The Truth About AI Data Center Costs

SITUATION:

Data centers are nothing new. Amazon, Microsoft, and Google have operated large server facilities for roughly two decades, running the systems behind email, cloud storage, and online search long before artificial intelligence became part of the national conversation. Those earlier buildouts happened gradually, one facility at a time, without drawing much attention outside the industry itself.

What has changed is the pace of construction, the size of each new facility, and the reaction from the public and from lawmakers. AI-driven data center construction is moving through years of typical industry growth in a fraction of the time, the facilities themselves are far larger and draw far more electricity per site than earlier generations, and legislatures in dozens of states have introduced bills this year alone aimed at slowing, studying, or restricting new projects.

Rural areas typically employ fewer than 100 permanent workers once construction is finished.

Hundreds of new data centers are currently under construction or in the planning stages across the country, and the buildout is being driven largely by a small group of companies, including Amazon, Microsoft, Google, Meta, and OpenAI. One widely cited industry estimate projects global spending on data center construction will approach $7 trillion by the end of the decade.

Electricity demand tied to these facilities has become the most visible source of public frustration. Utility companies asked state regulators to approve more than $29 billion in rate increases in just the first half of 2025, more than double what utilities had requested over the same period the year before. Nationally, residential electricity prices have climbed roughly 40% since 2021, and many people living near new data center construction believe the two trends are directly connected.

Cooling the servers inside these facilities also requires large amounts of water, and a single large site can use millions of gallons on a hot day. A significant share of new construction is concentrated in Texas, Arizona, and other states that are already dealing with drought conditions or limited water supplies, which has made water use a central issue in many local debates.

Specific projects have become flashpoints in their own communities. In San Marcos, Texas, more than a hundred residents turned out to a single public hearing to oppose a proposed $1.5 billion data center campus, and the project was ultimately rejected. In Denver, the city council voted to impose a one year moratorium on new data center zoning approvals. Elsewhere, developers have taken the opposite approach and sued local governments directly, including a Michigan township, in an effort to overturn zoning decisions that blocked their projects.

Local governments are typically persuaded to approve these projects on the strength of promised jobs and future tax revenue. In Georgia, the state’s own audit of expected economic benefits from data center development was later revised, with job creation estimates cut by roughly 70% from what had originally been presented to the public and to lawmakers.

THE TRUTH:

One of the most widely repeated claims in this debate holds that people living near data centers have seen their electricity bills rise by as much as 267% over the past five years. When fact-checkers examined that number, they found it actually describes wholesale electricity prices, which is what utilities and other large buyers pay to purchase power on the open market, not the retail price that shows up on an individual household’s monthly bill. Wholesale costs do eventually work their way into what customers pay, but not on a one to one basis, and several fact-checking organizations rated the 267% claim, as applied to household bills, mostly false.

That correction does not mean there is nothing to the underlying concern. National residential electricity prices have in fact climbed close to 40% since 2021, which independent analysts describe as the fastest five year increase on record. Utilities requested $31 billion in rate increases in 2025 alone, more than double the amount they had sought the year before, and close to half of those requests were still working their way through state regulatory approval as of early this year, meaning a further round of bill increases is already in the pipeline regardless of what happens with data center construction going forward.

Not every researcher agrees on how much of that increase should be attributed to data centers specifically. A 2026 study from the Electric Power Research Institute, an independent nonprofit funded by a mix of government, academic, and utility industry sources, concluded that data centers were not the primary driver of rising electricity prices nationwide. That study pointed instead to a combination of aging grid infrastructure that needs costly replacement, rising costs for the equipment utilities need to build and maintain power lines, and state clean energy mandates that require investment in new generation sources.

At the same time, the picture looks different in places where data center construction is heavily concentrated. PJM Interconnection, the organization that operates the electrical grid across fourteen mid-Atlantic and Midwest states including Virginia and Maryland, found that data center demand was the primary reason behind $23 billion in customer price increases that will affect ratepayers in that region through at least 2028. In other words, the national picture and the picture in a data center heavy region can both be true at once. Nationally, data centers may not be the single biggest factor, but in the specific places where they are concentrated, they can be.

In response, at least a dozen states, including Oregon and Delaware, have begun requiring data centers to pay separate, higher electricity rates than other customers. The intent behind these policies is straightforward: regulators want the companies whose demand is driving the need for new power plants and transmission lines to shoulder more of that cost directly, rather than spreading it evenly across all ratepayers, including households that see no benefit from the data center’s presence.

Water raises a similar pattern of a modest share nationally paired with serious concentrated impact locally. Data centers account for a small fraction of total water use across the country as a whole. But direct water consumption for cooling alone reached an estimated 17.4 billion gallons in 2023, and that figure is projected to climb as high as 73 billion gallons annually by 2028 as construction continues to accelerate. Those numbers do not include the additional water consumed indirectly through the electricity generation needed to power the facilities in the first place.

