The Economist Special Report: US Data Centers Face Widespread Opposition, AI Computing Expansion Is Braking

Author: The Economist

Compiled by: Shenchao TechFlow

Shenchao Briefing: The Economist visited places on the ground such as Ohio and documented an anti–data center movement sweeping across the United States. In the first three months of 2026, at least $42 billion and 3.5GW of projects were canceled due to resident protests. From Democratic to Republican voters, opposition rates are all above two-thirds. This is not just a NIMBY effect—it is a collective public fury over AI infrastructure’s “physical intrusion.” For the AI industry, a computing-power bottleneck already exists, and the build-out side is being politically blocked.

Looking out from the top of a backyard slide, in April this year the view was still Ohio farmland, dense woods, and pretty wooden houses. Now this landscape has been swallowed up by six giant windproof tents. These tents are usually used to park fighter jets or set up temporary facilities in disaster zones. Soon, they will house cutting-edge semiconductors worth about $30 billion. Combined with supporting gas turbine generator sets, the entire site covers an area roughly equivalent to an airport terminal. If Meta’s “Prometheus” data center comes online as planned in 2026, it will put a full 1 gigawatt (GW) of power into artificial intelligence—enough to supply 1 million households with electricity, roughly comparable to the output of a large nuclear power plant.

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

超-large data centers planned to train frontier AI models before 2030 will not be built in existing server clusters in places like Virginia or California. Instead, they will land in the emerging “Silicon Heartland” made up of Michigan, Wisconsin, and Ohio—or in southern states such as Louisiana, Mississippi, and Texas. Amazon, Google, Meta, Microsoft, and Oracle together are pouring as much as $750 billion into these efforts, and data center operators such as CoreWeave—as well as real estate developers funded by Wall Street—are also moving in. According to Moody’s estimates, from 2026 to 2030 global investment in AI data centers will total about $3 trillion, with the bulk in the United States.

This money will raise the total amount of U.S. AI computing power—from today’s less than 12GW to about five times by the end of the decade. And almost in every corner of the country, people with every political viewpoint are furious about it.

Caption: The Economist illustration—data center construction and expansion plans across the United States

$42 Billion in Projects “Killed” in Three Months

The reasons people find it so infuriating are plenty: ugly construction; the noise from generators and cooling systems; transmission towers cutting up the skyline like a skeletal army; and worries about water-source pollution. Polls show Americans would rather live next to a nuclear power plant than next to a data center. Political heat around this issue has surged, and gubernatorial candidates facing elections in November are being repeatedly questioned about their positions.

Local activists have already notched wins. In the first three months of 2026, at least 20 data center projects were canceled, involving $42 billion in investment and 3.5GW of power. Over the past three years, projects totaling $85 billion have been cut, including smaller proposals submitted by Amazon and Meta. Residents in Cedar Rapids, Iowa are resisting Google’s construction plans. Multiple townships in Michigan issued moratoriums after OpenAI broke ground in Saline despite local opposition.

Not Just a NIMBY Effect

This resistance goes far beyond “don’t build it in my backyard.” In April, a Pew Research survey found that Americans who had only “heard of data centers” opposed them to the same degree as people living within five miles of one.

For years, philosophers have worried that an out-of-control AI would drain all human resources for a single goal, covering the earth with servers. Sam Altman and Dario Amodei spent years warning that AI would render most people unemployed or be weaponized. Now, they need infrastructure—and it is showing up at ordinary people’s doorsteps, looking like something hauled in from a war zone. Residents across the country stand at municipal meetings begging for projects to be slashed, hoping to slow the rollout of this technology. Will they succeed?

Energy Minister: “We Must Stay Ahead of China”

This is not only an issue for the AI industry. In an interview with The Economist, U.S. Energy Secretary Chris Wright said, “We must keep a lead over China.” Ensuring U.S. leadership in artificial intelligence is his “overriding goal” for his term. “We have to get these data centers permitted, built, and powered.”

At present, about 1–2GW of data center capacity in the United States is used to train frontier models—serving major providers such as Anthropic, OpenAI, and Google, as well as challengers such as Meta and xAI. Based on rough estimates, about 10GW of capacity can be used for inference, allowing customers to ask questions of models, write code, or perform other tasks. But after demand for AI tools surged in early 2026, available computing power became seriously insufficient. Anthropic limited model usage; OpenAI cut high-compute video tools; and Microsoft raised the price of its programming assistant to a level that pushes some programmers to write code themselves again.

New data centers were supposed to relieve this pressure. Large projects currently under construction are expected to add nearly 30GW of capacity by the end of 2028. But the computing power needed to train new models is also climbing rapidly. In a white paper last year, Anthropic pointed out that training a frontier model could require 5GW by 2028. Research firm Epoch AI estimates this could rise to 16GW by 2030. If that is true, most of the additional capacity coming online in the next few years will be consumed by training.

Can the Supply-and-Demand Gap Be Closed With Technology?

Several factors could curb demand for more new projects. Chip energy efficiency improves over time, meaning the same amount of electricity can produce more computing power. Crypto mining facilities are being repurposed for AI. Cerebras CEO Andrew Feldman points out that when inference is needed but training models is not, massive facilities like these may not be required in the same way—part of the capacity of existing data centers can be used instead.

