Opposition to nearby AI data centers has risen from 42% to 71% in less than a year. Texas shows why the backlash began. Florida shows how quickly it is reshaping American politics.
Last August, Americans were almost perfectly divided about whether they would welcome a data center near their homes. Forty-three percent supported one. Forty-two percent opposed it.
By this spring, support had collapsed to 21% while opposition had surged to 71%.
In less than a year, an issue most Americans barely thought about moved from a statistical tie to a 50-point opposition advantage. In a country where public opinion on guns, abortion, immigration and nearly every other major issue appears welded into place, approximately seven out of 10 Americans now say they do not want an AI data center built near where they live.
David Wallace-Wells and Robinson Meyer recently explored the reversal in a New York Times Opinion conversation provocatively titled, “Is This the Fastest Opinion Shift in American Politics?” The question is not hyperbole. Political attitudes can move when a new president takes office, the economy collapses or a war begins. They rarely move this far, this quickly, on an issue that did not exist in most voters’ political vocabulary a year ago.
Data centers have gone from obscure background infrastructure to one of the most unpopular forms of local development in America.
The AI industry did not see this coming. Neither did the politicians who fell in love with announcing multibillion-dollar projects. They were so focused on winning the AI race that they failed to notice they were losing the public.
The Fastest Backlash in American Politics?
The polling comes from Heatmap, the energy and climate publication founded by Meyer. Heatmap surveys a few thousand Americans several times a year and asks whether they would support or oppose a data center being built near where they live.
The question is important. It does not prove that 71% of Americans oppose all data centers, AI development or national investment in AI infrastructure. There is a strong local component to the response. NIMBY may be a loaded term, but some people who oppose a data center down the road might be perfectly comfortable with one built somewhere else.
The coalition is also not ideologically uniform. Water and electricity costs unite opponents across political parties and generations. Environmental concerns resonate more heavily with Democrats and younger voters. Older residents and conservatives may be more focused on noise, property values, utility bills, land use and the ability of local communities to control their own development. Younger Americans are more likely to connect the facility with their broader distrust of AI and anxiety about employment.
Heatmap’s research found that water is the leading stated concern, followed by electricity. Yet when an actual project is canceled, noise is often the issue that stops it. The national debate may be about AI, climate and billionaire power, but the deciding moment at a local zoning meeting can be a homeowner asking what a building full of cooling equipment will sound like at 2 a.m.
Americans also understand the potential benefits. They know data centers provide the infrastructure behind streaming and other digital services. They recognize the possibility of construction employment and additional local tax revenue. Their reaction becomes more negative when the project is described as supporting AI.
That distinction should concern the industry. The public is not simply rejecting buildings filled with computers. It is reacting to the scale, purpose and political economy of the AI buildout.
The View From the Other Side of the Fence
Omaira Garcia did not learn that one of the most important AI projects in the world would be built beside her small Texas ranch through a town hall, environmental hearing or notice in her mailbox.
She learned about it when the construction dust started blowing across her property.
Garcia, an Air Force veteran and mother of two, lives outside Abilene beside the massive campus that became the first site associated with OpenAI’s Stargate initiative. Construction began before Stargate was publicly unveiled as a $500 billion joint venture among OpenAI, SoftBank and Oracle. Today, the stacks of a natural-gas power plant built to serve the campus rise approximately 500 yards from her home.
Garcia tried to sell. According to a WIRED and Floodlight investigation, she did not receive a single offer. A realtor suggested that she convert the home into an Airbnb for Stargate workers, as if losing the ability to comfortably occupy or sell your property were merely an entrepreneurial opportunity waiting to be discovered.
“We weren’t given any time to understand what this impact was going to be on us,” Garcia said. “We’re trapped here.”
That is the AI infrastructure boom seen from the other side of the fence.
From Silicon Valley, Washington and Wall Street, Stargate represents American ambition. It is compute capacity, national security, technological leadership and a chance to beat China. From Garcia’s kitchen window, it is an industrial power plant beside a home she can no longer sell.
Both perspectives are real. The problem is that only one of them appears to have mattered when the important decisions were made.
Texas Chose Its Winners
Texas did not accidentally become ground zero for the AI data center boom. State leaders wanted it, recruited it and created an environment in which hyperscale projects could be built with fewer obstacles.
Governor Greg Abbott called the industry a “gold rush.” Texas offered abundant land, natural gas, accommodating regulation and political leadership eager to announce enormous investments. The industry asked for speed, tax concessions, confidentiality, dedicated power and permitting flexibility. It largely received what it wanted.
