This is partner content, brought to you by Arbor Energy. Brad Hartwig is the company’s CEO and co-founder
This year, Corpus Christi, Texas was on track to become the first American city to run out of water. Recent floods bought it another year, but nobody there is calling it solved, not with industrial users alone responsible for more than half of regional demand. That’s the backdrop against which every new data center announcement now lands.
I wrote recently that the industry has treated community concerns as a communications problem when they should be treated as design requirements, worked out before a site is chosen rather than explained after. Water is where that argument gets tested. You can’t talk your way out of how many gallons a facility pulls from a stressed aquifer or a municipal system that’s already rationing. The number is the number, and it either works for the community or it doesn’t.
Traditional data center cooling relies on evaporative systems and chillers that draw enormous volumes of water, usually from the same treated supply residents drink from. A single hyperscale campus can use up to 5 million gallons a day (comparable to a town of 10,000 to 50,000 people) according to the Environmental and Energy Study Institute. In places like South Texas or the broader Southwest, that water is coming out of the same system residents are already being asked to cut back on.
Developers have options here, and some of the best ones are already running at scale.
Liquid cooling is the most direct lever. Direct-to-chip systems and full immersion cooling move heat away from servers without evaporating water at all. Immersion in particular can eliminate water use in cooling altogether, since it doesn’t rely on evaporation the way traditional systems do. These systems also handle the higher rack densities modern AI chips require, which is part of why developers are adopting them regardless of the water benefit. Several hyperscalers have deployed it at scale, and new builds should be designed around it from day one.
Closed-loop, zero-water designs eliminate fresh water withdrawal altogether. Rather than evaporating water to reject heat, these systems circulate a fixed volume of water or refrigerant between servers and air-cooled chillers, eliminating fresh water withdrawal for cooling entirely. Microsoft has committed to this approach across its new datacenter designs, with zero-water pilots underway in Arizona and Wisconsin, two states with very different but equally real water pressures.
Sourcing matters as much as volume. Facilities that still need water for cooling don’t have to pull from the drinking supply. Reclaimed municipal wastewater and other non-potable sources work just as well in a cooling tower. Some developers are starting to choose sites specifically near wastewater treatment plants or municipal effluent outfalls, so recycled water is available on-site instead of trucked in. Recirculating that water instead of discharging it after a single pass can add up to 75 percent in additional savings according to World Economic Forum research.
This is proven technology. What’s missing is developers planning for water alongside power from the start, rather than defaulting to whatever is cheapest to permit.
Arbor takes this a step further: Instead of minimizing water use, we generate it. Our modular turbines run on oxy-combustion, burning fuel with pure oxygen instead of ordinary air. Because there’s no nitrogen in the mix, the exhaust stream is mostly carbon dioxide and water vapor, which we condense and capture as usable water.
Each turbine generates 25 megawatts, and they’re built to stack, so a 100-megawatt installation could yield roughly 100 million gallons of water a year. Sited alongside a data center, those turbines could supply the water for the data center’s closed-loop cooling system directly, so the facility wouldn’t draw from the community’s supply at all.
Efficiency gains still leave a data center pulling more water out of a scarce system than it puts back. A facility that produces more than it needs changes the conversation entirely, from “how much less will you take” to “what do we do with the surplus.” In a place like Corpus Christi, that’s the kind of improvement that decides whether a community rejects a project or welcomes it.
Power remains the hardest technical problem in AI infrastructure, but water is quickly becoming the hardest political one. The developers who solve for it in the design, not in the press release, will be the ones still building five years from now.
Arbor is building zero-emission power systems designed to meet the requirements communities are setting. See how at arbor.co.
