Key Takeaways
- Water availability is becoming as important as power availability when deciding where AI data centers can be built and kept running long-term.
- A large share of data centers built since 2022 sit in regions already facing water stress, including parts of Texas and Arizona.
- Conventional cooling forces a tradeoff between water-hungry evaporative systems and power-hungry, density-limited air cooling.
- Waterless architectures remove that tradeoff by using closed-loop immersion cooling instead of evaporative water use.
- Understanding how water scarcity affects permitting, regulation, and siting risk can help infrastructure planners avoid costly surprises down the road.
Every AI infrastructure conversation seems to circle back to power: how much of it a facility needs, where it comes from, and whether the grid can keep up. But now it is water use that is taking up the airtime. High-volume water usage for cooling is just as urgent a limit on where AI data centers can be built and how long they can keep running. But can new tech developments in how AI data centers are built solve the issue? The experts at Triton Thermal explain.
AI’s Water Bill Is Coming Due
Global data centers consumed roughly 560 billion liters of water in 2023, and that figure could rise to 1.2 trillion liters by 2030, enough to supply more than four million U.S. households. That trajectory matters because water, unlike electricity, cannot always be shipped in from somewhere else when local supply runs short. A region can add transmission lines or new generation capacity over time, but it cannot manufacture more groundwater once an aquifer is depleted.
A meaningful share of new and under-construction data centers already sit in regions facing rising water scarcity, which turns what used to be an environmental footnote into an operational risk. Planners who treat water as an afterthought are, in effect, betting their facility’s future on a resource that is getting harder to count on in exactly the places where AI infrastructure is expanding fastest. Water availability now belongs in the same planning conversation as power capacity and land costs.
Why Cooling Drinks So Much Water
Cooling, not computing itself, is where most of a data center’s water disappears. Understanding the mechanics behind that water use helps explain why the problem is so hard to shake with conventional infrastructure.
The Evaporative Cooling Tradeoff
Most data centers rely on evaporative cooling, where water is run through cooling towers and allowed to evaporate as it draws heat out of the building. This method works, but it is thirsty: a single hyperscale facility can consume three to five million gallons of water per day, with around 30 to 40% of that lost to evaporation rather than returned to its source. That means the water is not borrowed; it is consumed, and it has to be replaced continuously for as long as the facility runs.
One Facility, Millions of Gallons a Day
The scale involved can be hard to picture until it is compared to something familiar. A single hyperscale facility using evaporative cooling can draw three to five million gallons of water per day. One especially striking example: Loudoun County, Virginia’s data centers alone used over 1 billion gallons of water in 2023. Multiply that kind of demand across a fast-growing cluster of facilities, and it becomes clear why local water utilities and communities start paying close attention.
Building in the Driest Places
The uncomfortable irony is that a lot of new AI infrastructure keeps landing in places that can least afford heavy water use. That pattern deserves closer scrutiny before more capital gets committed to sites that may not be able to sustain it.
Data Centers Built Since 2022 Cluster in Water-Stressed Regions
A meaningful share of data centers built since 2022 have been located in water-stressed regions, including hot, dry climates like Arizona. This clustering is not random; it tends to follow cheap land, favorable tax incentives, and available power, with water availability often treated as a secondary concern until it becomes a problem during permitting or, worse, after construction is already underway.
Texas, Arizona, and the Aquifers Running Dry
Texas provides a vivid case study. The state’s rapid data center expansion has already raised legal and regulatory concerns about water usage, with reporting suggesting data centers could account for a significant share of Texas’s total water use by 2030 if current trends continue. Texas and Arizona are among the regions most exposed to water pressure tied to AI data center growth. For a company based in Houston, this is not an abstract policy debate; it is a regional reality shaping how infrastructure gets sited across the same part of the country where much of the AI buildout is happening.
The Choice Operators Keep Getting Wrong
Faced with these pressures, many operators still default to a binary decision that leaves them exposed no matter which side they pick.
Efficient and Thirsty, or Dry and Power-Hungry
Conventional cooling tends to force a choice between two imperfect options. Water-based and evaporative systems deliver strong energy efficiency, but they consume real, ongoing volumes of water that must be sourced, treated, and continuously replenished. Air-based systems go the other direction: they conserve water, but they demand more electricity and hit a density ceiling well before modern AI accelerator racks actually need them to. Most infrastructure decisions end up settling for one tradeoff or the other, efficient and thirsty, or dry and power-hungry, without escaping the underlying limitation. Neither option removes the risk; it just decides which resource the facility gambles on.
Designing Cooling That Skips the Tradeoff
A growing number of operators are asking whether the water-versus-power tradeoff is even necessary anymore. Industry leaders Triton Thermal say the answer now points toward architectures built to eliminate water dependency altogether rather than manage it more carefully.
Closed-Loop and Immersion Approaches Removing Water Entirely
Waterless cooling covers a growing set of technologies designed to remove the water dependency altogether. Closed-loop or “zero-water” air cooling rejects heat to ambient air through sensible heat transfer instead of evaporation, cutting out cooling towers and potable water draw. Direct-to-chip and immersion liquid cooling take a different route, circulating a dielectric fluid in direct contact with the hardware to move heat away without water in the loop at all. One prominent hyperscaler’s zero-water, closed-loop liquid cooling system for AI data centers is designed to save more than 33 million gallons per facility annually.
Atlas One, a fully waterless architecture, follows this same logic through single-phase immersion cooling. There are no cooling towers and no evaporative water consumption at any compute density, and the platform is designed to hit a targeted PUE of 1.0-1.1 without requiring the water tradeoff that evaporative systems demand. That efficiency comes from direct fluid contact with the hardware rather than from evaporating water to reject heat, which allows the architecture to sidestep the tradeoff instead of just managing it more efficiently.
Regulatory and Permitting Risk Water Creates
A low PUE achieved through evaporative cooling and a low PUE achieved through a waterless architecture can look identical on a spec sheet, but they expose an organization to very different risks. Facilities that depend on water face a growing list of operational threats that have nothing to do with raw compute performance:
- Local water availability shifting due to drought or competing municipal demand
- Drought-driven restrictions cutting into permitted withdrawal volumes
- Permitting friction in already water-stressed regions
- Rising regulatory and public scrutiny over industrial water use tied to AI expansion, with some states now considering legislation that would require closed-loop systems outright
None of that shows up in a PUE calculation, but all of it shows up in how easily a facility can be sited, approved, and kept running for the life of a deployment. Removing water from the cooling loop removes that entire risk category rather than just trimming a utility bill.
Water Availability Now Decides Where AI Gets Built
Power will keep dominating headlines, but water is quietly becoming the constraint that decides whether a proposed AI facility ever gets approved, or whether an existing one can keep expanding. Sites in water-stressed regions face longer permitting timelines, tougher public hearings, and the real possibility of restrictions kicking in mid-lease. Operators who build water risk into their siting criteria now, rather than after a project stalls in review, put themselves in a far stronger position as AI compute demand keeps climbing.
For infrastructure planners weighing where and how to build next, it helps to evaluate waterless AI data center cooling options early in the site-selection process, before water becomes the deciding factor rather than a line item.
Triton Thermal
3350 Yale St.
Houston
Texas
77018
United States
