How much water does a hyperscale data center actually use?
A hyperscale data center is a single building, or a cluster of buildings, that draws around 100 megawatts of electrical power or more. Industry analysts use the term for campuses built by the largest cloud and AI operators — Google, Microsoft, Meta, Amazon, Oracle, and specialist firms like QTS and EdgeConneX.[1]
Water enters the picture because servers produce heat, and heat has to leave the building before the hardware fails. The cheapest way to move heat out is evaporation. A cooling tower sprays water over warm air; some of that water evaporates into the sky, the rest is recirculated through the chilled-water loop.
Peer-reviewed and government figures put the typical range at 1 to 5 million gallons of water per day (MGD) for a hyperscale campus using evaporative cooling. The most commonly cited federal number comes from the Lawrence Berkeley National Laboratory's 2024 data-center energy report, which modeled on-site water use across U.S. operators.[2] Operator disclosures in 10-K filings and sustainability reports tend to confirm the range, though they are usually expressed as annual totals rather than peak-day figures.
Putting the number in scale
The raw figure — one to five million gallons a day — is hard to feel. A few rulers can help.
- Olympic pools. A regulation Olympic swimming pool holds about 660,000 gallons. A 3 MGD facility evaporates about four and a half pools every 24 hours.
- Households. The U.S. Environmental Protection Agency's WaterSense program puts average U.S. residential indoor use at roughly 300 gallons per household per day.[3] A 3 MGD data center uses as much water as 10,000 households.
- Small towns. A Texas town of 15,000 people uses roughly 3 million gallons per day for all residential purposes combined.
Daily water use, compared
Evaporative versus closed-loop
Not every data center uses water at that rate. There are three main cooling architectures, and they produce very different water bills.
- Evaporative cooling (open-loop cooling tower). Cheapest to run in hot climates. Heavy water use, lower electricity use. Most common in Texas.
- Closed-loop (dry coolers or air-cooled chillers). Near-zero water consumption. Higher electricity use because fans and chillers do the heat-rejection work that evaporation otherwise does for free.
- Direct-to-chip liquid cooling. A newer approach where coolant is piped directly to the server chips. Can be designed as closed-loop at the building level but still may need an evaporative tower for ultimate heat rejection, depending on design.
The tradeoff is real. A closed-loop facility will use perhaps 10 percent more power than an evaporative facility of the same compute capacity — meaning higher costs for the operator and more demand on the electrical grid. Which cooling method an operator selects is usually an economic decision, shaped by local water prices, power prices, climate, and permit conditions.
What operators actually disclose
Google's 2024 Environmental Report puts its fleet-wide water-use effectiveness (WUE) at 1.15 liters per kilowatt-hour of IT energy delivered.[4] Meta's 2023 sustainability report reports on-site water use of roughly 2.5 billion gallons across its global data-center fleet.[5] These are reported in annual aggregates and make it possible to back out rough per-facility numbers, though the company averages smooth over site-by-site differences that matter a great deal to a single county.
The cleanest place to find a site-specific peak-day number is the water-permit application filed with the local utility, the Groundwater Conservation District (if one exists), or the Texas Commission on Environmental Quality. Operators are required to disclose projected peak-day draw as part of the permitting record. Those filings are public.
Why the cooling question matters in Texas
Eight of the top twenty Texas counties most targeted by new data-center proposals draw from aquifers that the Texas Water Development Board's 2022 State Water Plan projects will be in structural shortage by 2040.[6] That does not mean every data center will cause a water crisis. It does mean the difference between an evaporative and a closed-loop facility — between 3 MGD and near-zero MGD — is a meaningful number to the county the facility sits in.
The right question is not whether water use is "big." It is: which method, and where does the water come from?
What to ask before a vote
If a data-center proposal is on your county commission, city council, or groundwater-district agenda, a resident can reasonably ask for three numbers in the public record:
- Projected peak-day water consumption in MGD, not annual average.
- The cooling architecture — evaporative, closed-loop, direct-to-chip, or hybrid.
- The water source — municipal, surface water, aquifer, or reclaimed.
These are all questions the operator can answer. Any one of them missing from a proposal is itself public-record information worth noting.
Subscribe for alerts
One email when a data-center, rezoning, or water-permit item lands on an agenda in your Texas county. Free. No spam.
Sources
- Uptime Institute, Global Data Center Survey 2024.
- Shehabi et al., 2024 United States Data Center Energy Usage Report, Lawrence Berkeley National Laboratory, December 2024.
- U.S. Environmental Protection Agency, WaterSense Residential End Uses of Water, Version 2, 2016.
- Alphabet / Google, 2024 Environmental Report.
- Meta Platforms, 2023 Sustainability Report.
- Texas Water Development Board, 2022 State Water Plan, county-level needs projections.