Water use has become the single most commonly cited complaint in local disputes over proposed data center projects nationwide, ahead of concerns about noise, traffic, or other environmental impacts. In Texas specifically, data center facilities consumed more than 50 billion gallons of water in 2024 alone, an amount roughly comparable to what would be needed to supply a city the size of Austin for several months.

Some of the larger technology companies building these facilities have already begun shifting toward closed loop or air based cooling systems, which recycle the same water repeatedly through sealed pipes rather than allowing it to evaporate into the air the way traditional open cooling systems do. These newer systems can cut a facility’s water consumption by as much as 70% compared to older designs, which suggests the water problem, while real, is not an unavoidable requirement of building data centers, but rather a consequence of which cooling technology a company chooses to install.

Local fights over specific projects have produced real, documented outcomes on both sides of the debate, not just abstract controversy. In San Marcos, Texas, residents packed a public hearing to oppose a proposed $1.5 billion campus, citing concerns about water scarcity in the area and about a power demand that could have reached more than twice the city’s existing peak electrical load. The project was ultimately rejected. In Denver, the city council voted unanimously in May to impose its own moratorium on new data center zoning approvals, giving officials time to study the impact before allowing further construction.

A research firm called Data Center Watch tracks these outcomes nationally, and it found that $130 billion worth of data center projects were blocked, delayed, or withdrawn in a single quarter this year. That figure alone already matches the total amount of project value affected by opposition across the entirety of 2025.

Public polling backs up the scale of this resistance. A Gallup survey found that 71% of Americans oppose having a data center built in their own area, a higher opposition rate than Americans report for nuclear power plants, which are often considered one of the more controversial types of infrastructure to site locally. The opposition also runs across party lines rather than breaking down along the usual political divide, with the same poll finding 75% of Democrats and 63% of Republicans opposed to a data center in their community.

The scale of the legislative response reflects that public sentiment. More than 300 separate data center bills were introduced across at least 30 state legislatures in just the first six weeks of 2026. These bills range from outright construction moratoriums, to new zoning restrictions specific to data centers, to requirements that companies disclose more information about water and electricity use before receiving approval to build.

Developers and technology companies have pushed back just as forcefully where they believe local governments have overstepped. Companies have filed lawsuits against townships in Michigan and North Carolina seeking to overturn local zoning votes that blocked their projects, arguing the decisions were improper or exceeded local authority. In Texas, a state senator pressed the state attorney general to intervene and block a county from imposing its own temporary moratorium, illustrating how far these disputes can escalate beyond the local level.

The gap between what is claimed and what is actually delivered is sharpest when it comes to jobs and tax revenue, which are the two benefits most commonly used to justify local approval and public subsidies in the first place. Good Jobs First, a nonpartisan watchdog organization that tracks economic development subsidies nationwide, found that not a single state currently discloses both the number of jobs a data center project promises at the time it is approved and the number of jobs that project actually creates once it is built and operating.

That matters because the tax breaks involved are substantial. A tax abatement is a reduction or full exemption from taxes a company would normally owe, typically in exchange for a promise of investment, jobs, or both. Texas alone is estimated to have lost roughly $1 billion dollars in tax revenue during fiscal year 2025 because of exemptions granted to data center companies, and the state discloses only the names of the companies that received the benefit, without publishing the dollar amounts involved, the number of jobs promised, or even the locations of the projects receiving the exemption.

Virginia, which hosts the largest concentration of data centers in the world, grants sales and use tax exemptions worth close to a billion dollars in state and local revenue each year specifically for data center equipment. Despite the size of that public cost, the state does not disclose which individual companies are receiving the benefit, which makes it effectively impossible for Virginia taxpayers to verify whether the companies benefiting from the exemption are the same ones that promised specific numbers of jobs or investment in exchange for it.

Georgia offers the clearest documented example of promised benefits failing to hold up under closer review. The state’s own economic development audit initially projected that data center development would create 28,350 construction jobs, 5,471 permanent data center operations jobs, and $474 million in tax revenue, figures that were used publicly to justify the tax incentives offered to attract that development in the first place.

A later revision of that same audit, conducted by the University of Georgia’s Carl Vinson Institute of Government, found the original projections were far too optimistic.

The revised estimate for construction jobs came in roughly 70% lower than originally touted, the estimate for permanent operations jobs was similarly reduced, and the tax revenue estimate ended up 48% lower than the number originally presented to the public. It is worth noting this correction came from the state’s own auditors conducting a more careful review, not from outside critics of the industry.

Separate academic research from Georgia Tech reinforces the same pattern nationally. Because data center construction is extremely capital intensive, meaning it requires enormous upfront investment in buildings and equipment relative to the number of people needed to keep the facility running, data centers built in rural areas typically employ fewer than 100 permanent workers once construction is finished, regardless of the far larger job figures often used to win local approval in the first place.

The historical context matters in explaining why all of this feels new. Data centers have operated for roughly two decades with comparatively little public controversy or political attention, not because the facilities themselves were silent or unobtrusive, since large server operations do generate real noise and require significant physical infrastructure, but because their smaller size, slower construction pace, and more modest power and water requirements rarely strained local electrical grids, water systems, or zoning processes enough to draw sustained public notice.

