The Data Center Fight Is a Proxy War
Why the fight over server farms isn’t really about server farms — and what it tells us about where society is splitting.
The poll question Gallup should have asked
Earlier this year, Gallup found that 71% of Americans oppose building a data center in their own area — more opposition than Americans have ever expressed toward a local nuclear power plant. That’s a striking number on its own. But it’s the reasons people give that are more interesting than the number itself. Half of opponents cite excessive resource use — water, energy. Others mention noise, traffic, and property values. On its face, this looks like a textbook NIMBY fight: a community weighing costs against benefits and deciding the math doesn’t work.
So I ran my own extremely unscientific poll in Maryland and Georgia — about four dozen people across both states, which makes this more anecdotal than a real survey. I asked people who were already opposed to new data centers in their area a different question: if a proposed data center produced its own power, made no noise, had no external lighting, and used no water — would you allow it?
About 90% said no.
That result should worry anyone who thinks this fight is really about water tables and decibel levels. If it were, stripping out those externalities should have flipped most of the opposition. It didn’t. Whatever is driving this, it isn’t NIMBY in the classic sense.
And my little sample isn’t alone. A University of Virginia survey of more than 3,200 Americans asked data center non-supporters a blunter version of the same question: how much money would it take to change your mind? Of the 2,642 respondents who didn’t support a local data center, the median asking price to change their mind was $10,000 but the study also found a group it had to set aside entirely: people who said nothing could buy their support, at any price.
Diffusion theory, and the case for something closer to zero
Everett Rogers’ Diffusion of Innovations (1962) explains adoption rates using five variables: relative advantage, compatibility, complexity, trialability, and observability. Rogers treated these as continuous dials — an innovation is more or less compatible with existing values, more or less complex to understand. The theory was built to explain why hybrid corn spread through Iowa at the rate it did, not to explain outright, organized refusal.
But a body of research that runs alongside Rogers has spent decades asking what happens when compatibility doesn’t just dip, it goes negative — when an innovation isn’t merely unfamiliar but experienced as a violation of who you are. Thomas Robertson’s typology (1967) of continuous, dynamically continuous, and discontinuous innovation was an early attempt to formalize this: a discontinuous innovation doesn’t ask you to adjust your behavior; it asks you to change who you are. Geoffrey Moore’s Crossing the Chasm is really Rogers' model applied to exactly this failure mode — the reason so many technically excellent products die between early adopters and the early majority is that the early majority requires Rogers’ “compatibility,” which the product lacks.
The sharpest tool for this, though, is Ram and Sheth’s Innovation Resistance Theory (1989), which reframes compatibility and complexity not as metrics of adoption rate but as sources of active resistance that can block adoption outright. Innovation Resistance Theory (IRT) splits resistance into functional barriers (usage, value, risk — Rogers’ “complexity”) and psychological barriers (tradition, image — Rogers’ “compatibility”), and the psychological ones are the interesting case: they don’t respond to better information, because the objection was never about information.
There’s a vivid real-world precedent for what “negative compatibility” looks like: the deaf rights movement’s organized resistance to cochlear implants, which framed the device not as a hearing aid but as a threat to Deaf identity and language. That’s not a complexity objection or a cost-benefit calculation. It’s a rejection of what the innovation means.
Which brings us to data centers
The data center backlash of 2026 has been extraordinary by any measure. Opponents blocked or delayed roughly $130 billion in projects in the first quarter alone — matching the entire prior year in three months. The number of organized opposition groups more than doubled, to over 800 across 49 states. And unusually for anything in American politics right now, it’s genuinely bipartisan: roughly three-quarters of Democrats and nearly two-thirds of Republicans oppose local data center construction, with some polling finding hardline conservatives closer to progressive Democrats on this than to moderate members of their own party.
Analysts increasingly describe this as a proxy fight rather than a land-use dispute. Brookings, citing a Vox headline that summed it up — “Americans don’t know how to fight AI. So they’re fighting data centers” — argues that the mega-centers have become the symbol of citizen frustration with AI itself: the most visible manifestation of the technology, and therefore the proxy for public feedback on it. A data center is the first node in the entire AI/internet stack you can actually show up and object to.
