There’s a lot to dislike: the ugliness of the buildings; rumbling of generators and cooling systems; Skeletal armies of new transmission towers are crisscrossing the landscape; Fear of contaminated water. Surveys show that Americans would sooner live next to a nuclear plant than in a data center. The matter has come to a head. Gubernatorial candidates facing voters in November are routinely questioned about where they stand.
Local activists are already claiming the skull. At least 20 data-center projects worth $42 billion, which would have used 3.5 gigawatts of power, were canceled in the first three months of 2026 after local opposition; Projects worth $85 billion have been canceled over the past three years, including smaller hubs proposed by Amazon and Metra. Residents of Cedar Rapids, Iowa, are protesting Google’s plans to build one there. Several townships in Michigan have passed moratoriums after OpenAI started a project in Saline despite local opposition.
Yet the resistance goes far beyond simple NIMBYism. A survey by pollster Pew Research in April found that Americans who have only heard of data centers are as opposed to them as those who live within five miles of one. Philosophers have long worried that a rogue AI – focused on a single goal like maximizing the production of paperclips – would ruin humanity by destroying all its resources, flooding the Earth with servers and rendering the planet uninhabitable. OpenAI’s Sam Altman and Anthropic’s Dario Amodei have warned for years that AI will put most people out of work or cause massive harm. Now the infrastructure they need is arriving at people’s doorsteps, and it feels like something out of a war zone. Residents across America are standing up at council meetings and begging for projects to be shut down in hopes of slowing down the progress of technology. Will they succeed?
This question is not just on the minds of the AI industry. “We need to be much ahead of China,” says US Energy Secretary Chris Wright in an interview with The Economist. He says that ensuring the US is a leader in artificial intelligence is a “overriding goal” of his tenure. “We need to be able to get these data centers permitted, built and powered on.”
Currently perhaps 1-2GW of US data-center capacity is dedicated to training frontier models from the major providers – Anthropic, OpenAI, and Google – as well as those trying to keep up with them, including Meta and XAI. This means that perhaps 10GW should be available for inference – allowing customers to use the model to ask questions, write code, or perform other tasks. Yet as demand for AI tools surged in early 2026, the available “compute” proved woefully inadequate. Anthropic has scaled back its use of models, OpenAI has scrapped its compute-intensive video tools and Microsoft has raised the price of its coding assistants so sharply that some programmers are returning to the lost art of writing the software themselves.
It is believed that new data centers will relieve this pressure. Construction is already underway on larger projects, which should provide approximately 30GW of additional capacity by the end of 2028. Yet the amount required to create new models is increasing rapidly. In a white paper published last year, Anthropic argued that up to 5GW would be needed to train a single frontier model by 2028. According to Epoch AI, a research firm, this figure could increase to 16GW by 2030. If this is true, most of the capacity coming online in the next few years will be absorbed by training alone.
Some forces will help prevent the need for more projects. Chips become more efficient over time, generating greater computing capacity from the same amount of power. Crypto-mining facilities are being repurposed for AI. And as Andrew Feldman, boss of chip maker Cerebras, points out, training a model doesn’t require the huge facilities needed for inference, meaning it’s possible to use parts of existing data centres. Still, this may not be enough. Currently the only one that has really been disrupted by AI is coding. Other industries in which it could revolutionize – such as law, finance or media – are still finding new adopters.
Most of the data-center projects already underway were approved and started before the response reached its current peak level. Some – like the OpenAI site in Saline, Michigan – were scratched by the skin of their teeth: the project was rejected by the council and was only able to move forward because the county had no land zoned for industrial use, which violated “exclusionary” zoning laws.
Even sectors that have long embraced data centers have turned against them. In March 2025, Loudoun County in Virginia – known as “Data-Center Alley” – eliminated regulations that made data-center development easier. New sites now require “special exceptions” involving public hearings. In Texas, which hosts the second-highest number of data centers after Virginia, the city of San Marcos has banned them.
