The Tech-MadeWater Crisis
What do we mean by AI consuming water? Over the last few months the discussions surrounding AI and water shortages
globally have become so intertwined that it has become near impossible to
mention one without the other. But, what do we really mean when we say AI is
consuming our already limited fresh water supplies, why not just use ocean
water, and why can't we find alternatives?
Let's start by answering the first question, why do AI data centers need so much
water in the first place?
Well, to start, AI essentially is a hungry adaptive system that constantly needs
more data and information to continue its rapid development. Which means every
activity on the internet passes through a vast network of data centers that
collectively house billions of gigabytes of information. [1] Which in turn, consumes
enormous amounts of energy in order to store and process all this information.
Consequently, it generates a lot of heat.
In fact, Data centres built to power AI produce so much heat that they can raise
the surface temperature of the land around them by several degrees, creating
so-called data centre heat islands that may already be affecting up to 340 million
people.[2]
A typical data center uses 300,000 gallons of water each day (equivalent to the
demands of about 1,000 households), but large data centers can use an estimated
5 million gallons of water each day, equivalent to the needs of a town of up to
50,000 residents. [3]
Why fresh water?
Now, after reading that I'm sure some people may question why we are using the
limited fresh water that we need instead of our unlimited supply of ocean water. A
reasonable question; however, when saltwater evaporates, the salt and other
minerals are left behind during the evaporation process. In a data center cooling
system, this residue would quickly build up and clog the membrane, preventing
further evaporation. Effectively destroying the machinery. [4]
So..., any alternatives?
There are quite a few alternatives actually, already being tested. [5]
1. Liquid Cooling Systems:
Liquid cooling systems take the original idea of using water as a cooling agent
and twists it to make it a sustainable system. Instead of evaporating the water the
system uses Direct-to-chip and rear-door heat exchangers to circulate liquid.
Which basically means running cool water directly on the surface of the chip in
sealed loops, thereby eliminating evaporative losses. Studies show that these
systems are actually more efficient, and can reduce cooling energy consumption
by 40–60% compared to conventional air cooling.
2. Immersion Cooling:
Still following the theme of water, except this time we’re going underwater. The
concept behind this system is fully submerging the AI servers in dielectric fluids
(i.e. fluids that don't conduct electricity) that have low boiling points. Essentially
the water just absorbs and soaks up all the heat from the data servers.
3. Free Air Cooling:
Finally, moving on from water, we go to open air. As the name suggests, this
method utilizes cool outdoor cold air for heat removal. Furthermore, facilities
employing free cooling report energy reductions up to 60%. However, this system’s
effectiveness is very climate-dependent, and depends on many variables as it
is influenced by air quality, humidity, and seasonal variation.
4. Geothermal and Alternative Water Sources:
For the most unique system out of these we find ourselves underground instead
of underwater this time. By tapping into naturally cool temperatures deep
underground, geothermal systems can exploit stable underground temperatures
or groundwater aquifers through closed-loop configurations.
The future of AI data servers
Currently, the tech industry is failing to meet the sustainable measures that we
must urgently implement to ensure that our fresh water supply doesn't run dry
and to prevent the rapid development of an already global water crisis, that isn't
only affecting millions of people but is also causing irreversible damage to our
planet and its climate.
However, despite all this, it's important to remember that we are moving in the
right direction and that alternatives already exist. If there’s one thing for sure, it’s
the fact that sustainability will surely pave the way for a brighter future for us and
the tech industry as a whole.
Bibliography:
1. CoreSite Blog Lettiere, Christopher. “Hot or Not? How Data Center Thermal
Standards Impact Energy Use and Costs.” CoreSite, 2024,
act-energy-use-and-costs. Accessed 31 Mar. 2026.
2. New Scientist Sparkes, Matthew. “AI Data Centres Can Warm Surrounding
Areas by up to 9.1°C.” New Scientist, 31 Oct. 2024,
ng-areas-by-up-to-9-1c/. Accessed 31 Mar. 2026.
3. Brookings Institution Kane, Joseph W. “AI, Data Centers, and Water.”
Brookings, 20 Nov. 2025,
www.brookings.edu/articles/ai-data-centers-and-water/. Accessed 31 Mar.
2026.
4. Data Center Knowledge Tozzi, Christopher. “Turning Off the Tap: The Case
for Seawater Cooling in Data Centers.” Data Center Knowledge, 16 Mar. 2026,
awater-cooling-in-data-centers. Accessed 31 Mar. 2026.
5. Authorea (PDF/Journal Article) “Sustainable Alternatives to Water Cooling
in Data Centers.” Authorea, 2024,
able%20Alternatives%20to%20Water%20Cooling%20in%20Data%20Centers/S
ustainable%20Alternatives%20to%20Water%20Cooling%20in%20Data%20Ce
nters.pdf. Accessed 31 Mar. 2026.
Writer: Leya akkoush
Editor: Lynn Almarzoog
Comments