The people living near the projects were treated too often as another development issue to be managed after the agreements were reached.
Texas reportedly has approximately 300 operating data centers and another 200 in development. The Abilene Stargate campus occupies roughly 1,100 acres and includes a 360-megawatt onsite gas plant. Including Stargate, at least 15 natural-gas plants tied to data centers are planned across the state.
Permits reviewed by Floodlight for nine of those plants could allow more than 130 million tons of greenhouse-gas emissions annually. Actual emissions may be substantially lower. Permitted capacity is not pollution already entering the air, and it would be inaccurate to suggest every generator will run continuously.
The scale nevertheless raises serious questions about the processes used to approve these facilities.
Stargate’s developers initially obtained minor permits covering 10 turbines and 62 backup diesel generators. Texas uses permits by rule and standard permits to streamline approvals for familiar or comparatively limited sources, including portable, emergency and standby generators. These processes do not require the environmental studies, public notice and comment periods associated with major-source permits.
Crusoe, Stargate’s developer, says the turbines are intended for backup power. It also points to tangible economic benefits, including funding for roads, school expansion and new fire trucks. Those benefits deserve recognition. So does the fact that 10 turbines and 62 diesel generators bear little resemblance to what most people imagine when they hear “backup power.”
The governance problem does not depend on proving that every permit is illegal. Power infrastructure of extraordinary cumulative scale is being assembled through mechanisms that can leave communities with little warning and even less influence.
Texas did not merely fail to anticipate that result. Its government made a choice about whose urgency, property and economic interests would receive priority.
Building the Shadow Grid
The power problem confronting the AI industry is legitimate. Data centers are being proposed faster than utilities can build generation and transmission capacity. Grid interconnections can take years. Major Texas transmission projects will not arrive until the 2030s. The AI race is being fought in months, not utility-planning decades.
The hyperscalers’ answer is to build their own power.
More than 20 gigawatts of behind-the-meter generation for Texas data centers was announced during 2024 and 2025, with approximately another 10 gigawatts announced during the first four months of 2026, according to Reuters. ERCOT is tracking more than 438 gigawatts of large-load requests, nearly 89% associated with data centers.
Many of those requests are speculative, duplicative or unlikely to become operating projects. Even after discounting the queue heavily, the scale is extraordinary.
Researchers have given the result an appropriate name: the shadow grid.
Unable or unwilling to wait for the public grid, hyperscalers are constructing private generating systems capable of powering cities. This may bring data centers online faster, but it also means America is creating a parallel energy system organized around the capital schedules of technology companies rather than a coherent public infrastructure strategy.
Government capture is beginning to appear in concrete, turbines, transmission lines and gas pipelines.
America needs more power. It needs generation, transmission, transformers, storage and a much faster process for connecting all of it. Natural gas will almost certainly play a role while nuclear, geothermal, renewable energy and storage expand. Pretending the demand can be wished away is not an energy policy.
But bypassing a broken grid is not the same as fixing it. “Temporary” bridge generation can become fossil-fuel infrastructure operating for decades. Private power plants may serve the hyperscalers while doing little to modernize the system everyone else depends upon.
“We Must Beat China” Is Not a Permit
The AI industry has built an extraordinarily effective political argument. Artificial intelligence is essential to economic competitiveness and national security. China is investing aggressively. Delaying American infrastructure could mean surrendering technological leadership.
Much of that is true. It does not settle every question about how, where and at whose expense this infrastructure should be built.
“We must beat China” cannot become a permission slip for every subsidy, permitting shortcut or transfer of costs. National importance does not eliminate the need to notify property owners, protect residential ratepayers, account honestly for water use or determine who pays for the generation and transmission capacity required by private facilities.
The largest technology companies in history are not impoverished pioneers asking communities to help them get started. If an AI data center requires dedicated generation, substations, water infrastructure, transmission upgrades and environmental mitigation, the presumption should be that the company pays for them.
The burden should not quietly migrate onto the utility bills, tax rolls and property values of people who had no meaningful voice in approving the project.
The deeper anger, however, extends beyond any single electric bill. Many Americans believe a small collection of immensely wealthy technology leaders is writing the country’s future without their consent. They cannot enter OpenAI’s boardroom. They cannot negotiate with Microsoft, Google, Meta or Amazon. They may believe Washington listens more carefully to Silicon Valley executives than to ordinary citizens.
They can still attend a zoning meeting.
The local data center has become the physical point at which people can grab hold of an AI revolution they otherwise feel powerless to influence. They can organize their neighbors, confront their elected officials and stop a project from being built down the road.