My informal polling suggests that proxy has two separate narratives, not one. About half of the objections I heard were economic-anxiety flavored — AI taking jobs, wages, relevance. The other half were something else entirely, and people were not shy about naming it. The direct quote I kept hearing, verbatim, in both states: evil billionaires who just want to enslave us.
Those are different species of objection, and both deserve to be taken at face value. “AI will take my job” is a relative-advantage and risk calculation — rational-actor territory, and a calculation nobody can currently claim to address with confidence, including the optimists. “Evil billionaires” is not a cost-benefit statement at all. The phrasing is theatrical; the question underneath it is not. It’s a claim about who, if anyone, gets to decide the shape of the (<-- a simple word I will revisit below) future, and by what right — which is roughly the oldest question in democratic politics. People asking that question about AI are not confused about the technology. They’re asking about consent. That’s textbook compatibility failure — a collision with a deeply held value (self-determination, suspicion of concentrated power) rather than a misjudgment of the technology’s usefulness. Time magazine has started calling this a brewing populist backlash, one explicitly framed around “the gap between the promises of the tech oligarchy and the reality of Main Street” — language that doesn’t map onto the usual left-right axis at all.
Why now, though? The cloud had a good run of thirty years.
Here’s where I think the deeper cultural story lives. For most of the internet era, “the cloud” did some interesting ideological work that few realized — not through conspiracy, but through convenience. It rendered something intensely physical (land, concrete, transformers, millions of gallons of water) into something that felt weightless.
That arrangement worked as long as the physical plant stayed out of sight. It no longer does. Hyperscale buildouts are pushing into exurbs and even suburbs, and for the first time, a huge number of people are encountering physical proof that the “virtual” world they’ve been living half their lives in was never virtual at all. That’s not a small correction to a mental model. For people who had made peace with the internet precisely because it felt immaterial, optional, apart from their physical world, the data center is an unwelcome intrusion into their physical reality.
It turns out the “there’s no permit mechanism for an algorithm” framing I’d been assuming was wrong, or at least incomplete. Municipalities already have a working precedent for zoning out a specific piece of software: since San Francisco’s first-of-its-kind ban on facial recognition, at least 16 more cities have passed local ordinances prohibiting facial recognition and biometric surveillance within their jurisdiction, some covering private businesses as well as government use. The “temporary” data center moratoriums we are seeing around the country are, in many cases, the exact same thing, but for some reason, those passing them are being quiet about the possible real intent: increasing the base cost of building a data center can effectively become a permanent ban.
Which is exactly why there’s now a live fight over whether that authority should exist at all for AI more broadly. State legislatures in at least nine states have introduced bills limiting local governments’ ability to regulate AI. Several — in New Hampshire, Ohio, South Carolina, and Virginia — share language traced to a conservative ALEC model bill and would impose a very high legal bar on any local ordinance, rule, or fee restricting AI. Illinois went further with a bill that would preempt local AI regulation outright, explicitly overriding home-rule powers. And Georgia’s SB 104 — relevant to one of the two states I polled — pairs privacy protections with preemptive provisions barring local governments from engaging with foreign-controlled AI systems. The details vary, but the direction doesn’t: the data center zoning fight isn’t happening in a vacuum where no permitting tool exists. It’s happening while a parallel fight is underway over whether local communities get to keep the one permitting tool they’ve already built.
The pacing problem
There’s a name in tech-policy circles for why this is intensifying rather than settling down: the “pacing problem” — the observation, going back to Larry Downes’ The Laws of Disruption, that technology changes exponentially while social, economic, and legal systems change incrementally.
