You Can’t Manufacture an Aquifer

Data Centers & Energy By EnergyInnoCore Editor · Published Oct 18, 2026 · 10 min read

For the past couple of years, most conversations about AI infrastructure have eventually come back to electricity. Where will the next 500 megawatts come from? How quickly can a substation be built? Is there enough transmission? Can a data center sit behind the meter next to a gas plant, a nuclear plant or its own dedicated generation? Those questions are legitimate and they are not going away.

But another resource problem is developing alongside the power shortage, and in some parts of the country it may prove harder to solve: water. A data center can build its own generator, sign a power purchase agreement with a plant hundreds of miles away, or install batteries, fuel cells and gas turbines. Creating a new water supply is different. You cannot manufacture a local aquifer or build a river, and when a growing data center competes with households, farms and ecosystems for the same water, the conversation quickly stops being purely technical. It becomes political, and developers are learning that lesson in real time.

Why Water Is Now Part of Site Selection

Modern data centers generate a great deal of heat. Nearly all of the electricity a processor consumes ends up as heat, and that heat has to be removed. Operators use several cooling technologies, but evaporative cooling has long been popular because it moves heat efficiently while using less electricity than many purely mechanical systems.

The catch is in the name. Water that evaporates is consumed rather than returned to the local system. How much a facility uses varies enormously with climate, design, computing load, cooling technology and season. A facility in Oregon does not behave like one in Arizona, a closed-loop system behaves differently from an evaporative cooling tower, and new AI facilities may be configured very differently from older cloud data centers. That makes simple claims such as "a data center uses X gallons of water" misleading. Regional studies, such as HARC's analysis of data center water demand in Texas, show how much the answer depends on local assumptions.

What developers increasingly have to answer during site selection is a more practical question: can the local water system reliably serve this campus for the next twenty or thirty years? That question gets harder as facilities grow. An enterprise data center of 20 or 30 megawatts once looked large; hyperscale and AI campuses can now approach several hundred megawatts or even a gigawatt. More computing means more heat, and unless cooling technology changes, more heat can mean more water. That brings developers face to face with something the technology industry has rarely had to manage: local resource politics. Tucson, Arizona and The Dalles, Oregon illustrate two very different versions of the problem.

Tucson: A Project Rejected, Then New Rules

It would be hard to find a more sensitive place to introduce a large new industrial water user than southern Arizona. Water scarcity is part of everyday civic life in Tucson. The Colorado River is under long-term stress, groundwater protection has been debated for decades, and conservation is part of the region's economic survival strategy rather than a slogan.

Then came Project Blue, a proposed large-scale data center that became one of the city's most controversial development debates of 2025. Water was not the only concern (electricity use, noise, land use and transparency were also debated), but a major new water-consuming facility landed differently in Tucson than it might have elsewhere. On August 6, 2025, the mayor and council directed staff to end negotiations over the proposed development agreement and annexation, and called for faster work on rules for large water users and data centers, as AZPM's one-year review recounts.

Tucson did not simply say no and move on. The city developed a Large Quantity Water User approach and spent about a year drafting broader regulations. On August 5, 2026, the council adopted rules for large-scale data centers by a 6-1 vote. Among other things, developers must demonstrate access to adequate water and power, and large facilities must go through neighborhood coordination, a zoning hearing and final council consideration, according to AZPM's coverage of the vote.

From a developer's perspective, that is a major shift. Water is no longer a detail settled late in the project between an engineering contractor and the utility; it can become part of the political approval. A developer may have land under contract, a willing electricity provider and billions of dollars available, and still not have the community's agreement to supply water.

For decades, communities competed for data centers because they brought investment, construction jobs and tax revenue without the traffic and population growth of a large factory. AI-scale facilities are changing that calculation. Residents are asking what they receive in return for committing scarce infrastructure to a facility that may employ relatively few people after construction. In a water-stressed city the question is sharper still: if households are asked to conserve and to accept higher rates for new water infrastructure, why should a data center receive millions of gallons? There may be good answers, but developers now have to provide them.

The Dalles: When the Data Center Is Already There

If Tucson shows how hard it can be to approve a large new project, The Dalles shows a different challenge: what happens when an existing data center becomes one of the largest water customers in town.

Google opened its first data center in The Dalles in 2006, attracted by relatively inexpensive electricity, fiber connectivity and a favorable climate. Its presence expanded substantially over the following two decades, and so did its water use. According to city records reported by Oregon Public Broadcasting, Google's annual water use rose from about 104 million gallons in 2012 to about 434 million gallons in 2024, and OPB reports that the company now accounts for roughly a third of the city's water use.

