A single large data center can drink as much water in one day as a town of 30,000 to 50,000 people. That number surprises most people, because we picture data centers as rooms full of blinking servers, not as thirsty industrial facilities. But servers generate enormous heat, and water remains one of the cheapest, most efficient ways to carry that heat away. So when people ask how much water does a data center use per day, the honest answer ranges from almost zero to more than five million gallons, depending on the cooling design, the climate, and the size of the building.
This matters more every year. Artificial intelligence workloads are pushing power density higher, and higher power means more heat and more cooling. Communities in Arizona, Texas, Virginia, Oregon, and Chile have all pushed back on new projects over water concerns. In this guide, you will learn the real daily water numbers for small, medium, and hyperscale facilities, how the different cooling systems compare, what water usage effectiveness (WUE) actually measures, how AI training changes the math, which companies publish real data, and what the industry is doing to cut consumption. You will also get the context most headlines skip, including how data center water use compares to farming, power plants, and golf courses.
What Data Center Water Consumption Actually Means
A typical mid-sized data center uses roughly 100,000 to 300,000 gallons of water per day, while a large hyperscale campus can consume 1 million to 5 million gallons per day during hot weather, and some small facilities use no water at all because they rely entirely on air cooling. That wide spread confuses people, but it makes sense once you understand what the water actually does.
Almost all of that water goes to cooling. Servers convert electricity into heat, and that heat has to leave the building or the equipment fails. Many facilities use evaporative cooling, which works like sweating. Warm air passes over wet media or water sprays into a cooling tower, some of the water turns to vapor, and evaporation pulls heat out of the system. The evaporated water leaves the site permanently. That is called consumptive use, and it is the number that worries local water managers most.
A smaller share of water goes to other uses. Humidification systems add moisture to prevent static electricity. Some sites use water for fire suppression testing, landscaping, and restrooms for staff. But those uses rarely exceed a few percent of the total. Cooling drives the story.
It also helps to separate two different measurements. Withdrawal means the total water a facility pulls from a source. Consumption means the portion that never returns to that source, usually because it evaporated. A cooling tower might withdraw 100 gallons, evaporate 80, and discharge 20 as concentrated wastewater called blowdown. Reports that mix these two numbers create a lot of confusion in public debates.
- Evaporative cooling water: 80 to 95 percent of total site water use at most water-cooled facilities
- Blowdown and discharge: water sent to sewer after minerals concentrate, typically 15 to 25 percent of withdrawal
- Humidification: small but steady, often less than 3 percent
- Domestic and landscaping: usually under 2 percent, since data centers employ few people per square foot
- Indirect water: water used at power plants to generate the electricity the data center consumes, often larger than on-site use
Daily Water Use by Data Center Size and Type
Size drives everything. Data center capacity gets measured in megawatts (MW) of IT load, not square feet, because power determines heat and heat determines cooling. A useful rule of thumb: a water-cooled facility using evaporative towers consumes roughly 1.8 to 2.5 gallons of water per kilowatt-hour of IT energy in warm climates, and less in cool ones.
Run that math and the daily numbers appear quickly. A 1 MW facility running around the clock uses 24,000 kilowatt-hours per day. At 2 gallons per kilowatt-hour, that is about 48,000 gallons daily. Scale that to a 100 MW hyperscale campus and you land near 4.8 million gallons per day at peak. Real facilities vary because operators shut off evaporation on cool days and lean on outside air instead.
Typical Daily Water Use Ranges
| Facility Type | IT Load | Estimated Daily Water Use | Household Equivalent |
|---|---|---|---|
| Server closet or edge site | Under 0.1 MW | 0 to 500 gallons | Fewer than 3 homes |
| Small enterprise data center | 0.5 to 1 MW | 10,000 to 50,000 gallons | 60 to 300 homes |
| Mid-sized colocation facility | 5 to 15 MW | 100,000 to 500,000 gallons | 600 to 3,000 homes |
| Large single data center | 30 to 50 MW | 800,000 to 2 million gallons | 5,000 to 12,000 homes |
| Hyperscale campus | 100 to 300 MW | 3 million to 9 million gallons | 18,000 to 55,000 homes |
| Air-cooled or closed-loop facility | Any size | Near zero for cooling | Office building only |
Notice the last row. Water use is a design choice, not a law of physics. Operators who choose air cooling, closed-loop chillers, or refrigerant-based systems trade water for electricity. They use more power to run compressors and fans, but they pull almost nothing from the local water supply.