Historically, technological transitions — electrification, the automobile, television — had multi-decade absorption windows. A generation could gradually normalize a change before the next one arrived. If AI, like other technologies with S-curve growth rates, is compressing that window, you don’t get one generation smoothly adjusting before the next shift lands. You get people deciding, quite reasonably, that the last era they trusted is the one they’ll stand on — because the pace no longer leaves room to genuinely evaluate each new wave before the next one arrives. While some of this may just be a failure or refusal to adapt, the data strongly suggests it is a judgment that perpetual and unexamined adaptation is itself the thing worth confronting.
A prediction: society is sorting into attractor states
Earlier, I suggested the word “the” was an important distinction. I think the drift we are experiencing isn’t going to resolve into some unifying consensus. I think it’s going to resolve into cultural bands organized roughly around which technological era a person feels a cultural affinity with. I’m not using “era” as a term related to age, but to large buckets of related technology that shared the same “velocity” of disruption.
Accelerationist — full speed ahead, AI and beyond
Pre-AI — comfortable with the internet, the smartphone, the modern connected world, but drawing a hard line at AI
Pre-internet — living deliberately closer to analog life, phones optional, connectivity is minimized and localized
Pre-modern agriculture — the existing end state, already stably occupied
A note on the labels before anyone bristles at them: these are not rankings, and the accelerationist tier is not the neutral default from which the other three deviate. Every tier is a value position: a considered judgment about what a good life requires and what it can do without. The Amish did not fail to understand the automobile; they evaluated it and concluded the costs to the community outweighed the convenience. The entire argument of this piece has been that the resistance we’re watching is not a comprehension, age, or generational problem. Taking that argument seriously means treating each attractor as a coherent answer to the same hard question, not as varying degrees of falling behind.
That fourth category isn’t hypothetical. It’s the Amish, who have held a fixed technological line for nearly 300 years and remain the most credible proof that a durable, self-reproducing culture organized around a specific tech cutoff is not just possible but sustainable across generations. Neo-Luddite scholarship has explicitly cited Amish communities as the model for what a deliberate technological epoch looks like at scale. The Amish have also demonstrated the ability to live according to their culture while physically embedded within another culture.
The second and third categories are visibly forming right now, which is what makes this feel less like idle futurism and more like a trend already underway. Chad Whitacre, a longtime open-source software developer, announced on his blog in May 2026 that he was retiring from tech entirely to live offline, writing that “AI took the last of the wind out of my Open Source sails.” Quartz picked up the story and described his framing of the move as going “neo-Amish,” aiming for a life closer to 1980 than 2026. He built his career at the center of the open web, which is what makes his exit hard to dismiss — this is a considered verdict from someone who knows exactly what he’s leaving, not a rejection born of unfamiliarity. This summer, a cluster of deliberately offline events — a play about the original Luddites, a mending workshop, a conference on AI’s ties to the military — ran under the banner “Summer of Ludd,” none of it advertised online. Meanwhile, a growing number of Americans are quietly ditching smartphones altogether in pursuit of “attention, presence, and meaning.”
The important nuance, and the reason I call these attractor states and not categories, is that the edges are fuzzy. Nobody signs a charter declaring themselves Pre-AI. Most people drift toward an attractor until they reach a stable orbit around it. They do that through hundreds of small daily choices — which tools they use, who they spend time with, what they avoid — and then get held there by ordinary social reinforcement: the communities and norms around them start pulling them deeper toward the center of whatever tier they’ve drifted into. Identity does the rest. Once “which tier are you” becomes a thing people ask each other, even implicitly, the basins deepen, and the boundaries harden.
Which is also, I think, the best explanation for why the data center fight is so intense right now. Most of the country isn’t opposing data centers from a settled position. It’s opposing them from the unstable zone between two attractors — no longer entirely at ease with where technology has taken us, still working out where the line belongs. Settled attractors don’t need activist coalitions. Contested boundary zones do. The fight over server farms in Maryland, Georgia, Virginia, and forty-six other states isn’t really a fight about data centers at all. It’s the sound of a very large number of people deciding, seriously and for the first time, which side of the next line they intend to live on. And I’ve seen that exact process play out in real time when I asked these probing questions.
