The Enemy Makes the Village

A collaboration between Lewis McLain & AI

What the fight over data centers reveals about the fastest, cheapest way to make people love each other


There is a moment, early in almost every land-use fight, when a stranger becomes a friend.

It happens in a folding chair. Someone leans over during the third hour of a zoning hearing and says something like can you believe this guy, and the person next to them — a person they have never met, whose politics they’d have found repellent at a dinner party — says no. No I cannot. And something clicks into place that would have taken a year of shared hobbies to build.

Social psychologists have a name for what just happened, and a body of evidence about how well it works. In 2006, Jennifer Bosson, Amber Johnson, Kate Niederhoffer and William Swann published a paper with the deceptively cheerful title Interpersonal chemistry through negativity. Across two surveys and an experiment, they found that discovering you share a negative attitude about a third party predicts liking a stranger more strongly than discovering you share a positive one — at least where the attitudes are not already strongly held. Shared dislike is a faster solvent than shared affection. It draws the in-group boundary in one stroke, flatters both parties, and — because most people are careful about voicing contempt — signals a level of candor that ordinary agreement doesn’t.

The paper’s most unsettling finding is the last one: participants seemed largely unaware the effect was operating on them.

I want to trace that effect through a live case — the American fight over AI data centers — because it is unusually well documented, unusually fast-moving, and because it has now run far enough to show you the whole arc. Not the arc from apathy to activism. The arc from activism to something else.

A note on method: the specific incidents below are real and sourced. The folding-chair scene above, and the composite passages marked as such, are illustrative — assembled from the recurring texture of these fights rather than any single meeting.


Stage one: the lawn sign

The mildest form of the bond is a sign.

Drive through Loudoun County, or Mason County, Kentucky, or the ring roads outside Atlanta, and you’ll see them: hand-lettered, coroplast, a mix of fonts that betrays a dozen different households ordering from a dozen different print shops. NO DATA CENTER. The Soufan Center, which has spent a year monitoring anti-data-center discourse, notes almost as an aside that you now see these signs from the highway across large stretches of the country.

A sign is a low-cost, high-visibility declaration of the boundary. It says: there is a we, and you can tell who’s in it. It is also, crucially, an invitation. Two neighbors who have exchanged nothing but nods for eleven years now have a subject.

The grievances underneath are not manufactured. Water consumption and contamination, electricity costs shifted onto ratepayers, grid strain, noise, the loss of farmland, eminent domain for transmission lines that serve the facility rather than the town. A Gallup poll in May 2026 found roughly seven in ten Americans oppose a data center being built in their area — higher opposition than for a nuclear plant. Data Center Watch counted at least 75 projects worth around $130 billion blocked or delayed by local opposition in the first quarter of 2026 alone.

This is what functioning civic life looks like. People noticing an imposition and organizing against it. Nothing in what follows should be read as a claim that the grievance is fake.

The claim is narrower, and stranger: the grievance is not what does most of the bonding work. The enemy does.


Stage two: the twenty-seven-hour meeting

On the morning of December 12, 2023, the Prince William County Board of Supervisors in Virginia opened a hearing on the Digital Gateway — a 2,100-acre data center campus proposed for the corridor between the Manassas National Battlefield and Conway Robinson State Forest.

The meeting ran more than twenty-seven hours, ending in the late morning of December 13. More than two hundred people spoke, in person and online. The final vote was 4–3, with one abstention, in favor.

Twenty-seven hours is not a deliberative procedure. It is an initiation rite.

Consider what it does to the people in the room. Sleep deprivation. Shared physical discomfort. A visible out-group at the front of the room with the power to ignore you. Repeated public testimony in which each speaker restates the collective position in their own words — a mechanism social psychologists have studied since Stoner’s 1961 “risky shift” finding, generalized by Moscovici and Zavalloni in 1969 under the heading of group polarization: when like-minded people deliberate together, they don’t average toward the middle. They move further out. Cass Sunstein calls the specific case of self-selected like-minded groups enclave deliberation, and the pattern is robust — the enclave exits more extreme than any individual entered.