Americans have turned against data centers partly because of understandable (if sometimes misguided) concerns about the impact on their neighborhoods and the environment. For example, the myth that AI data centers consume huge amounts of water was popularized by a book in 2025 that was based on a serious misconception. A medium-sized data center uses the equivalent of two golf courses of water annually, but it uses much less water if it includes water-recycling technology, as many do now.
Then there is the concern of power. According to SemiAnalysis, another research firm, there is currently one terawatt (1000GW) of large-load grid-connection requests for almost all data centers in US states. This is equivalent to almost the entire capacity of the US electricity grid, which can produce a maximum of 1250GW. And although US electricity demand averages about 470 GWh throughout the year, summer peak demand can reach 750 GWh and utilities prefer to maintain a 15-20% buffer above that. Furthermore, only about 975GW of capacity is reliably available on demand.
This has led to concerns over rising electricity prices for consumers and other businesses. There is so far little reliable evidence that this happens. The increased demand for electricity allows utility companies to spread the cost of upgrades over more users. And because data-center operators always establish a backup source in case their energy supply is disrupted, they can reduce their demand in the event of an extreme event such as a hurricane. “We want to be good grid citizens,” says Alistair Spears, who leads data-center build-out at Microsoft. He says the batteries typically attached to a hyperscaler’s data centers allow it to “choose” when to siphon and when to ingest from the grid.
Still, the huge investments made in data centers over the next few years will require the US to produce a lot of electricity – which is sure to result in infrastructure conflicts. The Energy Department expects the country will need to add 50 GW of generation by 2030 to support AI and an additional 50 GW for the manufacturing renaissance, the administration estimates. “We need to increase our capacity for dispatchable electricity,” says Mr. Wright, who is skeptical of wind and solar projects that produce only intermittent energy. To achieve this goal, he has prevented the closure of coal-power stations and supported the restarting of nuclear plants and the construction of more gas-fired plants. However by 2030 more than a third of data centers are expected to generate all of their electricity on-site, making these projects even more challenging, and the remainder will still be grid dependent.
Permission not granted
Ohio, now home to the fourth-largest concentration of data centers in the US, is more sensible in its approach than most states. Last July the state Utility Commission passed a requirement that each month data-center operators above a certain size must pay for at least 85% of the power capacity they’re asked to make available, even if they don’t use it, to assure Ohioans they won’t foot the bill for grid investments. This innovation became one of the ideas included in the “Ratepayer Protection Pledge,” which tech companies signed in the Oval Office in March.
Ohio’s provision is better than a pledge, because it is binding. Yet it has done nothing to provide relief to the residents of the state. Nearly three-quarters of Democrats and two-thirds of Republicans in Ohio still oppose local data-center development. The opposition is so great that, despite Donald Trump winning the state by 11 points in the 2024 presidential election, AI enthusiast Vivek Ramaswamy, who is running to become governor of Ohio, is facing a tough battle with his Democratic rival in the polls.
The Trump administration has the means to circumvent local opposition. In March the Energy Department announced a massive 10GW project on federal land in Piketon, rural Ohio, bypassing some of the normal permitting processes. The project will be funded by Japanese conglomerate SoftBank, run by Myoshi Son, which will build a gas-power plant to supply the world’s largest data center. “Imagine us standing in an Appalachian farmer’s field with Secretary Lutnick, Secretary Wright, Mr. Sun and me… and all the other hillbillies with mud on our shoes!” says Ohio state representative Adam Holmes, who attended the groundbreaking.
Piketon residents would hardly have hesitated: the area was home to America’s nuclear-enrichment program in the 1950s. Still, it’s not an easy time for state Senator Shane Wilkin, who represents the district and sits on the Ohio legislature’s data-center committee. “We had people from water utilities come in to testify and I asked them how many data centers in Ohio have discharge permits? One. And then I asked them a question I didn’t know the answer to – which is always a risk – did they have any violations? And they said two: for late paperwork.” Mr. Wilkin says he explains to his constituents that data centers won’t raise electricity prices because they bring their own electricity, and they can’t contaminate water supplies because they don’t spill their own water. But they just say: “Well, I don’t want that.”