Here is a practical scenario. Imagine a 40 MW facility outside Phoenix. On a 108-degree July afternoon, its cooling towers run hard and the site consumes about 1.6 million gallons that day. In January, when outside air sits near 55 degrees, the operator switches to free cooling and the same site uses under 200,000 gallons. Average the year and you get maybe 700,000 gallons per day, which is why annual averages hide the peaks that stress local utilities.
How Different Cooling Systems Change the Water Math
The single biggest factor in daily water use is the cooling technology. Two identical buildings with identical servers can differ by a factor of 100 based on this choice alone. Let me walk through the main options and what each one costs in water and power.
Evaporative and Cooling Tower Systems
These systems dominate older and mid-generation facilities because they are cheap to run and very effective. Water flows over a tower, air moves through it, and evaporation dumps heat into the atmosphere. Water use runs high, but electricity use runs low. In dry climates, evaporative cooling works especially well, which is exactly why so many data centers landed in deserts where water is scarce. That irony sits at the center of most local disputes.
Air-Cooled Chillers and Direct Air Cooling
Air-cooled systems reject heat straight to the atmosphere with fans and refrigerant coils. They use almost no water. The tradeoff shows up on the power bill, since these systems can raise total facility energy use by 10 to 30 percent compared to evaporative cooling in hot climates. Operators in water-stressed regions increasingly accept that trade.
Closed-Loop and Liquid Cooling
Closed-loop systems circulate the same water or coolant over and over, topping off only small losses. Direct-to-chip liquid cooling brings coolant right to the processor, which handles high-density AI racks far better than air. A closed loop might hold a few thousand gallons total and refill only a fraction of that each year. Immersion cooling goes further, submerging servers in a non-conductive fluid with essentially zero water consumption during operation.
Hybrid and Adiabatic Systems
Hybrid designs run dry most of the year and switch to evaporative assist only on the hottest days. This approach cuts annual water use by 60 to 90 percent while keeping peak-day performance. Many new builds now specify hybrid coolers as the default.
| Cooling Type | Water Use per Day (10 MW site) | Relative Energy Use | Best Fit |
|---|---|---|---|
| Open cooling tower | 300,000 to 500,000 gallons | Lowest | Water-rich, hot regions |
| Adiabatic or hybrid | 30,000 to 150,000 gallons | Low to moderate | Mixed climates |
| Air-cooled chiller | Under 2,000 gallons | Higher | Water-stressed areas |
| Closed-loop liquid | Under 1,000 gallons | Moderate | High-density AI racks |
| Immersion cooling | Near zero | Lowest for high density | Extreme density, edge sites |
One more variable deserves attention: water source. Some facilities use potable drinking water, which draws the loudest criticism. Others use reclaimed wastewater, greywater, harvested rainwater, industrial process water, or even seawater. A facility using 2 million gallons per day of treated municipal wastewater affects the community very differently than one drawing the same volume from a drinking water reservoir.
Understanding WUE, PUE, and the Numbers Operators Report
If you want to compare facilities honestly, you need to understand the industry’s two main efficiency metrics. They show up in every sustainability report, and knowing them helps you spot both real progress and clever spin.
Water Usage Effectiveness (WUE) measures liters of water consumed per kilowatt-hour of IT energy. A WUE of 1.8 means the site uses 1.8 liters for every kilowatt-hour the servers consume. Lower is better. Industry averages sit around 1.8 liters per kilowatt-hour for traditional facilities, while leading operators report 0.2 to 0.5. Some air-cooled sites report near zero.
Power Usage Effectiveness (PUE) measures total facility power divided by IT power. A PUE of 1.5 means the building uses 50 percent more power than the servers alone. Here is the catch that trips up a lot of readers: WUE and PUE often push against each other. Cutting water usually raises energy use, and cutting energy often raises water use. Good operators optimize both together rather than gaming one metric.
How to Convert WUE Into Daily Gallons
- Find the facility’s IT load in megawatts. Multiply by 24 to get megawatt-hours per day.
- Multiply that by 1,000 to get kilowatt-hours per day.
- Multiply kilowatt-hours by the WUE figure to get liters of water per day.
- Divide liters by 3.785 to convert to gallons.
- Adjust for seasonality, since summer peaks can run double the annual average.
Try it with a real example. A 20 MW facility with a WUE of 1.8 uses 480,000 kilowatt-hours per day, which equals 864,000 liters, or about 228,000 gallons daily. Now drop the WUE to 0.3 through hybrid cooling and reclaimed water, and the same site falls to about 38,000 gallons per day. That is the difference good design makes.
