Servers generate heat constantly. Left unmanaged, that heat destroys hardware within minutes, so every data center must continuously remove it. As covered in Essay #1, the most common and cost-effective way to do this at scale is evaporative cooling — running water through cooling towers where it absorbs heat and evaporates into the air. That evaporated water doesn’t come back. It has to be replaced, continuously, for as long as the facility operates. This is the simple mechanical fact behind one of the most contested resource fights in America today.
Water enters a data center’s footprint in three distinct ways, and understanding all three is essential to understanding the full scope of the problem.
The numbers involved are difficult to fully grasp, but a few figures put it into perspective. A federal report estimated that the indirect water consumption footprint from electricity use by U.S. data centers was roughly 211 billion gallons in 2023 alone — and that figure covers only the power-generation portion of the footprint, not on-site cooling.
This is not a hypothetical projection. It is already happening at facilities operating today, and the trajectory points sharply upward. A 2025 analysis from the sustainability research group Ceres projected that annual water use tied to data center electricity demand will increase 400 percent in the coming years — from roughly 2.9 billion gallons to more than 14.5 billion gallons. Water use specifically tied to on-site cooling operations is projected to grow even faster: an 870 percent increase, from 385 million gallons a year to more than 3.7 billion gallons, enough to supply a city the size of Flagstaff, Arizona for nearly two years.
No single place illustrates the scale of this problem better than Loudoun County, Virginia — the world’s largest concentration of data centers, often called “Data Center Alley.” In 2023 alone, Loudoun County supplied an estimated one billion gallons of water to its data centers, and most of that was treated, drinkable municipal water rather than reclaimed wastewater, because the county’s reclaimed water infrastructure simply wasn’t built to handle the demand.
That is the core of why residents in data center host communities get alarmed: when a facility draws from the municipal water system, it is drawing from the same supply that fills bathtubs, irrigates lawns, and keeps fire hydrants pressurized. In a drought, that competition becomes a crisis. In a non-drought year, it is still a quiet, continuous transfer of a public resource to a private industrial user — often at a price far below what residential customers pay per gallon.
The geography of the current data center boom makes the water problem substantially worse, because much of the explosive new construction is happening in the driest part of the country. As of early May 2026, more than 60 percent of the lower 48 states were in drought conditions, with the West facing the most severe shortages. Arizona, Colorado, Idaho, Nevada, New Mexico, Oregon, Utah, and Wyoming all recorded their lowest snowpack levels since modern monitoring began in the 1980s. The Colorado River system — the water lifeline for 40 million people across seven states — is at historic lows, with Lake Mead at roughly a third of capacity and Lake Powell below a quarter.
This is precisely the region where data center developers are building most aggressively, drawn by cheap land, available power, and accommodating regulators. The result is a direct collision: an industry whose core infrastructure depends on water arriving in exactly the places that have the least water to give.
“Hyperscale data centers pose many harms to our communities, including enormous consumption of electricity, loss of critical farmland, and unrelenting noise pollution. When we focus on water consumption, however, the data center buildout becomes immensely more alarming.” — Food & Water Watch analysis, Colorado River basin, 2026
Many data centers don’t draw from municipal systems at all — they drill directly into local groundwater, pumping from the same aquifers that farmers and rural homeowners rely on for wells. Heavy, sustained pumping by an industrial user can lower the water table across an entire region, a process that is often slow, invisible, and difficult to reverse. Unlike a river or reservoir, an aquifer that has been over-pumped for years can take decades to recover, if it recovers at all.
This dynamic has put data center developers in direct conflict with the agricultural sector in several states, since farms and data centers frequently compete for the same underground water resources. The math is stark: a data center generates dramatically more revenue per gallon of water than a farm does, but a farm produces food and supports a community’s economic base in ways that a server room — staffed by a few dozen people — does not.
To their credit, the major tech operators have responded to public pressure with sustainability pledges. Several, including Google and Microsoft, have committed to becoming “water positive” by 2030 — meaning they aim to replenish more water than they consume through investments in wetland restoration, aquifer recharge projects, and community water programs. Google has stated a goal of replenishing 120 percent of the water it consumes in the regions where it operates.
These commitments are worth acknowledging, and some of the underlying projects are genuinely beneficial. But they deserve scrutiny on several fronts. Replenishment pledges are typically measured at a global or company-wide level, not at the local watershed level where the actual depletion is occurring — meaning a company can fund a wetland restoration project a thousand miles from a community whose well levels are dropping, and count it toward the same sustainability target. The timelines are also long, the verification is largely self-reported, and the pledges are voluntary commitments rather than binding regulatory requirements.
The good news, covered in greater depth in Essay #14, is that water-intensive evaporative cooling is not the only option. Air-cooled designs and closed-loop systems use dramatically less water, though typically at the cost of higher electricity consumption. Liquid immersion cooling, still relatively rare in large-scale deployment, can cut both water and power use significantly. The technology to build less thirsty data centers exists. What’s missing, in most states, is any regulatory requirement that developers use it.
Water is not an abstract environmental talking point in this fight — it is the single resource constraint most likely to determine whether a community can sustain both its existing population and a new generation of hyperscale data centers. In wetter regions, the tradeoffs are real but manageable. In the drought-stricken West, where most new construction is concentrated, the math increasingly doesn’t work. Communities weighing a proposed facility have every right to demand specific, binding answers about water sourcing and consumption before approval — not vague sustainability language attached after the fact.