What the Data Center Debate Gets Wrong

The backlash to AI infrastructure is growing louder—and more disconnected from the policies that actually govern its impacts.

Courtesy George Frey/AFP via Getty Images

In a remote part of northern Utah, where sagebrush extends for miles and the nearest homes are few and far between, officials just voted to approve one of the largest data center projects in American history.

The proposal, backed by Canadian investor Kevin O’Leary, would transform a large swath of Box Elder County into a massive “hyperscale” data-center campus. At full buildout, it could reach a capacity of nine gigawatts—more than double the state’s current electricity consumption—and cover an area the size of Washington, D.C.

The project—like many others across the country—has become a lightning rod. County officials had delayed the vote after pushback from residents, and crowds packed local meetings to voice concerns. Environmentalists warned of dire consequences for water, air quality, and the surrounding region.

And yet, as Utah governor Spencer Cox put it last week, the site itself is about as ideal as it gets. It is remote, adjacent to a major natural gas pipeline, and far from residential neighborhoods. “If you can’t put this here,” Cox said, “then we can’t put them anywhere.”

These tensions are not unique to Utah. Across the country, data centers are becoming a flashpoint in local and national politics, with communities raising alarms about water use, electricity demand, and the broader implications of artificial intelligence. Some policymakers are now calling for moratoria on new data centers. In a few cases the backlash has taken a dark turn, with threats and vandalism aimed at public officials and industry leaders.

Something about the issue has clearly struck a nerve. But the debate over data centers is not just heated—it is becoming increasingly detached from the policies and institutions that actually govern the centers’ impacts on surrounding communities.

The discourse tends to treat resource use in the simplest possible terms. Water is treated as if it were simply “taken” from a shared pool. Electricity demand is assumed to translate directly into higher residential energy bills. These framings are intuitive, but they are often divorced from how these resources are actually governed in practice. Data centers don’t operate in a vacuum. They operate within legal and institutional frameworks that determine who can use water, how power is supplied, and how competing demands are resolved.

The real issue, then, isn’t whether data centers use too much water or energy, but whether the policies and institutions that govern those resources are equipped to handle these new demands, and where they fall short. That is the conversation we are not having. Instead, the debate defaults to panic, moratoria, and blunt prohibitions, making it harder to see where real reform is actually needed.

Consider an issue that has drawn some of the most intense scrutiny: water use. Data centers rely on water to keep servers from overheating. That demand has drawn concern, especially in arid regions. In Utah, opponents of the Box Elder project have pointed to the rapid decline of the nearby Great Salt Lake and warned that the new data center could exacerbate already strained supplies.

In recent months, a growing body of analysis has pushed back against claims that data centers are uniquely water-intensive. In aggregate, they are not. Compared to agriculture, golf courses, or even beer production, total data center water consumption is relatively modest. In many regions, it is a rounding error. Earlier facilities relied on evaporative cooling, which continuously vents water to the atmosphere. But newer data centers, including the proposed Utah project, use closed-loop recirculating systems that cycle the same water repeatedly.

But the relevant question is not how much water data centers use in total. A data center could account for a tiny share of statewide water consumption and still trigger serious local conflicts if it draws from a scarce aquifer or competes with other users for a common supply. Conversely, it could use a meaningful amount of water without much controversy if that water is acquired through existing rights, transferred from lower-value uses, or returned to the system in ways that preserve downstream flows.

In other words, the impact of data center water use is not determined by gallons alone. It depends on the policies that determine how water rights are governed.

In Utah—as in much of the American West—water is not an open-access resource. It is governed by well-defined rights that can be bought, sold, and transferred. New users must acquire water not by simply diverting or pumping at will, but by purchasing rights from existing holders. This process forces a comparison between competing uses and creates a mechanism—price—for deciding which ones persist.

In the case of the Box Elder project, its developers have so far secured rights to 1,900 acre-feet of water—roughly what a mid-sized Utah farm might use annually to irrigate 400 to 500 acres of alfalfa or hay. Those rights were acquired from an agricultural user, not carved out of a common pool at others’ expense. The data center’s water use doesn’t increase total withdrawals from the system; it transfers an existing allocation from one user to another. The developers say they plan to purchase rights to roughly 3,000 acre-feet in total for the project.

The institutional details matter even more than that, however. When it comes to the shrinking of the Great Salt Lake, the relevant question isn’t how many gallons a project uses in the abstract. It’s how consumptive that use actually is compared to what it replaced.

For example, with agricultural irrigation, a significant portion of the water applied to a field is lost to evapotranspiration and never returns to the watershed. In a closed-loop data center, by contrast, consumptive loss is near zero, and periodic “flushing” of the system returns water to the watershed that was previously depleted by the agricultural operation. On balance, that means the project may be net neutral, or even a modest net positive, for the inflows to the Great Salt Lake.

Furthermore, the data center operates under the same basic water constraints as any other user. It cannot simply increase its consumption at will. If it needs more water than originally projected, it must secure additional, existing water rights from willing sellers.

The same pattern shows up in debates over energy. The Utah project’s scale has fueled fears that it will overwhelm the state’s electricity grid and drive up rates for existing customers. But recent legislation in Utah creates a framework that addresses precisely this concern, allowing projects like this one to generate its own power on-site rather than drawing from the existing grid. Under this model, the project’s energy demands don’t hit the existing grid at all, and officials say it may even supply excess power back to the grid, which could result in lower prices for residential consumers. Again, the issue is not whether the project uses energy. It is how that energy is supplied, and under what legal and policy constraints.

These distinctions are rarely part of the public conversation. Instead, concerns about water, energy, and land use are often bundled together and treated as if they call for a single, sweeping response. The result is less a coherent policy framework than a kind of ambient opposition to “data centers” as such, often resulting in calls for moratoria or outright bans.

That is why proposals to simply “pause” data center development are so misguided. As the progressive energy and climate scholar Holly Buck recently argued, bans on data centers do little to slow AI development. Instead, they simply shift it elsewhere, often to places with weaker safeguards, while sidestepping the real policy questions at hand. Writer and policy advocate Nat Purser put it more succinctly: “pausing isn’t policy.” Attempting to address many distinct issues through a blanket moratorium makes it less likely that any of them get addressed.

The policy lens reveals where resource governance systems are working and where they are not. In Utah, surface water rights are well-defined and tradable. The state has also closed the Great Salt Lake basin to new groundwater claims, meaning data centers can’t simply drill new groundwater wells to satisfy their water demands. Instead, they must compete in an existing market for scarce rights.

Not every place looks like Utah. In parts of Arizona, for example, the picture is more complicated. Most of the state’s existing data centers operate within a rigorous “assured water supply” policy framework that requires municipal water providers to demonstrate 100-year supply sufficiency before committing water to new large users. This helps ensure that data centers’ water demands don’t come at the expense of existing users. But a growing pipeline of proposed projects sits outside those boundaries, where groundwater regulation is limited or absent.

The importance of these institutional frameworks is illustrated by a recent episode in Tucson. When the city council voted unanimously to block a data center last year on environmental grounds, it voided a negotiated deal requiring the developer to fund $100 million in reclaimed water infrastructure and commit to returning more water to the system than it consumed. The project proceeded anyway under county jurisdiction—drawing on the same aquifer, but with fewer constraints.

The question is not whether data centers use water, or energy, or land. Everything does. The question is whether the systems governing those resources are equipped to handle new demands—and where they aren’t, what it would actually take to fix them. Figuring that out requires a different kind of debate than the one we’re currently having.