That may explain why the opposition feels more intense than a conventional land-use dispute. Residents are not only fighting a building. They are asserting some measure of control over a technological future they did not request and do not trust the billionaire class to design in their interest.
The Backlash Comes to Florida
Here in Florida, the shift is already arriving through our television screens.
Byron Donalds, the Trump-endorsed Republican candidate for governor, is running a major advertisement centered on protecting Floridians from AI data center costs.
“You don’t want to pay for AI data centers, and when I’m governor, you won’t,” Donalds says. He promises that developers that fail to protect Florida’s residents and their pocketbooks will not be permitted to build.
Consider what that means. This is not a progressive environmental group protesting outside a planning meeting. It is a leading Republican candidate deciding that data centers are sufficiently unpopular with Republican voters to anchor a statewide television advertisement.
Donalds’ Republican opponent, former Florida House Speaker Paul Renner, has gone further by advocating a moratorium on new hyperscale facilities. We do not need to choose between their policies to see what the competition tells us. Republican candidates believe there are votes to be won by standing up to the AI infrastructure industry.
The campaign-finance angle of this story is too delicious to skip.
The Miami Herald reports that American Mission, a pro-AI political group, spent at least $500,000 on approximately 1,700 television spots promoting Donalds earlier this year. American Mission is funded by Leading the Future, whose financial backers include Marc Andreessen, Ben Horowitz, Andreessen Horowitz, OpenAI President Greg Brockman and Brockman’s wife, Anna.
These were independent expenditures and could not legally be coordinated with the Donalds campaign. There is no basis for claiming that the spending purchased his position.
The irony is sufficient. AI industry money helped promote a candidate who now feels compelled to assure voters that he will protect them from the costs of AI infrastructure.
Donalds is not creating the backlash. He is reading the polling.
Whose Money Machines?
President Trump has called data centers “money machines” and “LIQUID GOLD.” His point is understandable. They represent enormous capital investment, and states that turn them away may watch those projects move elsewhere.
But calling them money machines raises the defining populist question: Whose money machines are they?
If hyperscalers receive tax concessions, developers receive contracts, investors capture the returns and politicians receive industry-funded advertising, what do nearby residents receive? Lower electricity rates, durable employment and better infrastructure? Or noise, pollution, water pressure, grid expenses and homes that become harder to sell?
The answer will differ by project. Loudoun County, Virginia, expects data centers to provide an enormous portion of its local tax revenue. Some communities will welcome the investment. Some developers will pay for the required infrastructure, use cleaner power and locate projects where their effects can be responsibly managed.
“Data center” should not become a synonym for “bad project.”
The industry nevertheless invited this backlash by behaving as though every project is entitled to approval because AI is important. The 71% opposition number shows how badly it misread the public.
Make the Investment Serve the Country
The answer does not need to be an indefinite national moratorium or a retreat from AI. Robinson Meyer offers a more productive direction: Capture some of the extraordinary investment flowing into data centers and use it to build the grid America needs anyway.
That might involve an excise tax, a dedicated infrastructure fund, direct build-and-pay requirements or enforceable development agreements. The exact mechanism requires serious analysis. The governing principle does not.
Data center investment should expand public capacity rather than merely create private islands of power.
Projects should face transparent permitting, meaningful public notice and defensible environmental review. Residential customers and small businesses should not subsidize hyperscale loads. When developers receive tax concessions, the promised benefits should be measurable and enforceable. Companies should pay proportionately for the generation, substations, transmission, water systems and mitigation their facilities require.
I have compared the AI buildout to electrification because it may ultimately prove just as consequential. But electrification eventually produced a public compact. Utilities received protected positions, access to public rights-of-way and the ability to recover investments from customers. In exchange, they accepted regulation, service obligations, rate scrutiny and public accountability.
AI companies increasingly want utility-scale access to land, power, water and government accommodation while retaining the independence and economics of private technology companies. They want the privileges of essential infrastructure without accepting comparable obligations to the public.
That bargain is creating the backlash.
The AI industry has won extraordinary access to presidents, governors, legislators and economic-development agencies. It assumed that elite agreement amounted to public consent. It does not.
The industry may continue winning permits while losing the people who pay the electric bills and live beside the projects. That is not merely a public-relations problem. Without public legitimacy, every campus becomes a political fight, every utility investment becomes suspect and every delay makes America’s infrastructure challenge harder.
We should not kill the investment. We should make the investment serve the country instead of merely serving the companies making it.
