Power and Water Are the Limits
Investors are pressing Amazon, Microsoft and Google over the water and power demands of proposed data centres, turning what looked like a permissionless AI build-out into a local political and utility challenge (insurancejournal.com). That squeeze is already reshaping vendor value: Motley Fool argues companies like Bloom Energy are benefiting because distributed power solutions can make AI clusters more resilient to local utility bottlenecks (fool.com).
What looked like a simple race to build more artificial intelligence data centers is turning into a fight over two old utilities: electricity and water. More than a dozen investors are pressing Amazon, Microsoft, and Alphabet’s Google for site-level details on how much of both their new United States facilities will consume. (insurancejournal.com) The pressure is landing after all three companies recently walked away from multibillion-dollar data center projects that ran into local opposition. Reuters reported the flashpoints were not abstract climate goals but concrete complaints about power demand, water use, pollution, and land. (insurancejournal.com) A modern data center is a warehouse full of computers that run hot around the clock, so it needs huge amounts of electricity to power chips and huge cooling systems to stop those chips from overheating. In North America, data centers used nearly 1 trillion liters of water in 2025, a volume Reuters said was roughly equal to New York City’s annual demand. (insurancejournal.com) The investor argument is that companywide sustainability reports no longer answer the local question a town asks when a new campus shows up: how much water will this site pull, and where will the power come from. Trillium Asset Management said Alphabet’s current climate disclosures do not show enough about water strategy and local grid effects for proposed projects. (energynow.com) The companies have started changing the plumbing because the old model is getting harder to defend. Reuters said Meta, Google, Amazon, and Microsoft have all begun using closed-loop cooling systems, which recirculate water instead of constantly drawing fresh supplies, but the public reporting on those systems still varies widely by company. (insurancejournal.com) That uneven reporting is part of the conflict. Meta’s 2025 environmental report disclosed water use for sites it owned, but not for leased sites or projects still under construction, and Amazon reports water use per unit of power rather than total water consumption for each facility. (insurancejournal.com) Power is becoming the other bottleneck because a giant artificial intelligence cluster cannot wait years for a utility to add transmission lines, substations, and generation. That delay is why the value is starting to shift from land and buildings toward companies that can bring electricity to the site faster. (fool.com) Bloom Energy sits in that gap with solid oxide fuel cells, which are box-sized power plants that generate electricity on site instead of waiting for a full grid connection. Motley Fool argued that pitch is landing because artificial intelligence operators will pay for resilient, always-on power if the local utility cannot deliver enough capacity on time. (fool.com) Bloom’s appeal is speed as much as fuel. CNBC reported in January that the company had become a standalone onsite power supplier for electricity-hungry artificial intelligence data centers, offering an immediate alternative to strained public grids, and Reuters-linked coverage this week showed why that matters as local permitting fights intensify. (cnbc.com) (insurancejournal.com) So the new map of the artificial intelligence build-out is less about who has the boldest model announcement and more about who can secure megawatts, cooling water, and community approval for one address at a time. The companies still have the money, but towns, utilities, and investors are now forcing them to show the wiring diagram and the water bill before the concrete gets poured. (insurancejournal.com)