Watch for reporting gaps too. Many companies publish only annual, company-wide totals that blend efficient cold-climate sites with thirsty desert campuses. Others report withdrawal instead of consumption, or exclude leased colocation space. When a headline cites a shocking number, check whether it covers one building, one campus, or an entire global fleet.
Why AI Workloads Are Rewriting the Water Equation
Traditional server racks drew 5 to 10 kilowatts each. AI training racks packed with GPUs now draw 40 to 130 kilowatts, and next-generation designs push past 250 kilowatts. That density change is the biggest shift in data center cooling in twenty years, and it cuts both ways on water.
On one hand, more compute means more heat, and more heat means more cooling demand. Estimates suggest that a single large AI model training run can consume hundreds of thousands of gallons of water across weeks of processing. Researchers have also estimated that a short conversation with a large language model consumes somewhere between a few milliliters and half a liter of water, counting both direct cooling and the water used to generate the electricity. Those per-query numbers stay small individually but add up fast across billions of daily requests.
On the other hand, high-density racks force operators toward liquid cooling, and liquid cooling is usually a closed loop. A direct-to-chip system moves heat far more efficiently than blowing cold air through a room, which cuts the total cooling energy needed. Many new AI facilities pair closed-loop liquid cooling with air-cooled or dry heat rejection, producing sites with dramatically lower water intensity than the evaporative buildings they replace.
The net result depends on scale. Even if each unit of compute uses less water, total compute is growing so quickly that absolute water use still climbs. Analysts project global data center water consumption will keep rising through the rest of the decade, with AI accounting for the majority of the growth. That is why efficiency alone will not settle the debate. Siting decisions, water sourcing, and local water availability matter just as much.
- Rack density jump: from 5 to 10 kW per rack to 40 to 130 kW for AI clusters
- Cooling shift: air cooling struggles above roughly 30 kW per rack, pushing liquid adoption
- Training runs: weeks-long jobs that run at near-full power around the clock with no off-peak dip
- Inference load: steady, growing, and spread across many regions including water-stressed ones
- Heat reuse potential: liquid cooling produces warmer waste water that district heating systems can actually use
Real Companies, Real Numbers, and Real Communities
Abstract ranges only go so far. Public disclosures and local reporting give us concrete anchors for how much water major operators actually use.
Google has reported consuming billions of gallons of water annually across its global data center fleet, with individual campuses varying enormously. Its facility in The Dalles, Oregon, drew public attention when records showed it used hundreds of millions of gallons per year, a large share of the city’s total supply. Meanwhile, Google’s site in Douglas County, Georgia, runs largely on reclaimed wastewater, and several European sites use air cooling with minimal water draw.
Microsoft has published annual water figures in the billions of liters and set a goal to become water positive, meaning it replenishes more water than it consumes. The company has also piloted a closed-loop design that consumes essentially zero water for cooling. Meta reports its data center water use and has funded watershed restoration projects in the regions where it operates. Amazon Web Services has committed to water positive operations as well and uses reclaimed water at many sites.
What Local Communities Actually Experience
Numbers land differently depending on where you live. In Loudoun County, Virginia, home to the world’s densest data center cluster, the combined draw runs into millions of gallons daily, but the region has substantial water infrastructure. In drought-prone parts of Arizona, Utah, and Chile, the same volumes trigger genuine conflict with agriculture and residents.
Consider a hypothetical but realistic case. A county with 60,000 residents uses about 6 million gallons of water per day for homes and businesses. A proposed 100 MW data center campus would add up to 4 million gallons per day at peak. That single facility would increase county water demand by roughly two thirds, all for a building that employs maybe 100 people. You can see why councils ask hard questions, and why operators increasingly arrive with air-cooled designs and reclaimed water contracts already in hand.
The best outcomes usually come from early transparency. Communities that require water use disclosure in permits, cap potable water draw, and negotiate reclaimed water infrastructure tend to get facilities that use a small fraction of what older designs would have consumed.
Common Misconceptions About Data Center Water Use
This topic attracts a lot of half-truths from both critics and industry defenders. Clearing them up helps you read the news more accurately.
Myth: All Data Centers Use Massive Amounts of Water
Not true. Plenty of facilities, especially in cool climates and newer builds, use almost no water for cooling. Colocation sites in Nordic countries and many air-cooled enterprise rooms consume little more than an office building. The huge numbers come from evaporatively cooled hyperscale campuses in warm regions.
