Water has become the most contested question in the data center debate. As AI drives a construction boom into communities across the country, residents, regulators, and reporters are all asking the same thing: how much water do these facilities actually use? The answer is not simple, because it depends almost entirely on how a facility is cooled, and because a large share of the water is hidden upstream at the power plant. This is a plain-numbers reference on data center water use in 2026, what the real figures are, where they come from, and what is changing the math.
TL;DR
Data center water use varies enormously by facility size and cooling design. The essentials:
- A medium data center uses roughly 300,000 gallons of water per day. That is comparable to the daily water use of about 1,000 US households, and works out to roughly 110 million gallons a year for a single medium-sized site.
- A large hyperscale facility can use 1 to 5 million gallons per day. In hot climates and with evaporative cooling, the largest facilities reach the top of that range; Google’s largest US site has withdrawn on the order of 3.9 million gallons per day.
- A fully air-cooled data center uses almost no water on site. The cooling method is the single biggest driver of water use, and it is a direct tradeoff with electricity: air cooling saves water but uses far more power.
- Most of the water is consumed, not just borrowed. In evaporative cooling, roughly 70 to 85 percent of the water is lost to evaporation and never returns to the local supply; only the remainder is discharged as wastewater.
- A large share of the water is off-site and hidden. Beyond on-site cooling, data centers consume water indirectly at the power plants that supply their electricity, so air-cooled facilities push water use upstream rather than eliminating it.
The single most important thing to understand: there is no one number for data center water use, because water use is a design decision. A facility’s cooling system and local climate determine whether it consumes millions of gallons a day or almost none, and every choice trades water against electricity or capital cost.
The Headline Numbers
The most-asked question has a range, not a single answer, and the range is wide because facilities differ by orders of magnitude.
| Facility type | Typical on-site water use | Rough comparison |
|---|---|---|
| Medium data center (evaporative) | ~300,000 gallons per day (~110 million gallons per year) | About the daily use of 1,000 US households |
| Enterprise data center | ~300,000 to 500,000 gallons per day | A mid-sized municipal water customer |
| Large hyperscale (hot climate, evaporative) | ~1 to 5 million gallons per day | A small town’s daily water use |
| Largest individual hyperscale sites | Up to ~3.9 million gallons per day withdrawn (real reported figure) | Thousands of households |
| Air-cooled or closed-loop facility | Near zero on-site (water use shifts upstream to power) | Negligible direct consumption |
A few things anchor these figures. A medium-sized facility consuming around 300,000 gallons a day is the most commonly cited benchmark, roughly the daily water use of a thousand homes, a figure documented in university extension research. Enterprise facilities often run somewhat higher, in the 300,000-to-500,000-gallon range. And the largest hyperscale campuses, particularly in hot, dry climates where evaporative cooling works hardest, can reach 1 to 5 million gallons per day. These are not hypothetical: one of Google’s largest US data centers, in Council Bluffs, Iowa, has withdrawn an average of about 3.9 million gallons of water per day and consumed roughly 2.8 million of that, per Google’s own environmental report.
The key caveat on every one of these numbers is that they describe water-cooled facilities. A data center that uses air cooling or a closed loop uses almost no water on site, which is why any honest answer to “how much water does a data center use” has to start with “which kind.”
Why Cooling Method Decides Everything
The reason data center water use ranges from millions of gallons to nearly zero is that cooling is the single biggest factor, and the main cooling methods sit on a spectrum that trades water directly against electricity.
Evaporative Cooling: Low Power, High Water
Evaporative cooling (including cooling towers and adiabatic systems) works by evaporating water to carry away heat, the same principle that makes sweating cool your skin. It is highly energy-efficient, which is why it became the industry default: it keeps a facility’s power-usage effectiveness low. But it is water-intensive, because the whole mechanism depends on evaporating water. Roughly 75 to 90 percent of data centers worldwide have relied on water-based cooling as their primary method. In an evaporative system, roughly 70 to 85 percent of the water withdrawn is lost to evaporation and never returns to the local supply, with only the remaining 20 to 30 percent discharged as wastewater. This is why evaporative facilities dominate the alarming water-use headlines.
Air Cooling: High Power, Low Water
Air-cooled systems (large chillers and HVAC-style equipment) use almost no water. But that near-zero water figure is misleading if read in isolation, because air cooling uses substantially more electricity to run compressors and fans. And more electricity means more water, just somewhere else: at the power plant. Air cooling does not eliminate a facility’s water footprint so much as relocate it upstream. It is the right choice in water-stressed regions, but it raises both the power bill and the carbon footprint.