Twenty-seven hours of enclave deliberation, capped by a 4–3 loss, is close to an optimal recipe for converting a policy disagreement into an identity.

The Digital Gateway rezoning was later voided in Circuit Court, upheld unanimously by the Virginia Court of Appeals in March 2026, and the project finally died in July 2026 when the last developer withdrew its appeal. The opponents won. But by then the coalition they’d built had long outgrown the parcel it was built to defend.


Stage three: the enemy of my enemy sits down next to me

Here is where the bonding effect starts producing things that ought to be impossible.

The anti-data-center coalition in 2026 includes Democratic Socialists of America chapters and Steve Bannon. It includes Extinction Rebellion and Marjorie Taylor Greene. An activist group called Humans First, which organized a national day of protest, was reportedly seeded by AI-safety-motivated effective altruists and then handed off to America First conservatives — a lineage that makes no sense on any conventional political axis.

A DSA organizer in Corpus Christi, describing on a national call what she was hearing from self-described “MAGA mommies” in her area, relayed it about as plainly as the psychology allows:

“I’m not scared of the socialists. I have more in common with them than I do with any of the people making these decisions.”

That is a socialist’s account of what a conservative told her, which makes it a paraphrase filtered through an ally who was recently an opponent — and that fact is itself the evidence. The sentence had to travel across the old boundary to reach us.

Read that again as a technical statement. She is describing a re-categorization — Gaertner and Dovidio’s common in-group identity model, in which two previously antagonistic groups dissolve their boundary by discovering a superordinate one. It is one of the most reliable interventions in the prejudice-reduction literature. It is, in most contexts, a good thing.

The catch is that the superordinate identity here is defined negatively. The new “we” is not people who want X. It is people who are not the people making these decisions. And a group defined by what it is against has no internal mechanism for deciding when it has gotten what it wants.

Muzafer Sherif demonstrated both halves of this at Robbers Cave in 1954. Manufacture a competition between two groups of boys and you get instant, ferocious in-group cohesion — flags, names, raids, contempt. Then he showed you could dissolve it, but only with superordinate goals: tasks neither group could accomplish alone. Cooperation, not contact, was the solvent.

The data center coalition has the first half. It has no supply of the second.


Stage four: the moral upgrade

Composite scene, drawn from the recurring pattern:

By month eight the Facebook group has four thousand members and a moderator who did not exist as a public person a year ago. The early posts were about setback distances and decibel readings. The current posts are about what kind of person votes for this. Somebody has started making graphics. Somebody has found the county supervisor’s business address. The word “corruption” enters the vocabulary and does not leave, and it is not always wrong, and it is never again subject to evidentiary standards, because the group has already decided.

This is the transition that matters, and it is visible in the real record.

In May 2026, the progressive outlet More Perfect Union took Representative Alexandria Ocasio-Cortez to Morgan County, Georgia, to meet residents who said a Meta data center’s construction had fouled their water. Days later she held up two jars of brown water while questioning an EPA water administrator at a House oversight hearing. The video went viral.

That jar is a masterpiece of moral communication, and I mean that without cynicism — the residents’ complaint appears to be real and it deserved the amplification. But notice what the jar does that a groundwater table cannot. It converts a regulatory dispute into a moral violation, and moral violations are the single most transmissible category of content in a networked information environment. Rathje, Van Bavel and van der Linden, analyzing 2.7 million posts from news outlets and members of Congress, found that content about the political out-group was by far the strongest predictor of sharing — stronger than emotional language, stronger than anything about one’s own side. The platforms are not neutral carriers of this material. They are selection pressure on it.

