A liberal state does not distribute privileges or impose burdens according to popularity. Yet in recent years, data centers have been alternately privileged and burdened. When they were popular because of their promise of jobs and growth, states showered them with tax breaks. Now that they have become unpopular because of growing recognition of their demand for electricity and water, states and localities have rushed to impose peculiar regulations or even moratoriums on their construction.
Recently, New York State provided a prime example of how data center opponents can go too far with its selective moratorium on their construction. It pauses certain state permits for building or expanding data centers that can consume 50 megawatts or more, pending a special environmental review. At the same time, it exempts facilities primarily devoted to other activities, such as manufacturing and health care, even if they have the same electrical demand. But why should the same demands on electricity and water be disfavored when they support politically unpopular activities and accepted when they support politically popular ones? Exemptions should reflect only demonstrable differences in harm, not the political connections or popularity of those receiving them. This issue differs somewhat from simple NIMBYism, a political tendency that generally wants to stop or slow down all construction in the interests of incumbent homeowners or businesses, as discussed in the latest Law & Liberty Forum. The focus here is on targeting particular enterprises.
An enterprise’s demand for resources is not, by itself, evidence of market failure that the state needs to correct. Indeed, the market allocates resources to those willing to pay for them. If the poor need subsidies, they should be supported through taxation, not by special regulations that prevent profitable activity. Enterprises should be allowed to build under neutral rules. A liberal state protects the market economy because it reflects the voluntary choices of millions of its citizens.
Of course, data centers, like other enterprises, might harm others by imposing externalities like noise and pollution on their neighbors. But neutral laws can handle these effects too. There is no need to single out data centers for special regulation or a moratorium. One virtue of common-law torts is that they apply equally to all polluters and noisemakers. If these restraints are thought too weak, other laws can supplement them, such as permit requirements. The key is to apply all the laws based on common conditions.
The way regulation works should also give pause to supporters of New York-style policies. Utility regulations can allow the costs of new infrastructure to be passed on to all customers, effectively shifting some of the cost of serving a new customer onto existing customers. Higher market prices caused by additional demand are not themselves a subsidy; making existing customers pay for facilities built to serve a new business is. Data centers should pay the costs attributable to serving them, rather than shift those costs or investment risks onto captive customers. A well-known solution in utility rate regulation is to require businesses to pay the cost of new dedicated infrastructure. But this requirement should apply regardless of the business’s identity. The solution again is classical liberalism’s insistence on neutral laws.
These principles should address the specific complaints against data centers. They do increase demand for water, but the resulting price changes are the market at work. If new infrastructure is needed, data center businesses can pay for it or construct it themselves. While data centers may need water during periods of peak demand, when it costs more to supply, appropriate metering and rates that are sensitive to time allow extra charges to reflect that timing. If water is not now correctly priced at times of such demand, that is the reform needed. Correct pricing will encourage new supply, such as reclaimed water, as well as better cooling systems.
The answer is much the same with electricity. Infrastructure costs should be allocated according to the demands that they create and the benefits they provide. Neutrality does not require identical treatment of enterprises that impose different costs. A customer requiring a substation for their own use should make infrastructure payments that a household does not. But the relevant distinction is the burden imposed, not the business’s identity. Making an enterprise bear its costs is not discrimination.
Hard questions deserve our political and civic energy. The debate over data centers redirects that energy elsewhere.
Higher prices should prompt new ways to increase supply. In many localities, excessive government regulation is one of the biggest constraints. If localities streamline permitting for new transmission, everyone can benefit from greater supply that brings down rates. Georgia Power recently announced that its contracts with new large users, including data centers, are expected to lower residential electricity costs beginning in 2029.
Some data centers cause noise pollution, and a Virginia audit in 2024 on the issue suggests that continuous low-frequency noise is not subject to sufficient regulation. The solution is more comprehensive noise regulation that data centers, like everyone else, must follow. Federal and state law can address air pollution and carbon emissions if they are considered harmful. Those laws should also be applied neutrally to data centers. More generally, opponents of data centers should be challenged to identify externalities that existing law does not address. Those externalities can then be debated as a general matter rather than by focusing on data centers.
The liberal state should be as neutral in its privileges as in its regulations. When data centers seemed to be just a source of jobs and investment, states gave them tax breaks. The Virginia audit identified almost a billion dollars in these tax savings in fiscal year 2023 alone. But such selective subsidies undermine the liberal order. They invite political favoritism. They create a business culture in which enterprises compete for government favors rather than customers.
And the government does not have the knowledge to choose which industries will be winners in the long run. Economists Cailin Slattery and Owen Zidar found no strong evidence that tax incentives for specific firms produce broader economic growth at the state or local level. That does not mean states cannot promote business investment generally, but they are not qualified to choose particular kinds of enterprise.
The one exception for special privileges is facilitating national security. AI and the infrastructure that supports it are important for improving our military technology, particularly because our greatest geopolitical rival—the People’s Republic of China—is also our greatest rival in the AI race. National interests may support a case for federal subsidies. They are not a license for states and localities to compete to provide special benefits for particular business sectors.
Why has the sharp turn toward data center moratoriums occurred? One reason is the general movement toward economic illiberalism. Rent control has become a very popular cause. To be sure, rent control and data center moratoria have different effects. Rent control distorts market prices; data center moratoria suppress market entry. But both share the error of assuming that political direction of economic decision-making will improve social outcomes, rather than trusting decentralized market decisions.
Opposition to data centers may also serve as a proxy for opposition to AI. But local moratoria are a blunderbuss. They burden ordinary computing as well as AI and may redirect investment in centers abroad without controlling dangerous capabilities. An international restriction on computing infrastructure could conceivably form part of a justified policy promoting AI safety. But it would need to be defended on that basis. An argument about water use or electricity prices is a smokescreen.
Moreover, any restrictions would have to address issues that do not come up in the data-center discussion. Slowing AI in the United States does not mean AI will slow elsewhere. As a result, our geopolitical adversaries, like China, would benefit and our national security would suffer. Even apart from risks posed by other nations, one danger of AI is that it can empower malevolent nonstate actors. Advanced AI may also strengthen defenses against AI-enabled attacks. A policy that delays defensive tools while leaving attackers access to powerful models could make us less safe. The question is which arrangements best combine technical expertise, incentives for safety, and public accountability. A local moratorium does not begin to answer it.
Which approach would best address these risks—existing product-safety law, industry self-regulation, or AI-specific regulation? The answer will be found in the arrangements best combining technical expertise, incentives for safety, and national accountability. A local moratorium is not a sensible response along any of these dimensions.
A full assessment of AI must weigh its potential benefits as well as its risks. Recent breakthroughs in mathematics show AI’s capacity to transform science. It will boost medical discovery and longevity. It will likely help develop better solutions to other pressing problems, like climate change, and help invent technologies to prevent natural disasters. The AI balance sheet is only complete once we calculate its likely benefits.
These hard questions deserve our political and civic energy. The debate over data centers redirects that energy elsewhere. Classical liberalism limits government’s role to preventing external harms and providing domestic public goods where markets and civil society fall short. A politics entitled to address everything will likely fail to solve our most pressing dilemmas.
