the conversation gap — the infrastructure behind the intelligence

artificial intelligence may live on our screens, but the infrastructure beneath it is reshaping power grids, water systems, land, and the communities being asked to support it

Artificial intelligence still feels almost frictionless. A question goes into a box, and an answer appears seconds later. A company automates a process. A student summarizes a reading. A doctor searches through information that once would have taken hours to review. From the user's perspective, very little about this experience feels physical. But behind every interaction are buildings filled with servers, cooling equipment, electrical substations and backup generators, all connected to power grids, water systems and physical land.

That infrastructure is becoming difficult to ignore. On August 11, the U.S. Energy Information Administration projected that American electricity consumption will reach new records in both 2026 and 2027, as demand from data centers and other large users grows. In June, Lawrence Berkeley National Laboratory estimated that data centers could account for about 11.8 percent of all U.S. electricity use by 2030, with modeled scenarios ranging from 9.5 to 15.3 percent.

The political response is beginning to move just as quickly. On August 11, Chicago Mayor Brandon Johnson called for stronger rules governing data centers, specifically pointing to electricity affordability, water use, environmental justice and the public benefits communities receive from new projects. On July 14, New York imposed the country's first statewide temporary moratorium on new hyperscale data centers while it develops a regulatory framework addressing ratepayer costs, infrastructure and environmental impacts.

For years, the AI debate has focused mostly on what the technology can do. Increasingly, there is another question underneath it: what has to be built around AI, what resources does that infrastructure require, and who absorbs the costs?

the electricity has to come from somewhere

For much of the past two decades, American electricity demand was relatively flat. That era appears to be ending. Berkeley Lab's June analysis estimates that the country's data centers could consume almost three times their 2023 share of U.S. electricity by the end of the decade. The increase is not solely an AI story. Data centers also support cloud computing, storage, streaming and other digital services, while manufacturing and electrification are contributing to rising demand elsewhere. Still, the scale and concentration of new computing facilities are forcing utilities and regulators to rethink how the grid is built and paid for.

Virginia offers one of the clearest examples. The state contains the world's largest data-center market, and Dominion Energy's fuel costs have risen nearly 90 percent over five years, from $2.31 billion in 2021 to a projected $4.35 billion by mid-2027, as demand rises and the utility relies more heavily on wholesale electricity. Reuters reported on August 11 that a typical residential bill could rise about 13 percent to roughly $195 a month. Data centers are not solely responsible for that increase, but their rapid expansion is a major source of new electricity demand in the state.

Virginia regulators have already changed the rules in response. Beginning January 1, 2027, Dominion customers demanding 25 megawatts or more, including hyperscale data centers, will enter a separate rate class. Large users will be required to pay at least 85 percent of contracted transmission and distribution demand, even when they consume less electricity, in an effort to prevent infrastructure costs from shifting to other customers.

Federal regulators are confronting the same problem. On June 18, the Federal Energy Regulatory Commission ordered all six regional grid operators under its jurisdiction to justify or reform their rules for connecting data centers and other large loads to the grid. FERC explicitly identified consumer protection and cost allocation as central concerns, including the possibility that expensive network upgrades could otherwise find their way into bills paid by families and small businesses.

The policy question is more consequential than the technical language suggests. When a private company requires extraordinary amounts of power to expand its business, who should pay for the infrastructure required to supply it?

the water question is intensely local

Electricity is only one part of the physical footprint. Many data centers also use water for cooling, although the amount varies substantially by facility, climate and cooling technology. What matters is not simply the industry's national water use, but how that demand is concentrated within individual water systems.

In Virginia, researchers Eric Bonds, Braderick Hatch Jr. and Fiona Steffens reviewed public records and filed Freedom of Information Act requests in Louisa, Spotsylvania, Caroline and Stafford counties. They reported on June 19 that those local governments had allocated at least 19.6 million gallons of water per day in capacity for Amazon data-center cooling. That figure represents water capacity promised to the projects, not proof that the facilities are consuming 19.6 million gallons every day, but the distinction is part of the larger issue. Communities still have to plan water systems capable of meeting the demand they have committed to serve.

The geography changes the meaning of those numbers. Water demand concentrated in one county cannot be understood in the same way as water use averaged across the country. Local governments have to consider drinking-water supply, population growth, drought risk, wastewater infrastructure and competing industrial demands at the same time that developers are seeking increasingly large allocations.

Technology can reduce that pressure. Amazon's data center under construction in Gilroy, California, for example, is being paired with new recycled-water infrastructure intended to sharply reduce its reliance on potable water. The City of Gilroy says its existing infrastructure and planned upgrades can support the project's needs.

That variation is important because there is no single environmental footprint for a data center. Where one is built, how it is cooled and what resources already exist around it can determine whether the burden is modest or substantial.

the digital economy still needs land

The footprint extends well beyond utilities. An April 22 analysis by energy-data company Enverus identified approximately 136,000 buildable acres controlled by data-center developers and their subsidiaries across the lower 48 states. The figure represents potential development rather than land that has already been converted, but the land bank alone illustrates how physical the next phase of digital expansion has become.