The Soufan Center’s own monitoring caught the compounding directly. Their dataset shows a baseline of anti-data-center discourse with a sharp surge beginning in April 2026 that then stayed elevated, and they describe the mechanism precisely: each new incident now “generates much greater secondary amplification than it did earlier in the year — meaning once a story breaks, they compound onto each other, feeding back into the information environment and causing secondhand outrage.”

Secondhand outrage. That is the load-bearing phrase. At this stage most of the anger in the system no longer belongs to anyone whose water it was.


Stage five: the official becomes the enemy

Saline Township, Michigan. Population about 2,300. Farmland. In 2025 a developer proposed a $16 billion data center campus tied to the Oracle–OpenAI Stargate build-out.

The township board denied the zoning change. The developer sued. Facing legal exposure that residents were told worked out to something like $29,000 per household over a decade, the board settled. Recall petitions went up against three members.

On May 13, 2026, at a packed board meeting, Treasurer Jennifer Zink announced her resignation in tears, citing death threats. Among them, she said, were wishes that she die of Lyme disease. The township clerk described similar harassment. Zink’s resignation took effect May 29.

A study published in April 2026 by West Point’s Combating Terrorism Center identified the shift underneath this: grievance is migrating from tech executives to local officials, because data centers are what makes AI physical, and they concentrate its costs at the municipal level. The CEO is an abstraction three time zones away. The county treasurer shops at your grocery store.

Ballotpedia’s mid-year tally counted 58 local officials across seven states facing data-center-related recall petitions — roughly 17 percent of all recall efforts in 2026, a record year.

That is democracy working, and I want to be clear that recalls and elections are exactly the right instrument. But a woman crying at a folding table in a room of two thousand three hundred neighbors, resigning an unpaid-ish rural office because strangers wished disease on her, is not a policy outcome. It is what happens when a coalition that formed around an enemy runs out of enemy and finds a nearer one.


Stage six: the far end

The last stage is small in volume and impossible to unsee.

On April 10, 2026, prosecutors say Daniel Moreno-Gama threw a Molotov cocktail at the San Francisco home of OpenAI chief executive Sam Altman, then traveled to the company’s headquarters and threatened to burn it down and kill the people inside. Four days earlier, at around 12:45 a.m., Indianapolis City-County Councilman Ron Gibson found thirteen rounds had been fired into the front door of his home — he and his eight-year-old son were unhurt — after he publicly backed a data center in his district. A handwritten note reading “No Data Centers” was left on the doorstep. In June, a disrupted plot to kill senior government officials at a White House event was, per the charging documents, motivated by a grab-bag of grievances that included “water-guzzling data centers.” In July, Extinction Rebellion activists in Amsterdam threw balloons of hydrogen peroxide and acetic acid at a hyperscale construction site expected to house Microsoft.

The Soufan Center is careful, and their caveat deserves full weight: explicit, credible threat rhetoric remains a small fraction of anti-data-center content, most of which is ordinary, lawful, and often well-founded civic opposition. Most direct threats target infrastructure rather than people. Much of what looks violent online is posturing — fire GIFs, guillotine emojis, deliberate misspellings to dodge moderation, the verb “luigi-ed” deployed in a register that is clearly rhetorical.

But note the shape of the ladder. Lawn sign. Hearing. Coalition. Moral frame. Recall. Threat. At no point does anyone cross a line. Each rung is a small, locally reasonable step from the one below it, taken in the company of people who have become your friends because you took the previous one together.

That is the mechanism. Not radicalization as a leap — radicalization as a series of loyalty payments.


What the bond is actually made of

Swann, one of the co-authors of the negativity-bonding paper, later developed a theory called identity fusion: the state in which the boundary between personal self and group self becomes porous, so that a threat to the group is processed as a threat to the body. Fused individuals, in his research with Jetten, Gómez, Whitehouse and Bastian, show markedly elevated willingness to fight and to sacrifice for the group.

The path from sharing a negative attitude with a stranger to fusion is not long, and every stage above is a step along it. Which is the uncomfortable point of this essay:

The properties that make common-enemy bonding so good at building community are the same properties that make it bad at stopping.