A conclusion and a warning
My conclusion from these conversations is that this trend can’t be ignored or legislated away. People will lose elections over this issue. The politicians attempting to limit their citizens’ ability to locally control what happens in their communities among them. Acknowledging that this isn’t a “data center” issue but a debate about what kind of world you want to live in is incredibly important to working out a detante of sorts between those attractor states.
Part of the reason I wrote this and why it is a fascinating topic is from a conversation I had with my father. I am now 57, and Dad is about to turn 80. Sixty years of smoking have put him in failing health. As that happens, conversations get more meaningful because you know your time is limited. I was sitting with my parents while they watched Yellowstone, and just flippantly, I asked them if they could have lived anywhere in the world at any time, what would that be? My Mom said, “In the house when you and your sister were babies”. My Dad said, “On a 100,000-acre ranch in Wyoming with no other human beings for a hundred miles with nothing but a horse, some coffee, and a rifle.” I knew my Dad well enough that I knew he absolutely was NOT joking. He was happiest when he was either tending the garden that provided most of our vegetables or in a treestand hunting for what provided most of our protein. I know exactly what attractor state my Dad would instantly flee to if it existed. Fortunately for me, my sister, and my Mom, he decided to swallow that deep desire and be a great and awesome Dad. Still, I have a deep unfillable wish that he could have had that for himself.
Sources and further reading
Diffusion theory and its extensions on compatibility/complexity failure
Everett Rogers, Diffusion of Innovations — the foundational five-factor model. Overview: TheoryHub
On continuous vs. discontinuous innovation and category knowledge effects: Consumer evaluation of continuous and discontinuous innovation
On Innovation Resistance Theory applied to a modern adoption context: Barriers to Digital Health Adoption in Older Adults
On radical innovation and organized cultural resistance (the cochlear implant case): Market rebels and radical innovation — McKinsey
The data center backlash, in numbers
Data Center Watch: $64 billion of projects blocked or delayed
Brookings: Data center backlash signals a fight over AI power
Harvard Gazette: Why are communities pushing back against data centers?
Heatmap News: Americans now overwhelmingly oppose new data centers near them
TechXplore: Data center fights pit social values, democracy, and capitalism against each other
National Security Data and Policy Institute: Data Center Opinion Project (the “won’t accept any price” finding)
Rainey Center: Voters are persuadable, but they want accountability (a contrasting result worth weighing against conditional-support polling)
AI anxiety, job loss, and the “tech oligarchy” framing
The cloud as a physical thing we agreed to forget
Institute of Network Cultures: Unsettling the Cloud — On Data Centers and Counter-Narratives
Where the Internet Lives: Data Centers as Cloud Infrastructure
The pacing problem
Wikipedia: Collingridge Dilemma (includes the pacing-problem lineage back to Larry Downes)
Mercatus Center: The Pacing Problem and the Future of Technology Regulation
Evidence for the “tiers” already forming
Chad Whitacre: I Am Retiring from Tech to Live Offline (primary source — the “neo-Amish” open-source developer, in his own words)
Quartz: A growing anti-tech movement is powering down in the AI era (”Summer of Ludd” and coverage of Whitacre’s announcement)
Psychology Today: Pushing Back Against Technology — The Rise of Neo-Luddism
Deseret News: What are ‘neo-Luddites’? Americans ditching their smartphones
Wikipedia: Neo-Luddism (cites Amish and Chipko movement communities as explicit models)
Giles Crouch: The Four Luddites of the AI Age (a related but distinct taxonomy of resistance types, worth reading against this piece’s framework)
Municipal algorithmic zoning, and the fight over whether it can continue
EFF: The Movement to Ban Government Use of Face Recognition (the precedent — 17+ municipal bans, best practices)
TechCrunch: Portland passes expansive city ban on facial recognition tech (a ban covering private, not just government, use)
Local Solutions Support Center: State Legislatures Attempt to Restrict Local Governments from Regulating AI (tracks Georgia SB 104 and similar preemption bills nationwide)