Myth: The Water Is Gone Forever
Evaporated water re-enters the atmosphere and eventually falls as precipitation. What matters is local hydrology, not global water totals. Water evaporated in a desert basin may fall hundreds of miles away, which removes it from that specific watershed. So the concern is real, but the framing of water being destroyed is inaccurate.
Myth: On-Site Water Is the Whole Story
Indirect water use at power plants often exceeds on-site cooling water. Thermoelectric power generation consumes water for its own cooling. A data center running on coal or nuclear power carries a large hidden water footprint, while one running on solar or wind carries almost none. Any honest accounting includes both.
Myth: Water Use Is Steady Year-Round
Daily use swings wildly with temperature and humidity. Peak summer days can consume two to four times the winter daily average. Utilities care about peak day demand, not annual averages, because that is what strains the system.
- Annual averages hide summer peaks that stress local supply
- Withdrawal and consumption are different numbers and often get mixed up
- Reclaimed water use changes the community impact even at identical volumes
- Cutting water almost always increases electricity use somewhere
- Company-wide totals blur huge differences between individual sites
How Data Center Water Use Compares to Other Industries
Context does not excuse high water use, but it does help you judge scale. Data centers consume a real but modest share of total water withdrawals in most countries, while agriculture and power generation dominate.
In the United States, agriculture accounts for roughly 40 percent of freshwater withdrawals and thermoelectric power for another large share. Data centers, by most estimates, account for well under 1 percent of national water consumption. The problem is concentration. A million gallons per day spread across a state is invisible. A million gallons per day in one small desert county is a crisis.
| Water User | Approximate Daily Use | Notes |
|---|---|---|
| Large hyperscale data center campus | 1 to 5 million gallons | Concentrated in one location, peaks in summer |
| 18-hole golf course | 100,000 to 1 million gallons | Highly climate dependent |
| Coal-fired power plant (1,000 MW) | 10 to 30 million gallons | Mostly withdrawal, large consumption share |
| Semiconductor fabrication plant | 2 to 10 million gallons | Requires ultrapure water |
| 1,000 acres of irrigated alfalfa | 3 to 5 million gallons | Seasonal, fully consumptive |
| Beverage bottling plant | 500,000 to 2 million gallons | Includes product and cleaning |
| City of 50,000 people | 4 to 6 million gallons | Residential and commercial combined |
Notice how a single hyperscale campus sits in the same league as a small city or a large farm operation. That comparison explains both the criticism and the industry’s defense. Critics point out that the facility arrives suddenly and employs few people. Defenders note that the digital services it supports replace physically larger, more resource-intensive alternatives like data on paper, physical retail, or business travel.
The fairest way to think about it: data center water use is not the largest problem in water management, but it is one of the fastest growing, and it lands in specific places that often cannot absorb it.
Proven Strategies to Cut Daily Water Consumption
The good news is that operators have a deep toolkit, and many of these approaches pay for themselves. Here are the strategies that actually move the needle on daily water use.
Design and Technology Changes
- Raise the setpoint. Running server inlet air at 78 to 80 degrees instead of 68 dramatically extends the hours when free cooling works, cutting both water and power.
- Install hybrid or adiabatic coolers. Run dry most of the year and wet only on peak days to cut annual water by 60 to 90 percent.
- Adopt direct-to-chip or immersion liquid cooling. Closed loops handle AI density and consume almost no water in normal operation.
- Increase cycles of concentration. Better water treatment lets cooling towers reuse water more times before blowdown, cutting total draw by 20 to 40 percent.
- Capture and reuse condensate. Air handling units produce condensate that can feed back into the cooling loop instead of going to drain.
Sourcing and Siting Choices
- Contract for reclaimed or recycled municipal wastewater instead of potable supply
- Harvest rainwater and stormwater on large roof and site areas
- Use non-potable industrial or mine water where available
- Site new capacity in cool climates or water-abundant regions when latency allows
- Match renewable power procurement to reduce the indirect water footprint from generation
- Return warm waste heat to district heating networks so the energy does double duty
Measurement and Accountability
You cannot manage what you do not meter. Leading operators install submeters on every cooling loop, track WUE monthly, and publish site-level figures. Some sign water stewardship agreements with local utilities that cap draw during drought and fund watershed restoration to offset consumption.