It is fast, because contempt is high-information. It is cheap, because it requires no shared positive vision — only a shared negation. It is durable, because the enemy can always be found to be doing something new. It is portable, because when this data center is defeated the coalition does not disband; it looks for the next imposition, and it will find one, because there is always one. And it is self-concealing — Bosson’s participants had no idea it was happening to them.

None of this makes the coalition wrong. On the merits, a great deal of the anti-data-center case looks strong: ratepayers really are being asked to subsidize private compute, water really is being drawn from stressed aquifers, and the NAACP — represented by Earthjustice and the Southern Environmental Law Center — filed suit in April 2026 over twenty-seven methane turbines running without permits in Southaven, Mississippi, powering xAI’s Colossus 2 next to predominantly Black South Memphis neighborhoods carrying asthma burdens far above state and national norms. You do not need a psychological account of group dynamics to explain why people are angry about that. You need an air permit.

But being right is not protection. Sherif’s boys at Robbers Cave each had, from the inside, entirely adequate reasons to despise the other cabin.


What holds

If the diagnosis is right, the interventions are unromantic and mostly structural:

Keep the enemy institutional. A coalition aimed at a permitting process behaves differently from one aimed at a named treasurer. The moment the target acquires a face and an address, the escalation ceiling rises.

Insist on a positive ask. “No data center” has no completion condition. “Facility-funded grid upgrades, metered water disclosure, a 55-decibel property-line cap, and a decommissioning bond” does. Concrete demands can be met, and a coalition that can be satisfied is a coalition that can stand down.

Manufacture superordinate goals. Sherif’s actual finding, the one that gets forgotten in favor of the flashier half: cohesion built by conflict dissolves only under cooperation neither side can do alone. Genuine co-design — communities with real authority over siting, water and cost allocation, not consultation theater — is the only known solvent. It is also, notably, what several jurisdictions have started reaching for: New York’s statewide hyperscale moratorium, the Netherlands’ two-site restriction, Amsterdam’s freeze on new construction until 2030. Whatever else those are, they are attempts to put the decision somewhere a person can reach it.

Watch for the moment the meeting becomes the point. When attendance stays high after the vote is won — when the Facebook group’s engagement metrics are healthier than its policy agenda — the bond has decoupled from the grievance and is now sustaining itself.

The folding chair is not a trap. Some of the best civic energy in the country right now is being generated in exactly those rooms, and the neighbor you meet there may well become a real friend.

Just be honest about the introduction. You did not meet over a shared love of anything.


Sources

Reporting and data

Research

  • Bosson, J. K., Johnson, A. B., Niederhoffer, K., & Swann, W. B. (2006). Interpersonal chemistry through negativity: Bonding by sharing negative attitudes about others. Personal Relationships, 13, 135–150.
  • Rathje, S., Van Bavel, J. J., & van der Linden, S. (2021). Out-group animosity drives engagement on social media. PNAS, 118(26).
  • Swann, W. B., Jetten, J., Gómez, Á., Whitehouse, H., & Bastian, B. (2012). When group membership gets personal: A theory of identity fusion. Psychological Review, 119(3), 441–456.
  • Sherif, M. et al. (1961). Intergroup Conflict and Cooperation: The Robbers Cave Experiment.
  • Gaertner, S. L., & Dovidio, J. F. (2000). Reducing Intergroup Bias: The Common Ingroup Identity Model.
  • Moscovici, S., & Zavalloni, M. (1969). The group as a polarizer of attitudes. Journal of Personality and Social Psychology, 12(2), 125–135.
  • Sunstein, C. R. (1999/2002). The Law of Group Polarization. Journal of Political Philosophy, 10(2), 175–195.

Composite scenes are marked as such in the text. All named incidents, figures and quotations are drawn from the sources above.

Leave a Reply