This story was originally published by Grist. Sign up for Grist’s weekly newsletter here.

Data centres are notoriously thirsty. Researchers at the U.S. Lawrence Berkeley National Laboratory have found that, in 2023, the facilities consumed roughly 17 billion gallons (64.4 billion litres) of water for their operations in the U.S. alone.

But that’s only a small part of the picture, Grist reports. A much, much larger share of data centre water intensity is indirect, a byproduct of the facilities’ enormous appetites for energy. That’s because most power plants themselves require huge amounts of water to operate. This offsite, indirect water consumption amounted to a whopping 211 billion gallons (800 billion litres) in the Berkeley Lab’s 2023 tally—well over 10 times the direct onsite usage.

As Silicon Valley continues to pour hundreds of billions of dollars into artificial intelligence and demand for data centres grows, these water needs are only going to grow in tandem.

However, new research from Cornell University shows there’s a way to mitigate both the climate and water footprints of these facilities: Build them in places with lots of wind and solar energy.

“Location really matters,” said Fengqi You, an energy systems engineering professor at Cornell and co-author of the new study. Where companies choose to locate their data centres could alter their combined environmental footprints by a factor of up to 100.

In the course of their operations, data centres use water as a coolant. Energy-hungry servers generate substantial heat, and water circulates through cooling systems to prevent the equipment from overheating and breaking down.

But substantial amounts of water are also used indirectly through the generation of electricity to run the facilities. Thermoelectric power plants, regardless of whether they use coal, gas, or nuclear material, use that fuel to generate heat that converts water into steam, which is then used to spin a turbine and generate electricity. And since hydroelectric plants typically store large volumes of water in reservoirs behind dams, there are losses there as well, as water continually evaporates from the surface of reservoirs.

All told, water use during power generation can be responsible for more than 70% of a data centre’s total consumption, according to the new Cornell research. “That’s why the electricity power grid mix is very critical,” said You. 

You and his co-authors examined the energy and water use of data centres across the country to project where future investments should be made to reduce environmental impacts. The study makes the possibly precarious assumption that the data centre boom, which is being fuelled by staggering levels of investment in artificial intelligence, is unlikely to slow down anytime soon. Against that backdrop, the question the study then poses is: Where in the U.S. is the most environmentally sustainable place to build a data centre?

The researchers considered both the direct and indirect uses of energy and water as a result of building a data centre in a specific location. The most promising region they identified might turn heads: bone-dry West Texas. Because the region is sparsely populated, has groundwater that can be drawn on for use as a coolant, and produces ample wind energy, it scored highest on both energy and water stress metrics. The grid-related water footprint in West Texas is among the lowest in the country, thanks to the large amount of wind energy it produces, according to the study.

“From an energy and water efficiency perspective, the states that have enough dry renewables will be the best choice,” said You, adding that Montana, Nebraska, and South Dakota also appear to be prime locations for future AI servers.

Conversely, most parts of the Pacific Northwest didn’t score as well because of the region’s reliance on hydropower. Although the region’s electricity costs are low, the associated loss of water through power generation means building more data centres is likely to have a substantially larger water footprint than it would in other parts of the U.S.

Another recent study from researchers at Purdue University [pdf] came to a similar conclusion. They looked at the availability of water across the country and mapped out how that might change over time, particularly as climate change makes some regions hotter and drier. The paper also examined the water impact of existing Google data centres and found that the majority were located in areas with low water stress.

“Companies absolutely take the environment into consideration in their decisions—not just the economic factor,” said study co-author Yi Ding, an electrical and computer engineering professor at Purdue. “We infer that Google already somewhat considered water stress because they put most of the data centres in low-stress regions.”

Texas already has more than 400 data centres, second only to Virginia among U.S. states. Texas’ grid infrastructure, potential for renewables, and availability of cheap land has made it an attractive proposition for tech companies.

But the other states identified by the Cornell study as having a small environmental footprint—Nebraska, South Dakota, and Montana—have just 70 or so data centres combined, out of more than at least 4,200 nationally. That’s because a number of other factors, such as the policy environment and infrastructure considerations, are deterring companies from building new facilities there. You said that picture could change if those states geared their policy-making toward attracting data centres.