In Goochland County, Virginia, a proposed data-center campus would stretch across roughly 871 acres and contain seven development areas. Developer Tract has described a potential 900-megawatt project. The land is already within the county's Technology Overlay District, but the project still requires a conditional-use permit and public review. Residents in a neighboring community of more than 500 homes have raised concerns about noise, water demand, diesel generators, transmission infrastructure and the transformation of the surrounding landscape.

Public land is also entering the buildout. On June 26, the Bureau of Land Management approved the Townsite Data Center project on 88.5 acres of federal land in Nevada. The agency explicitly described the approval as part of federal efforts to accelerate permitting for data-center infrastructure.

The footprint of these projects also extends beyond the property line. Large campuses may require new substations, transmission corridors, roads and utility connections, so the physical impact of a data center is not limited to the building that contains the servers.

That tension is especially visible in Gilroy. Amazon Web Services is constructing two data-center buildings totaling roughly 438,500 square feet on a 56-acre site. The project also includes a substation or switchyard and transmission upgrades to the existing power system. The Wall Street Journal reported on August 7 that many residents did not become aware of the project's scale until construction was underway, even though it had gone through the city's required approvals and environmental review processes. The controversy has prompted city officials to reconsider whether projects of this size should require greater public participation in the future.

The conflict in Gilroy points to a broader question. When a project can reshape a community's electricity use, water planning and physical landscape for decades, how much say should the people living around it have before construction begins?

what does AI sound like from next door?

Noise has become one of the most tangible complaints emerging around data centers and the energy infrastructure that supports them. Cooling systems, transformers, generators and other industrial equipment can operate around the clock. In several communities, residents describe the result not as occasional industrial noise but as a persistent hum, roar or vibration that changes the experience of being inside their own homes.

In Southaven, Mississippi, residents living near a facility supplying power to xAI data centers have described months of disruptive noise. CBS News reported on July 16 that homeowner Jason Haley compared it to an airplane hovering near his house and said it repeatedly interrupted his sleep. Haley and two other residents filed a lawsuit in June alleging near-constant noise and vibrations from the facility. The allegations have not been resolved in court, but the dispute is not unique. CBS identified similar 2026 lawsuits involving residents near data-center facilities in Wisconsin and Massachusetts who also complained of excessive or sleep-disrupting noise.

In Louisa County, Virginia, homeowner Austin Newsom is suing Amazon over conditions related to the construction of a campus planned to include 10 data center buildings and 3 substations. Newsom has reported heavy truck traffic, dust, late-night noise, light pollution, and periods when water from his private well turned brown. His lawsuit also alleges that construction vibrations damaged his home. Amazon sought dismissal of the nuisance claim, but a federal judge allowed it to proceed. The lawsuit has not established that Amazon caused all of the conditions Newsom describes, but his experience captures something that discussions of megawatts and investment rarely do: infrastructure has neighbors.

A data center can look abstract when measured in computing capacity. From the house beside one, the calculation can include trucks outside the window, lights at night, industrial sound, new transmission infrastructure and a landscape that no longer looks or feels the way it did before.

what does the public get in return?

States and municipalities compete for data centers because the benefits can be substantial. They bring major capital investment, construction activity and additional property and utility tax revenue. The harder question is how those benefits compare with the public resources used to attract and support them.

Washington state's Joint Legislative Audit and Review Committee offered a useful example in July. Its review of an urban data-center tax preference found that beneficiaries had received an estimated $42.4 million in tax savings between 2023 and 2026. They reported adding 53 permanent family-wage jobs and supporting nearly 300 temporary construction jobs. The state also found increased assessed property value and property-tax collections at participating sites, while cautioning that it could not determine how much of the economic activity occurred because of the tax incentive itself.

That is why the economic bargain is more complicated than either jobs or subsidies alone. Direct employment is only one measure of value, but tax revenue is only one measure of public cost. Communities may also be making decisions about grid capacity, water infrastructure, land use and long-term industrial development when they approve these projects.

The question is therefore not simply whether a data center brings economic growth. It is whether the benefits and burdens of that growth are being distributed fairly.

the conversation underneath AI

For years, the argument over artificial intelligence has centered on what happens inside the model. We have debated jobs, misinformation, education, creativity and productivity while the physical architecture underneath those systems has expanded quickly enough to become an electricity, water, land-use and local-government story.

By August 2026, those questions are no longer hypothetical. FERC is forcing regional grid operators to rethink how enormous new power users connect and pay for infrastructure. Virginia has created a separate rate class for large electricity consumers. New York has paused new hyperscale data centers while it writes a new regulatory framework. Chicago is asking for stronger protections around water, electricity affordability and community benefits. Residents elsewhere are confronting projects through zoning hearings, public meetings and lawsuits.

None of those developments answer whether the AI buildout is ultimately worth its physical cost. That calculation will differ by project and by community. What they do show is that the way we talk about artificial intelligence has fallen behind what it has become.

AI may arrive through a screen, but its expansion is increasingly dependent on shared and finite physical resources. It needs electricity generated somewhere, water drawn from somewhere, land built upon somewhere, and communities willing to live beside what is constructed.

That is the infrastructure behind the intelligence. The question now is whether the rules governing it can catch up before the costs and benefits have already been decided.