The Economist Special Report: U.S. Data Centers Face Nationwide Backlash, Slowing AI Computing Expansion

Author: The Economist

Translated and edited by TechFlow

TechFlow Introduction: The Economist visited Ohio and other states to document an anti-data-center movement sweeping across the United States. In the first quarter of 2026 alone, at least $42 billion worth of projects—representing 3.5 GW of capacity—were canceled due to local protests. Opposition exceeds two-thirds among voters across both Democratic and Republican parties. This is not merely “Not In My Backyard” (NIMBY) sentiment; it reflects a collective public anger over the “physical intrusion” of AI infrastructure into everyday life. For the AI industry, computational bottlenecks already exist—and construction is now being politically blocked.

From the top of a backyard slide, the view this past April still showed Ohio’s farmland, dense woods, and charming wooden houses. Today, that landscape has been swallowed by six massive wind-resistant tents—typically used for parking fighter jets or erecting emergency shelters in disaster zones. These will soon house roughly $30 billion worth of cutting-edge semiconductors. Combined with accompanying natural-gas turbine power generators, the entire site covers as much area as an airport terminal. If Meta’s “Prometheus” data center goes online as scheduled in 2026, it will consume a full gigawatt (GW) of electricity for artificial intelligence—enough to power one million homes, roughly equivalent to the output of a large nuclear reactor.

The Rise of the “Silicon Heartland”: $3 Trillion Flowing Into AI Data Centers

The next generation of ultra-large data centers—designed to train frontier AI models through 2030—will not be built in existing server clusters in Virginia or California. Instead, they are converging on the emerging “Silicon Heartland,” comprising Michigan, Wisconsin, and Ohio—or in southern states like Louisiana, Mississippi, and Texas. Amazon, Google, Meta, Microsoft, and Oracle have collectively committed up to $750 billion. Data-center operators such as CoreWeave and Wall Street–backed real-estate developers are joining the rush. Moody’s estimates that approximately $3 trillion will flow globally into AI data centers between 2026 and 2030—with the bulk going to the U.S.

This investment will lift total U.S. AI computing capacity from under 12 GW today to roughly five times that level by decade’s end. And nearly everywhere across the country—across the political spectrum—citizens are furious.

Caption: Economist graphic showing planned data-center construction and expansion across the U.S.

$42 Billion Worth of Projects “Killed” in Three Months

There are plenty of reasons for opposition: ugly architecture; noise from generators and cooling systems; transmission towers slicing across skylines like skeletal armies; concerns about water contamination. Polls show Americans would rather live next to a nuclear power plant than a data center. Political attention has surged—gubernatorial candidates facing November elections are repeatedly grilled on their positions.

Local activists have scored victories. In the first quarter of 2026 alone, at least 20 data-center projects were canceled—amounting to $42 billion in investment and 3.5 GW of capacity. Over the past three years, projects totaling $85 billion—including smaller proposals from Amazon and Meta—have been scrapped. Residents of Cedar Rapids, Iowa, are resisting Google’s construction plans. Several townships in Michigan imposed moratoria after OpenAI broke ground in Saline despite local objections.

More Than Just NIMBYism

This resistance goes well beyond “Not In My Backyard.” A Pew Research survey from April this year found that even Americans who had merely “heard of data centers”—but lived nowhere near one—opposed them just as strongly as those residing within five miles.

Philosophers have long warned that an uncontrolled AI might consume all human resources in pursuit of a single goal—blanketing Earth with servers. Sam Altman and Dario Amodei spent years warning that AI could render most people unemployed—or weaponized. Now, the infrastructure they require has materialized on ordinary people’s doorsteps—and looks like something hauled straight out of a war zone. Across the country, residents stand before city councils pleading to kill projects, hoping to slow down this technology’s advance. Will they succeed?

Energy Secretary: “We Must Stay Ahead of China”

This isn’t just an AI-industry problem. Chris Wright, U.S. Secretary of Energy, told The Economist: “We must maintain a lead over China.” Ensuring U.S. leadership in artificial intelligence is his “overriding priority” in office. “We must get these data centers permitted, built, and powered up.”

Currently, roughly 1–2 GW of U.S. data-center capacity powers training for frontier models—serving major providers including Anthropic, OpenAI, and Google, plus challengers like Meta and xAI. By extension, about 10 GW is likely allocated for inference—allowing customers to query models, write code, or perform other tasks. But after AI-tool demand exploded early in 2026, available compute became critically scarce. Anthropic began rationing model access; OpenAI scrapped its high-compute video tools; Microsoft raised prices for its coding assistant to the point where some developers reverted to writing code themselves.

New data centers were meant to relieve this pressure. Large-scale projects currently under construction are expected to add nearly 30 GW of capacity by late 2028. Yet the compute needed to train new models is rising rapidly too. Anthropic’s white paper last year estimated that training a frontier model by 2028 may require 5 GW. Epoch AI estimates that figure could climb to 16 GW by 2030. If true, most newly added capacity over the next few years will be consumed entirely by training.

Can Technology Bridge the Supply-Demand Gap?