A practical example shows the combined effect. Take that 40 MW Phoenix facility from earlier, using 1.6 million gallons on a peak summer day. Switch it to hybrid coolers, raise setpoints to 80 degrees, source reclaimed water, and add closed-loop liquid cooling for the densest racks. Peak day use drops to roughly 250,000 gallons, and nearly all of it comes from treated wastewater instead of the drinking supply. The community impact changes from alarming to manageable, and the operator’s energy penalty stays under 10 percent.
Questions People Ask Most About Data Center Water Use
These questions come up constantly in community meetings, comment sections, and search results. Here are straight answers.
Does a data center use water 24 hours a day?
Yes, if it uses evaporative cooling, though the rate changes hourly with outdoor temperature and server load. Water flow typically peaks in the afternoon and drops overnight. Air-cooled and closed-loop facilities use water only for occasional top-offs and building services.
How much water does one server rack use per day?
A 10 kilowatt rack in an evaporatively cooled facility corresponds to roughly 400 to 500 gallons per day. A 100 kilowatt AI rack on a closed liquid loop may correspond to under 50 gallons per day of makeup water, though its indirect power-plant water footprint could be larger.
Do data centers use drinking water?
Many still do, especially older facilities connected to municipal supply. The trend is moving fast toward reclaimed and non-potable sources, and several jurisdictions now restrict potable water for cooling in new permits.
Can data centers recycle their cooling water?
Partly. Closed loops recycle continuously. Open towers recycle water several times before mineral buildup forces a discharge. On-site treatment can push that further, and some campuses now treat and reuse their own blowdown.
Is data center water use growing or shrinking?
Water per unit of compute is falling steadily thanks to better design. Total water use is still growing because compute demand is growing faster. Expect that pattern to continue until liquid and dry cooling become the default across the whole installed base.
What can I do as a resident or customer?
Ask local officials to require water disclosure and potable water caps in data center permits. As a business buyer, ask cloud providers for site-level WUE and water source data. Demand from customers has already pushed several operators toward zero-water designs.
What the Next Decade Looks Like for Data Center Water
Three forces will shape the answer to this question going forward: AI demand, regulation, and cooling technology. Each one is moving quickly.
AI keeps pushing power density upward, which makes air cooling impractical for the newest hardware. That accelerates the shift to liquid cooling, and liquid cooling systems mostly run closed. Ironically, the workload that triggered the loudest water concerns may end up driving the industry toward its least thirsty designs. But total compute growth will likely outrun those efficiency gains for several more years.
Regulation is catching up too. More jurisdictions now require water use reporting as a condition of permits, restrict potable water for cooling, or tie approvals to drought contingency plans. The European Union has moved toward mandatory reporting of data center energy and water metrics. Expect similar rules to spread across US states with tight water budgets.
Technology keeps surprising people. Two-phase immersion cooling, waste heat recovery for district heating, seawater cooling in coastal locations, underground and underwater deployments, and advanced water treatment that pushes cooling towers to 15 or more cycles of concentration all reduce daily draw. Meanwhile, better AI-driven controls optimize cooling in real time, saving both water and power without any hardware change.
- Zero-water designs becoming standard for new builds in water-stressed regions
- Mandatory disclosure of site-level water use spreading through regulation
- Liquid cooling adoption jumping sharply as AI rack density climbs
- Heat reuse partnerships turning waste heat into a community asset
- Water positive pledges from major operators funding watershed restoration
- Smarter siting that weighs local water stress alongside power and latency
So where does that leave us? The daily water use of a data center ranges from essentially nothing to more than five million gallons, and the deciding factors are cooling design, climate, size, and water source. A mid-sized facility typically lands between 100,000 and 300,000 gallons per day, while hyperscale campuses in warm climates reach into the millions. Metrics like WUE let you compare facilities fairly, and simple math converts those metrics into daily gallons you can actually picture. Just remember that summer peaks run far above annual averages, and that indirect water use at power plants often exceeds what happens on site.
This topic deserves your attention because the digital services we use every day carry a physical footprint in real watersheds. The encouraging part is that the solutions already exist and work. Hybrid coolers, closed-loop liquid cooling, reclaimed water contracts, higher setpoints, and smart siting can cut a facility’s daily draw by 80 percent or more without hurting performance. As AI drives the industry toward liquid cooling and communities demand real transparency, the water math is genuinely improving. Keep asking for site-level numbers, keep an eye on how new projects source their water, and you will be able to judge any data center proposal with far more confidence than the headlines alone allow.
