Data centre water use in Texas currently sits at around 0.4% of the state’s total consumption. According to research from the Houston Advanced Research Center (HARC), that figure could climb to as much as 2.7% by 2030, and even that estimate, based on 2025 data, is likely already out of date given the scale of new development moving into the state.

Dr Margaret Cook, water resources programme manager at HARC, says the 2.7% projection assumes operators are largely choosing natural gas combined-cycle power and average water-intensive cooling technologies. “We could actually see that high-end estimate or even higher in the coming years, because of the amount of planned growth that’s happening,” she says.

But the statewide number, Cook argues, is the wrong place to look for risk. “The impacts are really local,” she says. Texas absorbed a comparable jump in water use during the hydraulic fracturing boom around 15 years ago, largely by requiring reporting and pushing operators toward non-freshwater sources, a template Cook believes could work again.

The difference this time is geography: hyperscale data centre growth is landing not in Austin itself, but in the smaller counties surrounding it: Williamson, Caldwell, and Hays counties chief among them, where developments are landing largely outside incorporated cities.

“Where we’re seeing a lot of hyperscale growth is generally outside of municipalities… just outside of city limits, so that they’re in the county, but not the city, so that there’s less regulation,” Cook says. Many of those communities have no long-term water plan in place, and some are still working through ageing water infrastructure even before new industrial-scale demand arrives.

A regulatory gap, not just a resource gap

Cook is not surprised by the scale of community pushback data centres have attracted in Texas. She points to a pattern that mirrors the state’s earlier experience with oil and gas: friction appears when heavy industrial activity moves into areas unaccustomed to it. What’s different with data centres, she says, is the mismatch between the scale of what’s being built and the regulatory apparatus around it. Projects with the footprint of major industrial developments are being approved with “the mentality of a commercial building”, without the community advisory boards or engagement infrastructure that typically accompany industrial-scale projects in other sectors.

The clearest illustration Cook offers: a county judge recently testified to the Texas House Natural Resources Committee that he was unaware a hyperscale data centre was being built in his own county until roads needed closing for construction. “If the local government doesn’t even have a distinct avenue to engage in that economic development process, then the community at large definitely doesn’t have an opportunity to do that,” she says.

Disclosure remains largely voluntary. Cook argues that meaningful transparency would mean operators publishing average, annual and peak water and energy volumes, alongside their efficiency measures and technology choices, and that state-vetted reporting would carry more weight with communities than self-reported figures from the companies themselves.

What residents are actually saying

That gap plays out on the ground, according to Kamil Cook, climate and clean energy associate in Public Citizen’s Texas office. He works directly with Texas communities pushing back against data centre development. “I haven’t seen as broad a pushback to really anything else,” he says. “I’ve been in this space for about three years, and I haven’t seen anything like this before.”

Water is nearly always the first concern, he says, given how much of Texas is under drought or water scarcity. Close behind is a sense of powerlessness: local elected officials have little real ability to impose rules on data centres, and their main point of leverage is tax abatements. That frustrates residents further, he says, when a company drawing millions of gallons a year is also being offered a tax break in exchange for limits on its water use. Air quality, from backup diesel generators and behind-the-meter gas plants, and the effect on electricity rates from rising grid demand round out the list.

What frustrates communities most, Kamil Cook says, is how little they know before a project is approved. Large, high-profile developments tend to be visible early. But many smaller projects begin life under NDAs, often referred to only by a code name, “Project Red” or “Project Yellow”, with developers citing proprietary technology to justify the secrecy.

Residents frequently discover, only after the fact, that a routine-looking council item a year earlier quietly granted land or water rights to what turns out to be a data centre developer. He says he hasn’t yet seen an example of a company engaging well with an affected community, only ones that have been standoffish, a tension he puts down to developers wanting to get online as fast as possible, against communities who simply want more time.

Where the 20% trade-off undoes the fix

Cook’s research identifies four mitigation levers: dry cooling, sourcing water-lean energy, water reuse and offset schemes, and public-private investment in local infrastructure. She cites Amazon as an operator already using all four, pairing dry cooling with renewable energy sourcing, capturing rainwater on site, and investing in community water-replenishment initiatives. Google and Meta have adopted some of the same measures; smaller operators, she notes, often lack the resources to do so.

Of the four, Cook expects dry cooling and water reuse to scale fastest, but with a caveat that complicates the “water-positive” narratives some operators use. Switching to dry cooling can increase energy use by as much as 20%. If that extra power comes from a water-cooled plant, total water consumption can rise rather than fall.

“We only recommend that switch if it also is in tandem with a water-lean energy source,” Cook says, “so that you’re not accidentally just moving the water use upstream to your power plant.” Her recommendation is a combined water-use-effectiveness measure covering both the data centre and its power source, something no current Texas reporting requirement captures.

Asked whether any part of Texas is close to a genuine tipping point, Cook names only one: Corpus Christi and the surrounding Coastal Bend region, where decades of drought and a failed desalination scheme have already produced a supply crisis. Her reading is a warning for the wider industry: public trust and water-supply planning have to be built years ahead of a crunch, not during one. “It’s going to take a long time for that to actually get built,” she says of the delayed desalination plan, “which tells you that they were very, very delayed in making the decisions that they needed to make.”

The industry’s answer: Cooling as the fix

On the technology side, the shift Cook describes is already under way. Reza Azizian, co-founder and CEO of cooling start-up Ferveret, has watched the industry move from air cooling, with power usage effectiveness (PUE) around 1.4, meaning roughly 40% energy overhead, toward direct-to-chip liquid cooling, which circulates a water-and-glycol mixture through cold plates on the GPU and CPU. That shift has brought typical PUE down to around 1.2, Azizian says, though fans are still needed to cool the rest of the server.

Ferveret, founded in 2021 by Azizian and MIT nuclear engineer Dr Matteo Bucci, is positioning its “adaptive phase cooling” technology, which combines elements of single-phase and two-phase cooling, as the next step. Azizian says the approach draws on decades of heat-transfer research from the nuclear industry, and can be added to existing direct-to-chip racks by changing the casing at server level, without wider infrastructure changes.

Ferveret is currently running roughly six-month pilots with CleanSpark, Furiosa AI and Switch, and working with ODM and OEM partners to be included in hyperscalers’ reference designs. As with any young vendor’s own account of pilot-stage technology, these figures are Ferveret’s projections rather than independently verified production data.

On community pushback, Azizian’s reading echoes Cook’s and Kamil Cook’s: he points to Virginia, where he says household electricity prices “almost doubled” as a result of data centre demand, and acknowledges water consumption and site noise as legitimate resident concerns. He is unequivocal, however, that the underlying build-out needs to continue: “We do need data centres,” he says, “because right now it’s like everyone [is] racing toward AI and AGI.” Asked how long before technology like his could shift the water-stress picture in a market like Texas, Azizian puts the timeline at roughly one to two years, an optimistic estimate that assumes hyperscaler adoption follows the reference-design conversations Ferveret says are already under way.

The gap that remains

Between Cook’s account of the regulatory lag, Kamil Cook’s account of communities caught off guard, and Azizian’s account of the technology response, one thing all three point to is timing. Cook’s central lesson from Corpus Christi is that trust and planning have to precede the crisis, not follow it.

Whether Central Texas’s current wave of hyperscale development gets there in time may depend less on any single cooling technology than on whether the reporting and engagement structures both Cooks describe as missing get built before, rather than after, the next round of pushback.

RELATED STORIES 

State of the Sector: Data Centres

New York data centre moratorium: Who will get off unscathed?