That does not mean Google behaved improperly. The company has agreements with the city, and city officials have said repeatedly that their water-planning efforts are meant to serve population growth and long-term municipal needs, not simply Google's expansion. That distinction deserves respect. But when a single corporate customer uses that large a share of a city's water, public debate about water infrastructure inevitably becomes tied to that customer.

The Dalles is seeking to expand its reservoir, including through a proposed transfer of nearby Mount Hood National Forest land, as reported by the Oregon Capital Chronicle. City officials maintain that the expansion is part of long-term planning. Environmental groups, meanwhile, have raised concerns about withdrawals from the Dog River system, which serves as cold-water refuge for threatened fish.

Availability Is Not the Same as Acceptability

The Dalles illustrates a problem developers can face even when water technically exists. Engineers might conclude that a city can supply another million gallons per day, but the community may ask different questions. What happens during a drought? Who gets priority? Who pays for expanded treatment plants, reservoirs and pipelines? What happens to residential rates? And how much should the public know about the water use of an individual company?

That last question has already caused controversy. In 2021, The Oregonian requested records of Google's local water use. The city initially withheld them as a trade secret and went to court to block their release before dropping the fight in late 2022. The dispute turned routine utility data into a public question: is industrial water use private business information when the resource belongs to the community? For the next generation of AI campuses, developers should assume the answer will increasingly be no.

Developers May Need to Bring Their Own Water Strategy

Traditionally, developers screened sites for three things early on: land, fiber and electricity. Water was evaluated, but in many markets it was treated as just another utility connection. That hierarchy is changing, and for some projects water strategy will have to be developed at the same time as power strategy. The main options include:

• Reclaimed wastewater. Cooling with treated municipal wastewater that would otherwise be discharged, rather than drinking-quality water, is much easier to defend politically in many communities.
• Closed-loop cooling. Water circulates repeatedly instead of being continuously evaporated, sharply reducing consumption.
• Air or dry cooling. These designs use very little water but usually require more electricity, especially in hot weather.

That last point reveals one of the most interesting trade-offs in AI infrastructure: saving water can require more power, and saving power can require more water. The best answer in Oregon may be wrong in Arizona. A developer might reasonably choose evaporative cooling in a water-rich region where electricity is the binding constraint, and accept higher power consumption in a desert in exchange for much lower water use. There will probably never be one universal answer, but it is now clear that water engineering cannot be separated from energy engineering.

The Real Issue Is Speed and Concentration

The challenge facing the AI industry is not simply that data centers consume resources; many industries do. The problem is the speed and concentration of the demand. A city can absorb hundreds of new homes arriving gradually over ten years. It is much harder to absorb the infrastructure demand of tens of thousands of households arriving at a single industrial address within eighteen months.

Electric utilities are already experiencing exactly this, as hyperscale developers request hundreds of megawatts from systems that historically grew a few percent per year. Water utilities may face the same shock, and they cannot expand on venture-capital timelines. New reservoirs need environmental review, land and years of permitting. New wells may face pumping limits. Pipelines and treatment capacity must be designed and financed, and a drought can undo assumptions made years earlier. That makes water a genuine development risk, not a footnote in a sustainability report.

What This Means for Developers

The lesson from Tucson and The Dalles is fairly direct. Inexpensive land is no longer enough, and neither is transmission capacity. The most desirable AI data center locations will increasingly be those where several resources overlap: reliable electricity, adequate transmission, fiber, suitable land, and a water system that can support industrial demand without triggering a political battle. Those locations are limited, and local governments appear increasingly willing to decide that community water security comes first.

Tucson shows that a project can be economically attractive and technically feasible and still fail if residents and elected officials consider the resource trade-off unacceptable. The Dalles shows that even a data center that has operated successfully for decades can find its growth reshaping local debates about infrastructure, transparency and the environment. Neither case means America should stop building data centers; the computing demand behind them is real. But the next phase will require developers to show that they can coexist with the communities around them. In practice, that means:

• Assessing water before the project reaches a zoning hearing, not after.
• Considering recycled water before requesting potable supply.
• Explaining drought resilience before residents ask about it.
• Being prepared to disclose water use and to invest in water infrastructure earlier than in the past.

For the last few years, people following AI infrastructure have asked where all the electricity will come from. That remains a defining question. Increasingly, a second one sits beside it: where will all the water come from, and who decides who receives it? The communities of Tucson and The Dalles are beginning to answer, and the decision will not belong to the data center industry alone.

EnergyInnoCore Editor

Independent analysis of power, grid and policy for AI data centers. Every article lists its sources. About the editor · Editorial policy

This article is for general information only and is not engineering, legal, financial or investment advice. See the disclaimer.