Policy

Data Centers Face Impossible Choice: Water Use or Energy Demand

AI infrastructure operators must pick between depleting local water supplies and straining electric grids, with neither option winning public support.

Omega Editorial· July 9, 2026· 4 min read

The cooling dilemma

Artificial intelligence infrastructure has created an unavoidable resource trade-off. Next-generation computer chips generate so much heat that cooling them requires either millions of gallons of water or enormous amounts of electricity — and tech companies must choose which resource to tax.

The physics are straightforward: AI chips overheat without active cooling, just like any laptop pushed to its limits. A 2022 heat wave in the United Kingdom forced Google and Oracle to shut down cloud facilities when temperatures exceeded 100 degrees Fahrenheit. But the solutions create their own problems.

Water-based evaporative cooling can reduce a data center's power demand by up to 41,000 megawatt-hours, according to Google's own calculations — enough electricity to power 10,000 homes. The catch: that savings requires 264 million gallons of water. Choose air-based cooling instead, and energy consumption surges along with carbon emissions.

Why it matters

This resource conflict is fueling widespread opposition to data center development. Seven in ten Americans now oppose these facilities, with water use cited as a top concern, according to recent Gallup polling. In early 2026 alone, at least 75 data center projects worth $130 billion faced disruption from local resistance. County-level construction moratoriums are becoming routine. The industry's opacity about its cooling methods — and the lack of federal disclosure requirements — has only deepened public mistrust at a time when AI infrastructure buildout is accelerating.

How tech giants are choosing

Major operators are taking divergent approaches. Microsoft and Quality Technology Services have committed to zero-water cooling systems, accepting higher power consumption in exchange. Google and Amazon use water-based cooling to reduce grid strain in temperate climates, but avoid it in drought-prone regions like Phoenix and Cape Town.

Amazon Web Services plans a Louisiana facility that will use outside air cooling 87 percent of the year, switching to evaporative cooling only during peak summer heat when the electric grid is most constrained. The company reduced its North American water use by 946 million liters in 2024 — enough drinking water for 1.3 million people annually — while improving water efficiency by 17 percent.

Google's global data center fleet consumed 8.1 billion gallons in 2024, equivalent to irrigating 54 golf courses. Amazon used 2.5 billion gallons in 2025, roughly 5 percent of Seattle's annual consumption.

The closed-loop confusion

Newer AI chips generate even more heat, prompting companies to add closed-loop liquid cooling systems that run fluid through cold plates touching the chips directly. These systems have become a marketing point for developers promising "water-free" cooling.

But the term can mislead. While closed loops don't increase water use once filled, operators still must remove heat from both the liquid loop and the building itself. Amazon and Google use evaporative cooling for this heat rejection. Microsoft dissipates heat through air cooling, genuinely eliminating water use but consuming more electricity.

"There's a lot of confusion out there" about closed-loop systems, said Beau Schlitz, who leads water infrastructure initiatives at AWS. Communities often don't know to ask the specific question: How is the cooling plant rejecting heat?

The transparency gap

No federal law requires data centers to disclose their cooling methods or resource consumption. Companies that do share information through sustainability reports use inconsistent metrics and varying detail levels, making comparisons nearly impossible.

"No other major U.S. energy-consuming sector suffers from as many public data blind spots as U.S. data centers," Eric Masanet, a data center expert at the University of California, Santa Barbara, testified before Congress in February.

Two-thirds of U.S. data centers built since 2022 are located in areas experiencing high water stress, according to one recent analysis.

These details were first reported by POLITICO's E&E News.

#data centers#ai infrastructure#water consumption#energy efficiency#evaporative cooling#sustainability

This is an original analysis by the Omega editorial team. Source reporting: AI Watch.

Want systems like this working for your business?

Book a Call

More in Policy

Policy· 3 min read

Tech's $120B Off-Balance-Sheet Datacenter Debt Isn't Enron 2.0

A veteran accountant who worked through similar financing structures in the 1980s biotech boom explains why today's datacenter buildout poses different, more manageable risks.

Via AI Watch · Aug 23, 2026
Policy· 2 min read

State Farm's Outside Counsel Filed AI-Hallucinated Case Law

A retired attorney questions the oversight failures that allowed fabricated legal citations to reach a Los Angeles County court.

Via AI Watch · Aug 23, 2026
Policy· 3 min read

Flock surveillance cameras become flashpoint in 2026 midterms

Congressional candidates are attacking AI-powered license plate readers alongside data centers as voter anger over surveillance technology surges.

Via AI Watch · Aug 23, 2026