A typical hyperscale data center uses 100,000 to 5 million gallons per day, depending on size, climate, and cooling technology. The number that matters for your community is the disclosed local consumption against your local utility capacity.
Typical 100MW evaporatively cooled facility in a hot climate
Estimated total US data center water consumption, rising fast
Standard metric, but reporting excludes water embedded in upstream electricity generation
Servers turn nearly all of their electricity into heat. A single rack can dissipate 15 to 80 kilowatts of heat continuously, year-round. Removing that heat at scale requires either water, electricity, or both.
The cheapest method is evaporative cooling: outside air passes over wet media or through a cooling tower, the water evaporates, and the phase change carries heat away. This consumes water but uses relatively little electricity. The opposite end is refrigerant-based cooling: zero water, much higher electricity bill. Most large facilities are somewhere in between.
Three categories matter for community impact assessment:
The disclosure question for any new project: which category does this facility use, and how does that draw compare to existing residential and agricultural demand on the same source?
Three structural factors determine the answer:
The single most useful question to ask at a public hearing: who pays for any new water infrastructure required to serve this facility, and how is that cost allocated across rate classes?
Cooling technology choice is the biggest variable. Compared to a baseline evaporative system, direct-to-chip liquid cooling can reduce water consumption by 70 to 90 percent. Closed-loop and air-cooled systems can approach zero consumption. The trade-off is electricity: less water means more kilowatt-hours.
Climate is the second variable. The same workload in Council Bluffs, Iowa uses far less water than in Phoenix or Las Vegas. Operators choose hot climates anyway because the land and power are cheaper.
Workload type is the third. AI training and inference at high server density push power per rack to 30 to 80 kilowatts, which pushes per-site daily water totals up substantially compared to traditional cloud workloads.
Independent analysis. Same numbers for every reader.
LSARS uses published EPA, OEHHA, and CARB methodologies. The applicant pays for the analysis. The report does not change for the buyer. Communities, councils, and developers see the same scoreboard.
Read the full methodologyQuestions we hear
Sourced from public regulatory filings, academic studies, and operator environmental reports.
A typical hyperscale data center uses 100,000 to 5 million gallons of water per day, depending on size, climate, and cooling technology. A 100-megawatt facility in a hot climate using evaporative cooling can consume 1 to 1.5 million gallons per day. AI-training campuses at the high end can approach 5 million gallons per day. Closed-loop and direct-to-chip cooling systems use significantly less.
For a single large-campus facility, daily consumption commonly ranges from 300,000 gallons to 1.5 million gallons. The variation comes from outdoor air temperature, server density, and the cooling type. The same workload uses far less water in a cool climate than a hot one. Daily peaks during heat events can be two to three times the annual average.
AI training facilities tend to draw more power per rack and need more cooling per square foot. Estimates range from 0.5 to 4 gallons of water per kilowatt-hour of IT load, depending on cooling design. A 200-megawatt AI training cluster running at 80 percent utilization can use 1 to 4 million gallons per day. The total grows with every model generation.
Most large data centers use water for evaporative cooling: outside air passes over wet media or through a cooling tower, and the water evaporates to carry heat away from the building. A smaller share of water goes to humidification, fire suppression, and human use. Air-cooled, refrigerant-based, and direct-to-chip liquid cooling systems use far less water but more electricity.
In its 2024 environmental report, Google disclosed roughly 6.4 billion gallons of water withdrawn across its global data center fleet, with the per-facility average around 450,000 gallons per day. Individual sites vary widely. A facility in Council Bluffs, Iowa can use under 100,000 gallons per day, while one in The Dalles, Oregon or in Las Vegas can use much more. Public disclosures are partial and lag the calendar year.
Servers turn nearly all of their electricity into heat. Removing that heat at scale, year-round, is what consumes water. The cheapest way to dump heat to the atmosphere is evaporative cooling, where water phase-changes from liquid to gas and carries heat with it. Replacing evaporative cooling with air-only or refrigerant-only systems reduces water use but raises the electricity bill.