Closed-Loop and Zero-Water: Capital for Water
Closed-loop systems recirculate the same coolant continuously instead of evaporating it, so they consume only a small fraction of the water an evaporative system withdraws, on the order of 5 to 10 percent, because the loop is filled once and reused rather than constantly replenished. The newest direct-to-chip and immersion designs push this further to near-zero evaporative water, circulating a sealed coolant (often a non-toxic glycol-based fluid) directly to the chips. These systems trade water for capital cost and, in some cases, higher energy or equipment complexity, but they are increasingly the industry direction, especially in dry markets. The tradeoff is the whole story: there is no free option, only a choice of which resource to spend.
The cooling tradeoff, summarized:
| Cooling type | On-site water | Electricity | Best suited to |
|---|---|---|---|
| Evaporative | High | Low | Water-abundant, cost-sensitive markets |
| Air-cooled | Near zero on-site | High | Water-stressed markets (shifts water to power) |
| Closed-loop | Low (~5-10% of evaporative) | Moderate | Balanced water and power constraints |
| Direct-to-chip / immersion | Near zero | Low to moderate | High-density AI, water-stressed markets |
The Hidden Water: Electricity and Chips
The on-site cooling figure is only part of a data center’s true water footprint. Two large sources of water use sit outside the facility fence and are routinely left out of the debate.
Power-Plant Water
Generating electricity consumes water, especially at thermal power plants that use water-intensive steam cycles to drive turbines. Because data centers are enormous electricity consumers, they indirectly consume large volumes of water at the plants that supply them. The 2024 US Data Center Energy Usage Report from Lawrence Berkeley National Laboratory estimated that in 2023 US data centers consumed roughly 17 billion gallons of water directly through cooling, plus an additional 211 billion gallons indirectly through the electricity that powered them, an order of magnitude more water hidden upstream than the on-site figure shows. This is the crucial catch in the air-cooling story: an air-cooled facility that reports near-zero on-site water is still driving significant water consumption upstream through its higher electricity demand. Any honest accounting of data center water has to include this indirect, off-site share, which is why “we use no water” claims based only on on-site cooling can be misleading.
Chip Manufacturing Water
There is also embedded water in the hardware itself. Semiconductor fabrication is extraordinarily water-intensive: producing the ultrapure water used to make chips takes roughly 1.5 gallons of regular water per gallon of ultrapure water, on top of the large volumes fabs already consume. Every server rack in a data center therefore carries a supply-chain water cost incurred long before it is installed. This embedded water rarely appears in facility-level water debates, but it is a real part of the total footprint of the AI buildout.
The practical implication: a facility’s true water footprint is on-site cooling water plus power-plant water plus embedded manufacturing water. Focusing only on the first can either overstate the problem (for air-cooled sites that look clean on site but drive upstream use) or understate it (by ignoring the power and supply-chain water entirely).
Putting the Numbers in Context
Because the raw figures are hard to intuit, data center water use is usually explained through comparisons, and those comparisons are where a lot of the public debate gets muddled. Here are the ones that hold up.
A medium data center using ~300,000 gallons a day uses about as much water as 1,000 US households. That framing is accurate and useful.
Comparisons to other large water users are where context matters most, and most of the common ones have limits worth understanding. Data centers are sometimes compared to golf courses, but that comparison only works if you account for scale, since a typical golf course uses on the order of one-tenth the water of a medium AI data center, though there are far more golf courses than data centers. The most apt single-facility comparison is agriculture: the average large almond farm uses roughly as much water as a medium-sized AI data center, and agriculture remains by far the largest water user overall. Matching comparisons to the right scale, per facility versus per sector, is what keeps the water conversation grounded rather than alarmist in either direction.
At the industry level, the aggregate picture is what is driving attention. Data centers already rank among the ten most water-intensive commercial and industrial sectors in the United States, and projections suggest US data center on-site water use could grow by two to four times between the mid-2020s and 2028. The dynamic is less about any single facility than about the number of facilities arriving at once, sometimes concentrated in water-constrained regions, which is exactly why cooling design has become such a central consideration in where and how new capacity gets built.
The Fixes Changing the Math
The water question is not static. Operators themselves have driven most of the progress, and the industry is actively re-engineering the problem, so the 2026 picture is meaningfully different from a few years ago.
Closed-Loop and Zero-Water Designs
The biggest shift is the move away from evaporative cooling in new builds. Major operators have begun designing facilities that consume little or no water for cooling by using closed-loop and direct-to-chip liquid systems. The reported impact is large: zero-water evaporation designs can cut a facility’s annual water consumption by well over 100 million liters compared with evaporative designs, and design changes at desert facilities have cut water use substantially. Industry commentary in late 2025 and 2026 points toward zero-water-evaporation cooling becoming the standard for new data centers, a genuine structural change rather than a marginal efficiency gain.