AI driven demand changed that equation on every front at once. It compressed years of typical industry growth into a dramatically shorter construction timeline, while simultaneously pushing the physical size, electrical demand, and water use of individual facilities well beyond what many local grids, water utilities, and zoning boards were originally designed or staffed to accommodate.

That mismatch between the speed of AI driven construction and the pace at which public infrastructure and oversight can reasonably adapt, rather than data centers as a concept, is what (in part) produced the current wave of local opposition and state legislation.

WHY IT MATTERS:

A reasonable starting instinct for many readers is skepticism toward a company that shows up promising hundreds of jobs and asking for a lighter tax bill in return for local approval. In this particular case, the available record actually backs up that instinct rather than undercutting it.

Georgia’s own state auditors, conducting their own follow up review rather than responding to outside political pressure, concluded that the original job and revenue projections used to justify the state’s data center incentives were significantly overstated.

Public money and public infrastructure deserve honest accounting regardless of how advanced or promising the underlying technology sounds. When a state government exempts a company from close to a billion dollars in taxes without disclosing which specific company received that benefit, or how many jobs actually resulted from it, that is fundamentally a transparency problem rather than a referendum on artificial intelligence itself. The same standard of disclosure would reasonably be expected of any other industry receiving a public subsidy of that size.

Many of these fights are, underneath the specific details, ultimately about who gets to decide what happens inside a given community.

Denver’s city council voted for its own moratorium on new construction. Residents in San Marcos packed a public hearing and, in the end, their opposition carried the day.

Where developers have instead responded by suing townships directly to overturn local zoning votes, or by lobbying a state attorney general to override a county level moratorium, that shift in tactics from persuasion to legal and political pressure is worth taking seriously on its own terms.

Not every objection carries the same weight, and it is worth being honest about that distinction as well. Concerns rooted in documented water rights disputes or verified rate increases rest on specific numbers that can be checked against public records.

Other opposition is closer to general discomfort with an unfamiliar type of industrial development moving into a community, which is a legitimate reaction but a fundamentally different kind of concern, and conflating the two risks weakening the credibility of the objections that are backed by hard evidence.

There is also a genuine national interest at stake that deserves acknowledgment alongside the concerns raised above. Keeping the infrastructure behind artificial intelligence, along with the jobs and economic activity tied to it, located inside the United States rather than ceding that ground to other countries such as China is a legitimate goal that most readers likely already share, regardless of their views on the technology itself. Recognizing the value of that broader goal does not require accepting any individual local deal on the exact terms it was originally pitched to a city council or county commission.

The debate underneath all of the specific disputes covered here is not really a question of whether artificial intelligence infrastructure should exist in the United States at all. It is a question of whether the public is getting an honest account of what these specific projects actually cost local communities and ratepayers, and what they actually deliver in return, rather than what a company’s public relations department or a project’s political supporters initially claimed.

When a company or a sitting legislator overstates the benefits of a project or understates its true costs to the public, treating that specific claim with the same skepticism a reader would apply to any other unverified promise is not a position against progress or against artificial intelligence. It is simply the same baseline standard readers already apply to sales pitches everywhere else in daily life, and there is no good reason this industry should be treated as an exception to that standard.

SOURCES:

• Environmental and Energy Study Institute, “Data Center Power Demands Are Contributing to Higher Energy Bills”
• PolitiFact, “How much have data centers increased electricity prices?” (June 2026)
• WRAL, “Fact-check: Did data centers cause US electricity bills to rise 267% over five years?” (June 2026)
• Fortune, “Electricity prices are up 40% since 2021, but data centers shouldn’t get all the blame” (May 2026)
• Fortune, “Data centers have already hiked electricity prices on the public by $23 billion” (July 2026)
• Stateline, “With electricity bills rising, some states consider new data center laws” (February 2026)
• AGU Advances, “Data Centers Water Footprint: The Need for More Transparency” (February 2026)
• MOST Policy Initiative, “Data Center Water Use” (April 2026)
• Yahoo News, “AI’s Thirst Trap: Data Centers Guzzle Water While Droughts Drain Communities”
• Data Center Frontier, “Community Opposition Emerges as New Gatekeeper for AI Data Center Expansion”
• TechTimes, “AI Data Center Backlash: Zoning Notice Rule Kills World’s Largest Planned Campus” (July 2026)
• The Conversation, “How local communities are challenging Big Tech data centers’ noise, pollution and rising electricity bills”
• Fortune, “Harvard scholar: the data-center backlash is just getting started” (June 2026)
• Good Jobs First, “Most States Fail to Disclose which Data Center Companies Get Huge Tax Breaks” (November 2025)
• Good Jobs First, “Cloudy Data, Costly Deals: How Poorly States Disclose Data Center Subsidies” (November 2025)
• Yahoo Finance/WSB-TV, report on revised Georgia data center audit figures
• Futurism, “The Economics of Data Centers Creating Jobs Are So Bad That They Sound Like a Joke”