Sources fall into three categories: potable (treated drinking water), reclaimed or recycled (treated effluent that does not meet drinking standards), and non-potable (industrial, raw river or aquifer water). Operators prefer potable for ease but face the most community backlash for it. Reclaimed and non-potable sources reduce competition with residential supply but require additional treatment infrastructure on site.
It depends on three factors: whether the utility is cost-of-service or revenue-neutral on large industrial customers, whether new water infrastructure is required to serve the project, and how the project pays for that infrastructure. If costs are spread across the rate base, residential bills can rise. If the developer funds the new infrastructure directly, residential bills may not change at all. The disclosure question to ask: who pays for any new water infrastructure required to serve this project?
Annual totals range from tens of millions to hundreds of millions of gallons per facility. A 100-megawatt evaporatively cooled facility in a hot climate can consume 350 to 500 million gallons annually. A 100-megawatt air-cooled facility in a cold climate may use under 50 million gallons. Total US data center water consumption is estimated at 175 billion gallons per year and rising.
Hyperscale facilities, defined loosely as 100 megawatts or more of IT load, commonly use 1 to 5 million gallons per day at full operation. Campus-scale developments combining multiple buildings can multiply that figure. Microsoft, Google, Meta, AWS, and Oracle all operate hyperscale sites in this range; total disclosed water consumption exceeds 50 billion gallons annually across the largest operators.
Yes, in limited deployments. Microsoft has tested underwater data centers cooled by sea water. Surface-located coastal facilities can use sea water in heat exchangers without consuming it (closed loop). The challenges are biofouling, corrosion, and discharge permitting. Sea water cooling is the exception rather than the rule, and only a handful of operating facilities globally use it.
A closed-loop system circulates the same water continuously through cooling coils and a heat exchanger, with the heat ultimately dumped to outside air or refrigerant. There is no evaporation in the primary loop, so water consumption is near zero after initial fill and occasional makeup. Trade-off: closed-loop systems use more electricity to drive larger fans and chillers. They are most common in cold or temperate climates where outside air can do most of the cooling work.
AI training pushes server power density higher, which pushes cooling load higher, which pushes water use higher per square foot. A traditional cloud rack at 6 to 10 kilowatts is moving toward 30 to 80 kilowatts for AI training. Per kilowatt-hour, water use is similar between AI and traditional workloads under the same cooling design, but AI campuses run more kilowatt-hours through the same building footprint, so the per-site daily total is higher.
Cooling accounts for roughly 90 to 97 percent of a facility water budget. The rest covers humidification (server racks need 40 to 60 percent relative humidity), fire suppression test cycles, and sanitary use for staff. Some facilities also use water for groundwater recharge or stormwater management as required by local permits, which counts against permitted withdrawal even though it is not consumed.
The standard metric is Water Usage Effectiveness (WUE), measured as liters of water per kilowatt-hour of IT load. Hyperscale operators publish WUE annually but disclosure is partial: it usually excludes the water embedded in electricity generation upstream. Total consumption is typically reported in millions of gallons or megaliters per year, broken out by source. AB2588-style hot-spots reporting also captures water in the air toxics inventory in some jurisdictions.
Send this to someone who can act
Pre-written templates you can review, edit, and send. No signup, no tracking. Pick who you want to reach.
Email the developer or applicant
Hi, I am a community member with concerns about this proposed data center. The questions on my mind are specific: 1. How much water will the facility use per day, and where does that water come from? 2. Will the project compete with residential or agricultural demand on the same supply? 3. Who pays for any new water infrastructure required to serve the facility? I came across LSARS, an independent health and environmental data platform. The applicant pays for the analysis but the methodology is published (EPA AirToxScreen, California OEHHA), so the developer cannot change the numbers. The same report goes to the council, the community, and the applicant. Would you consider funding an independent LSARS analysis for this project? It would address the questions above transparently, before they become a public hearing problem. Reference page I read: https://www.lsars.com/data-center-water-use Developer overview: https://www.lsars.com/for/developers Thank you for considering it. [Your name]
The message opens in your default email app where you can edit before sending. Replace [Your name] and any bracketed placeholders. The reference link points back to this page so the recipient sees exactly what you read.
We publish project-specific water consumption, sourcing, and utility-impact data on public dashboards as projects move through permitting.
See your community page