Everyone in AI talks about the compute race. More GPUs, more data centers, more power. The assumption is that the constraints are capital and supply chain. Get the money, get the chips, build the buildings.
That assumption is wrong. The constraint is increasingly the people who live where the buildings are supposed to go.
According to a comprehensive report by Data Center Watch, $18 billion in data center projects have been blocked outright across the United States. Another $46 billion have been significantly delayed. The combined total: $64 billion in AI infrastructure stalled or killed by grassroots opposition.
This is not a fringe movement. It is 142 activist groups across 24 states. And here is the number that should make every AI infrastructure investor pay attention: the opposition is 55% Republican and 45% Democrat. This is not partisan. This is local.
The numbers are staggering
A $14 billion project in Arizona: withdrawn entirely. The company walked away. Fourteen billion dollars, gone because the community said no.
$24.7 billion in data center projects in Virginia, from QTS and Compass among others: delayed. Virginia is the data center capital of the world. It hosts more data center capacity than any other state. And even there, the resistance is grinding projects to a halt.
Amazon’s $6 billion King George, Virginia project: delayed. This is Amazon. They have functionally unlimited capital, the best lobbyists money can buy, and deep relationships with state and local government. They still could not push through on schedule.
Virginia alone has 42 activist groups fighting data center expansion. Forty-two. In one state.
It is not just protests. It is elections.
This is where it gets serious for the industry. The opposition is not just showing up at town halls. They are winning elections.
In Oregon, recall elections removed pro-data center officials from office. Elected officials who approved data center projects were recalled by their own constituents.
In Warrenton, Virginia, every single town council member who supported Amazon’s data center project lost re-election. Not some of them. All of them. The voters replaced the entire council.
That sends a very clear message to every local official in every town where a data center is proposed: vote for this project and you may lose your job. That is an incentive structure that no amount of corporate lobbying can easily overcome.
Why communities fight back
The opposition is not irrational. Data centers consume enormous amounts of electricity and water. A single large facility can use as much power as a small city. In regions already dealing with grid strain, adding a multi-gigawatt data center means either building new power infrastructure, which takes years, or diverting capacity from existing users.
Water is the other pressure point. Data centers require massive amounts of water for cooling. In drought-prone areas, that puts them in direct competition with agricultural and residential water supplies. When residents learn that a proposed facility will consume millions of gallons per day from their aquifer, the abstract promise of “economic development” becomes a lot less persuasive.
Then there is the jobs question. Data centers are capital-intensive, not labor-intensive. A $7 billion facility might employ 50 to 100 people during operations. Compare that to a manufacturing plant of similar investment that might employ thousands. The economic argument that data centers bring jobs does not hold up well under scrutiny, and local communities have gotten very good at scrutinizing it.
NPR reported on the growing tension between federal AI ambitions and local resistance, noting that even in politically favorable environments, communities are pushing back when the physical costs land on their doorstep.
The consent problem
The AI industry has treated infrastructure as an engineering and capital allocation problem. Build more, faster, bigger. Raise the money, secure the chips, break ground. The playbook assumes that if you have the capital and the technology, the rest follows.
But you cannot build a $7 billion data center if the town votes you out. You cannot break ground if the county board denies your permit. You cannot operate if the state legislature passes new restrictions on water usage or power consumption in response to constituent pressure.
The Decoder reported that the total value of projects facing resistance is approaching $98 billion when including projects in early planning stages that are likely to face opposition. The $64 billion figure counts only projects where opposition has already materialized and caused measurable impact.
This is a consent problem. And consent problems do not scale the way capital problems do. You cannot solve them by spending more money. In fact, spending more money often makes them worse, because it signals to communities that the project is large enough to disrupt their lives in ways they have not yet imagined.
What the industry is not pricing in
Every AI company’s growth model assumes that compute capacity will grow to meet demand. The valuations, the revenue projections, the product roadmaps, they all assume the data centers will get built.
The $64 billion in blocked and delayed projects suggests that assumption needs a discount. Not a catastrophic one, but a real one. Projects will take longer. They will cost more. Some will not happen at all. The timeline for compute buildout is going to be slower than the industry is currently pricing.
This does not mean AI growth stops. It means the physical layer has a cost that is not on any balance sheet. The permitting delays, the legal battles, the electoral consequences, these are real costs that compound over time. A project delayed by two years is not just late. It is two years of lost revenue and two years of compute capacity that the rest of the industry was counting on.
The strategic risk
For companies betting on massive compute expansion, this is worth watching carefully. The opposition is organized, bipartisan, well-funded by local standards, and electorally effective. It is growing, not shrinking. And it has a structural advantage: data centers need specific locations with access to power, water, and fiber. The communities in those locations have leverage that is difficult to route around.
The AI industry talks constantly about scaling laws, training costs, and chip supply. It should be talking just as seriously about whether the towns where it wants to build will let it.
Sixty-four billion dollars says they might not.
Sources: Data Center Watch, Time, TechRepublic, NPR, The Decoder