainability of the global AI datacenter boom.

The approval landed with little fanfare, but it marks a significant step in the ongoing expansion of large-scale AI infrastructure. This week, local officials in Mount Pleasant, Wisconsin, signed off on Microsoft’s plan to build additional server farms near its existing datacenter campus. The project, part of an announced $3.3 billion investment, extends an already massive footprint that the company has described as on track to become one of the world’s most powerful AI datacenter hubs.

The approval comes at a time when water consumption by hyperscale facilities is under tight scrutiny. Cooling high-density AI systems requires substantial volumes of water, and that demand is only projected to increase.

The Wisconsin development sits on land initially set aside for Foxconn, a reminder of how quickly economic development plans can pivot when hyperscalers show interest. According to CNBC, the village board supported the expansion despite some pushback from community members concerned about long-term environmental impacts. Whether this signals a broader shift toward accommodating AI infrastructure at almost any cost remains an open question.

Elsewhere, Microsoft continues to accelerate its European build-out. It has committed ÂŁ2.5 billion (approximately $3.2 billion) for AI datacenter expansion in the United Kingdom between 2024 and 2027, alongside significant multibillion-dollar investments in Germany and Spain. The company has said it plans to drastically increase its datacenter capacity across Europe. That pace is blistering, even by cloud-era standards.

Of course, momentum rarely comes without consequences. The company has already acknowledged that its emissions were up roughly 30 percent from a 2020 baseline, attributing the rise in part to new datacenter construction. The water challenge may be even more stark. Reporting from the New York Times indicated Microsoft’s internal forecasts show annual water consumption across global campuses potentially tripling to 28 billion liters by 2030. Microsoft later said updated projections incorporating new conservation techniques reduce that estimate to around 18 billion liters—still a massive increase from 2020 levels.

Not everyone is convinced the mitigations are enough. Late last year, more than 230 organizations across the United States signed a letter calling for a national moratorium on new datacenter construction. Their argument hinges on the idea that the current build-out—driven in large part by AI workloads—poses mounting environmental and social risks, from water depletion to energy grid strain.

A brief tangent here: the political dimension is becoming harder to ignore. Regulators and political leaders have increasingly scrutinized major technology companies regarding the impact of infrastructure costs on local communities and utility rates. This pressure helped spur Microsoft’s Community-First AI Infrastructure initiative, announced earlier this month, which includes commitments to reduce and replenish water use and increase transparency around consumption.

Transparency itself is becoming an issue. Some local officials across the U.S. have been asked to sign non-disclosure agreements related to datacenter developments, making it difficult for communities to understand the impact of these large facilities. The Register recently reported on such NDAs, which have the effect of keeping certain operational details out of public debate. Whether this is sustainable over the long term is unclear.

Interestingly, not all industry players face the same level of scrutiny. Smaller colocation operators often build in dense urban areas with established utility infrastructure, while hyperscalers tend to sprawl across rural or semi-rural land, where water tables and grid capacity may be more vulnerable. That split is becoming a flashpoint in the broader conversation about AI’s physical footprint.

Inside Microsoft, vice chair and president Brad Smith emphasized the company’s plans to improve local transparency, saying communities deserve clear information about datacenter water usage. But translating that promise into actionable, timely reporting, especially across multiple geographies, may prove challenging.

Still, demand for AI infrastructure shows no signs of slowing. The Bank of England’s Financial Policy Committee has already warned about the systemic risks associated with the rapid growth in AI investment, but the build-out continues. Companies are racing to meet customer appetite for model training and inferencing capacity, even if the environmental calculus isn’t fully resolved.

That said, the current pace won’t continue forever. If AI spending cools, or if regulatory pressure tightens, approvals like the one in Wisconsin may face tougher scrutiny. For now, though, hyperscale expansion continues to surge forward—even as questions about sustainability grow louder.