Reclaimed and Non-Potable Water
A second fix is switching the water source rather than the volume. Many new facilities are being designed to use reclaimed water (treated wastewater) or other non-potable sources instead of drinking water, so their consumption does not draw down the municipal drinking supply. This does not reduce total water use, but it changes which water is used, addressing the most acute community objection, that data centers compete with residents for drinking water.
Replenishment Commitments
Third, major operators have made water-replenishment or water-positive commitments, pledging to restore more water to stressed watersheds than their operations consume, through watershed-restoration and reuse projects. Water-reuse projects have cut potable water use at some facilities dramatically and returned large volumes to community use. These commitments represent significant investment and real progress. They also tend to operate at the portfolio level, so a common point of discussion is how portfolio-wide replenishment maps to any individual watershed, which is part of why operators increasingly pair global commitments with facility-level and community-level water strategies.
The Density Connection
Finally, the shift toward liquid cooling is tied to the AI hardware itself. The high-density GPU racks that AI training demands generate more heat than air cooling can efficiently remove, which is pushing facilities toward direct-to-chip and immersion liquid cooling. That transition happens to reduce evaporative water use, so the same density pressure driving the AI buildout is also, indirectly, moving the industry toward lower-water cooling. The move to liquid cooling is a hardware-refresh event as much as a plumbing one, and facilities retrofitting for it are also cycling through the cooling and compute equipment that a water-focused redesign displaces.
Frequently Asked Questions
How much water does a data center use per day?
It depends heavily on size and cooling method. A medium-sized data center using evaporative cooling consumes roughly 300,000 gallons of water per day, comparable to the daily use of about 1,000 US households. Enterprise facilities often use 300,000 to 500,000 gallons per day, and large hyperscale facilities in hot climates can use 1 to 5 million gallons per day, with the largest individual sites withdrawing close to 4 million gallons daily. By contrast, a fully air-cooled or closed-loop facility uses almost no water on site. There is no single number, because cooling design and local climate drive the difference by orders of magnitude.
Why do data centers use so much water?
Data centers use water primarily to cool the servers, which generate enormous heat running continuously. The most common method historically has been evaporative cooling, which removes heat by evaporating water, the same principle as sweating. It is energy-efficient but water-intensive, and roughly 75 to 90 percent of data centers have used water-based cooling as their main method. The AI boom has intensified this because high-density AI hardware produces far more heat than earlier equipment. Data centers also consume water indirectly through the power plants that supply their electricity, so their true water footprint extends well beyond on-site cooling.
How much water does a hyperscale data center use?
Large hyperscale data centers can consume 1 to 5 million gallons of water per day, with the top of that range reached in hot, dry climates using evaporative cooling. These are real figures: one of Google’s largest US data centers, in Council Bluffs, Iowa, has withdrawn an average of about 3.9 million gallons per day and consumed roughly 2.8 million of that. Across a full portfolio the numbers are large: Google’s global data centers withdrew on the order of 7.8 billion gallons in a recent year, consuming roughly 78 percent of it through evaporation. Individual site figures vary widely, from under 500,000 gallons per day at some sites to more than 2 million at others.
Is data center water use consumed or returned?
Most of the water an evaporative data center withdraws is consumed, not returned. In evaporative cooling, roughly 70 to 85 percent of the water is lost to evaporation and never returns to the local water supply; only the remaining 20 to 30 percent is discharged as wastewater. This distinction matters, because “consumption” (water withdrawn and not returned) is the figure that actually reduces a community’s available water, and it is higher for evaporative systems than the withdrawal figure alone suggests. Closed-loop systems, which recirculate the same water, consume only a small fraction of what they initially withdraw.
Do air-cooled data centers really use no water?
Air-cooled data centers use almost no water on site, but that does not make them water-free. Air cooling uses substantially more electricity than evaporative cooling to run compressors and fans, and generating that additional electricity consumes water at the power plant, especially at thermal plants using steam cycles. So air cooling relocates a facility’s water footprint upstream rather than eliminating it. Air cooling is often the right choice in water-stressed regions, because it protects the local water supply, but it raises the power bill and the carbon footprint, and its total water footprint (including power-plant water) is not zero.
How do data centers reduce water use?
Operators reduce water use through several approaches. Closed-loop and direct-to-chip liquid cooling recirculate coolant instead of evaporating it, cutting water consumption to a small fraction of evaporative designs, and newer zero-water-evaporation systems eliminate evaporative water almost entirely. Many facilities switch to reclaimed or non-potable water so they do not draw on drinking supplies. Major operators have made water-replenishment or water-positive commitments to restore water to stressed watersheds. And the industry shift toward liquid cooling, driven by high-density AI hardware, is moving new builds away from water-intensive evaporative cooling. Industry commentary points toward zero-water cooling becoming standard for new facilities.
How does data center water use compare to other industries?
