Water Wars in the AI Age – Data Centers vs. Drought

The “future” isn’t the first thing you notice when you’re close to a hyperscale data center. It’s the simplicity. A low, windowless building that squats behind cameras and fences, with a half-empty parking lot and a steady, mechanical hum permeating the air.

There is frequently a second feeling if it is summer in a dry location: heat that is motionless and shimmering over asphalt as the building silently battles its internal weather, dragging cooler air in and forcing warmer air out hour after hour.

Add drought now. Put up lawn signs urging neighbors to practice conservation. Add a nearby reservoir that is encircled by bare, white shoreline that resembles a bathtub that has been over-drained. At that point, the AI boom begins to feel more like a resource dispute than a technology story, with the “cloud” suddenly having a water meter and a zip code.

Perhaps the buildout has become blunt, which is why the tension is at its highest. More than 160 new AI data centers were built in the US in a short period of time, according to Bloomberg, with many of them ending up in areas already vying for limited water supplies.

The reasoning is well-known: inexpensive land, amicable permits, fiber and power availability. However, the local response is evolving. When it comes to something that feels personal, like a shower that doesn’t work well in the morning, people can put up with a lot of invisible infrastructure.

A single pipe supplying a single building is how the water problem is frequently explained. It isn’t. In addition to directly drawing water for cooling, data centers also indirectly draw water through the electricity they use, particularly when that power is generated using a lot of water. Years ago, a team from the University of California, Riverside stated clearly that cloud computing uses water both on-site and upstream at power plants.

In “Making AI Less ‘Thirsty,'” researchers have attempted to quantify the thirsty footprint of the supply chain, which includes the manufacturing of chips and servers.

Evaporative cooling is the most contentious detail and the one that is brought up in community meetings. It functions well in hotter climates and is energy-efficient. Additionally, it produces local vapor from local water. According to experts quoted by Bloomberg, data centers typically evaporate the majority of the water they use, which is typically estimated to be around 80%, and only a smaller portion is returned for treatment.

Despite the engineers’ insistence that the math is more intricate, people visualize waste when they hear the word “evaporate.”

It is easier to picture the conflict when you travel south to Querétaro, Mexico. The city is well-known for its lengthy stone aqueduct, which is a reminder that water has always played a political role in this area, with arches marching across the landscape.

As residents worry about drought, blackouts, and who will be responsible for the strain, the area has also become a data center magnet in recent years, attracting big-name tech interest. A community asks where the water will come from, the government sells growth, and the answers are given in technical terms that most people don’t trust. This creates an awkward rhythm to the story.

The atmosphere is a little different in the American West; it’s more frontier and “build it fast.”

For example, the Nevada desert has been marketed as a new type of AI infrastructure boomtown, causing local advocates and tribes to express concern about already overburdened water systems. It appears as though the AI economy is resurrecting the old extractive instincts of mining towns, albeit with cleaner branding and fewer visible workers, as the pattern of industry rushing to places with cheap land and flexible regulations is repeated.

To be fair, tech companies aren’t ignoring the issue. In order to reduce evaporation, some are moving toward closed-loop water cooling and higher-temperature systems. According to Reuters’ Breakingviews, when those methods take the place of conventional evaporative setups, there can be significant savings—up to about 90% in some situations. It sounds comforting. The speed at which older facilities are retrofitted when the incentives are skewed toward building new capacity is still unknown.

Then there is the silent accelerant, the transparency issue. Annual totals that are vague enough to avoid local scrutiny and clean enough to appear responsible are frequently used for water reporting. For instance, Google’s most recent environmental reporting provides a precise estimate of the amount of water used, which is projected to be around 8.1 billion gallons in 2024.

It also details the sources of withdrawals and the classification of risk. Although helpful, it doesn’t always provide the answer to a town’s main query: what would happen to our watershed if three more facilities were to be built?

This fight is about more than just gallons, which is an uncomfortable reality. It has to do with legitimacy. When people feel safe and included, they are willing to make compromises. When they feel controlled, they rebel.

This disparity is exacerbated by AI, which is both glamorous and opaque. Unless you’re already in the tech industry, the benefits seem abstract, the servers are hidden behind fences, and the employment figures frequently appear small in comparison to the footprint.

In ten years, the industry might reflect on today’s water outrage and see it as a short-term growing pain that can be resolved with better planning, cooling, and possibly even regulation. Or it could appear to be the early warning that it was. In any case, neither the AI buildout nor the drought are waiting. The debate has already shifted from the question of “how much water?” to the more pertinent one of “who gets priority when the river runs low?”