But this may still not be enough. Right now, only the programming field has genuinely been disrupted by AI. Other industries AI could transform—such as law, finance, and media—are still in the early stages of adoption.

Even “Data Center Cities” Turn Against Them

Most of the projects currently under construction were approved before the resistance reached the level it has today. Some projects barely cleared the threshold. The OpenAI project in Saline, Michigan was rejected by the municipal council, but ultimately moved forward by exploiting a legal loophole: the county lacked industrial land-use planning and violated “exclusionary” zoning laws.

Even regions that have long welcomed data centers are reversing course. In March 2025, Loudoun County in Virginia (known as the “Data Center Corridor”) scrapped rules friendly to data center development, and new projects now have to go through a “special exception” process that includes public hearings. Texas has the second-highest number of data centers in the country, and the city of San Marcos has issued a moratorium.

The Water Myth and the Reality of Electricity

Americans oppose data centers partly because of concerns about their impact on communities and the environment—some are reasonable, and some are misleading. For example, the claim that “AI data centers consume large amounts of water” was popularized in 2025 by a book, but the book was based on a serious calculation error. A mid-sized data center uses roughly the same amount of water per year as two golf courses, and if water-recycling technology is used (many new projects have it), water use could drop dramatically.

The electricity issue is more concrete. According to SemiAnalysis, U.S. states currently have about 1 terawatt (1000GW) of large-load grid interconnection applications, and almost all of them come from data centers. That is equivalent to the entire generation capacity of the U.S. power grid (peak roughly 1250GW). While Americans use about 470GW of electricity on average throughout the year, the summer peak can reach 750GW, and power companies also need to maintain a 15–20% buffer on top of that. Truly reliable, always-adjustable electricity available at any given moment is only about 975GW.

This has raised concerns about rising electricity prices for consumers and other businesses. At the moment, there is not much credible evidence that this has already happened. As demand grows, utilities can spread the costs of upgrades across more users. Data center operators always install backup power to guard against supply disruptions, and in extreme cases (such as storms) they can cut their own usage. Alistair Speirs, head of data center construction at Microsoft, says, “We want to be good grid citizens.” He adds that batteries connected to hyperscale data centers give Microsoft the ability to “choose when to sip and when to gulp.”

But the huge data center investments planned in the coming years do require the United States to substantially increase power generation, and the infrastructure build-out will inevitably trigger new waves of opposition. The Department of Energy expects that by 2030 the U.S. will need an additional 50GW of generation capacity to support AI, and another 50GW for the manufacturing renaissance the government is aiming for. Minister Wright is skeptical of wind and solar projects that generate only intermittently. To that end, he blocked the shutdown of coal-fired power plants, supported restarting nuclear power plants, and backed the construction of new gas-fired power plants. It is expected that by 2030 more than one-third of data centers will generate their power entirely on their own, but that will make these projects even more conspicuous—while the remaining projects still rely on the grid.

The Ohio Experiment: Using Electricity Bills to Block Public Grievances

Ohio is now the fourth-largest concentration of data centers in the United States, and its strategy is smarter than most states. Last July, the state public utilities commission passed a rule: data center operators above a certain size must pay each month at least 85% of the cost for the reserved electricity capacity they applied for, regardless of whether they actually use it, so as to ease residents’ concerns about locals effectively paying for grid investment. This innovative approach was later incorporated into the “Ratepayer Protection Pledge” signed at the White House by technology companies this March.

Ohio’s provision is better than the White House pledge because it is legally binding. But even so, it has not calmed residents. About three-quarters of Democrats and two-thirds of Republicans in the state still oppose building data centers locally. So intense is the opposition that even though Trump won the state by 11 percentage points in the 2024 election, AI supporter Vivek Ramaswamy is polling neck-and-neck with his Democratic rival in the governor race.

The Federal Government Works Around Local Opposition

The Trump administration has ways to bypass local resistance. In March, the Department of Energy announced a 10GW mega-project on federal land in rural Piketon, Ohio, thereby avoiding the usual permitting process. The project is funded by SoftBank, which is controlled by Masayoshi Son, and will build a gas power plant to supply electricity for what will be the world’s largest data center.

“Just imagine—mud on your shoes in the fields of Appalachian farmers, with Commerce Secretary Lutnick, Energy Secretary Wright, Masayoshi Son, and me standing there… and every other person from up in the hills!” said Ohio Congressman Adam Holmes at the groundbreaking ceremony.

Piketon residents are not easily intimidated—this area was once the site of the United States’ nuclear enrichment program in the 1950s. Even so, state senator Shane Wilkin, who represents the district, is not relaxed. He serves on the state legislature’s data center committee in Ohio, and he told a story: “We asked the water department people to testify, and I asked him how many data centers in Ohio had obtained discharge permits. One. Then I asked a question whose answer I didn’t know—there’s always risk—whether they had any violation records. He said twice: both times were because paperwork was submitted late.”

Wilkin said he explained to voters that data centers would not raise electricity prices because they come with their own power; they would not pollute water sources because they do not discharge water. But residents’ response was: “I just don’t want it.”