Data centers already rank among the ten most water-intensive commercial and industrial sectors in the United States, and their footprint is growing. Single-facility comparisons help: a medium data center uses about as much water as 1,000 households, or roughly as much as a large almond farm. Comparisons to golf courses are often cited but incomplete, since a typical golf course uses roughly one-tenth the water of a medium AI data center. Agriculture remains the largest water user overall by far, but data center water use is concentrated in specific communities and growing rapidly (US on-site use is projected to grow two to four times by 2028), which is why it draws outsized local concern.
Does AI increase data center water use?
Yes, significantly. AI workloads require high-density hardware that generates far more heat than traditional servers, increasing cooling demand. Some analyses estimate AI-focused facilities can consume many times more cooling water than traditional server farms, and training large AI models consumes substantial water through the cooling of the facilities that run them. At the same time, AI’s high heat density is pushing facilities toward liquid and direct-to-chip cooling, which uses far less evaporative water, so AI is simultaneously increasing cooling demand and accelerating the shift to lower-water cooling methods. The net local effect depends on which cooling design a given AI facility adopts.
Where does a data center’s water actually go?
A data center’s water footprint has three parts. The largest visible part is on-site cooling water, most of which, in evaporative systems, evaporates and does not return to the local supply. The second, often overlooked, is power-plant water: the water consumed generating the electricity the facility uses, which is why air-cooled sites still have a real water footprint. The third is embedded water from chip manufacturing, since producing the ultrapure water used to make semiconductors consumes roughly 1.5 gallons of regular water per gallon of ultrapure water, on top of fabs’ already large use. A complete accounting includes all three, not just on-site cooling.
The Bottom Line
There is no single answer to how much water a data center uses, because water use is fundamentally a design decision. A medium facility using evaporative cooling consumes around 300,000 gallons a day, the water use of about 1,000 households; a large hyperscale site in a hot climate can reach 1 to 5 million gallons a day; and an air-cooled or closed-loop facility uses almost none on site. The cooling method decides everything, and it is a direct tradeoff: evaporative cooling saves electricity but consumes water, air cooling saves water but consumes electricity and pushes water use upstream to the power plant, and closed-loop and direct-to-chip systems trade capital cost for both. A facility’s true footprint is on-site cooling water plus power-plant water plus the embedded water in its hardware, so the simplest claims in either direction, that data centers are water catastrophes or that they use no water, are both usually incomplete.
What is changing is the engineering. The AI buildout is driving higher heat density that pushes new facilities toward liquid and zero-water cooling, reclaimed-water sourcing is displacing drinking-water use, and replenishment commitments are offsetting consumption at the portfolio level. Industry commentary now points toward zero-water-evaporation cooling becoming standard for new builds. The water problem is real, concentrated in specific and often water-stressed communities, and growing in aggregate as facilities multiply, but it is also being actively re-engineered, and the numbers for a data center built in 2026 look very different depending on which of these choices its designers made.
How ROC Telecom Fits In
ROC Telecom is an R2v3, RIOS, NIST 800-88, and ITAR-compliant ITAD specialist. Water and cooling are facility-operations questions rather than disposition ones, but the two connect at one point: the industry-wide shift from evaporative to liquid and direct-to-chip cooling is a hardware event as much as a plumbing one. Facilities retrofitting for liquid cooling, or refreshing high-density AI hardware that demands it, cycle through cooling and compute equipment that has to be decommissioned, sanitized, and recovered responsibly.
That is where ROC helps:
- Data center decommissioning with 48-hour rapid-response mobilization, including cooling-driven retrofits and high-density hardware refreshes
- Asset recovery across routing, switching, optical transport, and compute, with direct buyer relationships that protect residual value
- NIST 800-88 data destruction with per-asset serialized Certificates of Destruction and full chain-of-custody documentation
- R2v3 Appendix E materials recovery with in-house dismantling and direct-to-refiner processing
- Mass-balance recovery reporting for the ESG and sustainability disclosures that facility resource use increasingly demands
15+ years of ITAD experience, $25M+ in client capital recovered, 45M+ pounds diverted from landfill.
Request a Free ITAD Program Assessment
Tell us about your infrastructure and what you are building, retrofitting, or retiring. A specialist will reach out to discuss decommissioning timelines, recovery value, and certified disposition. No commitment, no spam.
Prefer to talk directly? Call 585-406-1249 or email info@roctelecom.com.
"*" indicates required fields
Related reading:
- What Most Data Centers Miss About E-Waste
- Nuclear Power and Data Centers: The 2026 Energy Overview
- Why AI Is Shortening Your Hardware Lifecycles (and What to Do About It)
- The Environmental Impact of Responsible Circuit Board Recycling
- Data Center Asset Recovery: The 2026 Strategic Guide
- E-Waste by the Numbers: Why It’s a Growing Problem
- The Phoenix Data Center Market: A 2026 Overview