Several factors could curb demand for additional new projects. Chip efficiency improves over time, delivering more compute per watt. Crypto-mining facilities are being repurposed for AI workloads. Andrew Feldman, CEO of chipmaker Cerebras, points out that inference doesn’t require the same colossal infrastructure as model training—and can leverage spare capacity in existing data centers.

But that may not be enough. So far, only programming has been genuinely disrupted by AI. Other sectors AI could transform—law, finance, media—remain in early adoption phases.

Even the “Data-Center Capital” Has Turned Against Them

Most projects now under construction received approvals before the anti-data-center wave reached its current intensity. Some barely cleared regulatory hurdles. OpenAI’s project in Saline, Michigan, was rejected by the city council—but advanced anyway by exploiting a legal loophole: the county lacked industrial zoning rules, violating “exclusionary” zoning statutes.

Even regions long welcoming to data centers are reversing course. In March 2025, Loudoun County, Virginia—the so-called “Data Center Corridor”—repealed pro-data-center development rules. New projects now require a “special exception” process, including public hearings. Texas hosts the nation’s second-largest concentration of data centers—and the city of San Marcos has enacted a moratorium.

The Water Myth and the Electricity Reality

American opposition stems partly from legitimate—and partly misguided—concerns about community and environmental impacts. For instance, the claim that “AI data centers consume vast amounts of water” went viral in 2025 after a book popularized it—but the book relied on a serious calculation error. A midsize data center uses about as much water annually as two golf courses; many new facilities employ water-recycling technology, slashing consumption further.

The electricity issue is more concrete. According to SemiAnalysis, U.S. states currently host roughly 1 terawatt (1,000 GW) of large-load interconnection applications—nearly all from data centers. That equals the entire U.S. grid’s generating capacity (peak ~1,250 GW). Though average annual U.S. electricity demand stands at ~470 GW, summer peaks reach ~750 GW—and utilities must maintain 15–20% reserve margins atop that. Thus, reliably dispatchable power totals only about 975 GW.

This raises concerns about rising electricity rates for consumers and other businesses. So far, there’s little credible evidence this has occurred. Growing demand lets utilities spread upgrade costs across more users. Data-center operators always install backup power to guard against outages—and can curtail usage during extremes like storms. Alistair Speirs, Microsoft’s head of data-center construction, says: “We want to be good grid citizens.” He adds that batteries connected to Microsoft’s hyperscale data centers let the company “choose when to sip and when to gulp.”

Yet the massive data-center investments planned over the coming years truly do require substantial U.S. generation expansion—and associated infrastructure builds inevitably spark fresh opposition. The Department of Energy expects the U.S. to need 50 GW of new generation capacity by 2030 just to support AI—and another 50 GW to fuel the administration’s hoped-for manufacturing revival. Secretary Wright remains skeptical of intermittent wind and solar projects. To that end, he has blocked coal-plant retirements and backed nuclear-plant restarts and new gas-fired plants. By 2030, over one-third of data centers are projected to generate power entirely on-site—making those projects even more conspicuous—while the rest remain grid-dependent.

Ohio’s Experiment: Using Fees to Quell Public Anger

Ohio—now the nation’s fourth-largest data-center hub—has adopted smarter strategies than most states. Last July, its Public Utilities Commission approved a rule requiring data-center operators above a certain size to pay monthly for at least 85% of their reserved electric-capacity allotment—even if unused—alleviating residents’ fears that they’d foot the bill for grid upgrades. This innovative approach later formed part of the “Ratepayer Protection Commitment” signed by tech firms at the White House in March.

Ohio’s rule is stronger than the White House pledge because it carries legal force. Yet even this hasn’t calmed residents. Roughly three-quarters of Democrats and two-thirds of Republicans still oppose local data-center construction. Opposition runs so deep that although Trump won the state by 11 points in the 2024 election, AI advocate Vivek Ramaswamy polls neck-and-neck with his Democratic rival in the gubernatorial race.

Federal Government Bypasses Local Opposition

The Trump administration has tools to sidestep local resistance. In March, the Department of Energy announced a 10-GW mega-project on federal land in Piketon, rural Ohio—skirting normal permitting procedures. Funded by SoftBank, chaired by Masayoshi Son, the project will build a gas-fired power plant to supply the world’s largest data center.

“Imagine standing in an Appalachian farmer’s field, mud on your boots, alongside Commerce Secretary Lutnick, Energy Secretary Wright, Masayoshi Son, and me… and everyone else from the hills!” said Ohio State Representative Adam Holmes at the groundbreaking ceremony.

Piketon residents aren’t easily intimidated—the region hosted America’s uranium-enrichment program in the 1950s. Even so, State Senator Shane Wilkin—who represents the district and serves on Ohio’s Data Center Committee—faces tough questions. He recounts: “We invited water-department officials to testify. I asked how many data centers in Ohio hold discharge permits. Answer: one. Then I asked a question I didn’t know the answer to—which is always risky: Do they have any violations on record? ‘Twice,’ he said—both for filing documents late.”

Wilkin tells constituents data centers won’t raise electricity rates—because they bring their own power—and won’t pollute waterways—because they don’t discharge water. But residents reply: “I just don’t want one.”