Ai Data Centers Could Decide The 2028 Election

I kept coming back to a number that did not feel like a tech story at all. Seventy-one percent. That is the share of registered voters, in a recent national survey, who said they would oppose an artificial intelligence data center in their own area. Nearly half said they would oppose it strongly. Only about a quarter were willing to support one nearby. Sit with that for a second. A technology that boardrooms treat as destiny is, on the ground, about as welcome as a new landfill.

That gap is why a well-known software chief executive told an investor podcast that the next presidential contest will be, in his words, a referendum on artificial intelligence. Not a side debate. Not a panel at a policy conference. The election itself. He argued that 2028 has to be the AI election, because the buildout is arriving faster than the public story that is supposed to justify it. A veteran technology executive sitting beside him went further on the language. Supporters, he said, own none of the vocabulary. Pause. Swarms. Rogue. Every sharp word already belongs to the people who want the project slowed or stopped.

I have found that political fights are rarely won by the side with the better spreadsheet. They are won by the side that can say the fear in one sentence. Right now, the fear has the shorter sentence.

Why 2028 Could Become A Referendum On The Buildout

The case for calling the next race an AI election is not that voters will suddenly start quoting model benchmarks. Most people will never care what a parameter count means. They will care about the physical footprint. Substations. Transmission lines. Water for cooling. Night noise. Truck traffic during construction. And, above all, the electric bill that shows up whether or not they ever open a chatbot.

That is a different political object from the last decade of technology fights. Social platforms could be blamed, regulated, or ignored from the couch. A data center campus cannot. It occupies land. It asks the grid for power on a scale that used to belong to heavy industry. It asks neighbors to accept a building that employs fewer people per acre than a warehouse, while consuming more electricity than a small city. The promise is national. The inconvenience is local. Campaigns live and die in that mismatch.

Perhaps the most interesting aspect is how little room there is for a middle tone. Boosters talk about competitiveness, productivity, and not handing the lead to a rival power. Skeptics talk about blackouts, layoffs, and machines that do not pay property taxes the way a factory full of workers would. Both stories can be partly true. Only one of them currently fits on a yard sign.

The Messaging Problem Supporters Have Not Solved

The executive’s point about vocabulary is sharper than it first sounds. Political language is not a neutral container. Once a word sticks, it does half the arguing for you. Pause sounds prudent. Rogue sounds like a thriller. Swarms sounds like something you would not want over a school. None of those words were chosen by the people writing capital-expenditure plans. They were chosen by critics, and they traveled because they are easy to repeat.

Try the other side out loud. “We need accelerated permitting for high-density compute campuses co-located with new generation.” Accurate, maybe. Useless on a doorstep. The pro-buildout case requires a chain: more compute, more productivity, higher wages somewhere, strategic advantage, and a grid that somehow does not stick households with the tab. Every link is contestable. The anti-buildout case needs one link. Your bill. Your street. Your job.

The side that can name the fear in a single breath usually sets the terms. The side that needs a diagram is already late.

A pattern visible in almost every infrastructure fight

In my experience covering these mood shifts, complexity is not the same thing as weakness. It does mean the advocates have to pick a plain claim and defend it until it bores them. Right now they are still speaking in strategy memos. The other camp is speaking in warnings. Voters remember warnings.

What The Polls Are Actually Saying

Two snapshots from early autumn are hard to shrug off. In one survey of registered voters, 71 percent opposed a data center in their area, including 46 percent who felt strongly about it. Support sat at 26 percent. A separate national poll found the same 71 percent believing the costs of data centers outweigh the benefits, against 29 percent who saw the benefits as larger. That second survey also found 64 percent calling artificial intelligence bad for society, and 36 percent calling it good.

Those are not niche readings. They are majority positions with room to spare. They also travel across the usual partisan lines more than technology elites expect. A homeowner in a swing suburb and a renter in a rural county can dislike the same substation for different reasons and still land in the same column. That is how a single-issue mood becomes a referendum even if the ballot never uses the word referendum.

Does that mean artificial intelligence is politically dead? No. It means the default setting in much of the country is suspicion, and suspicion is easier to activate than to retire. Candidates who treat the buildout as an obvious good are walking into a room that has already decided the obvious good looks expensive.

Numbers like these move. A hot summer of outages, or a visible drop in bills, could shift them. What they describe today is a starting line, not a finish. Still, you do not get to invent a friendlier starting line because the capital plan prefers one.

Electricity Prices Are The Pocketbook Test

If there is a single hinge, it is the bill. People will tolerate an abstract risk for a while. They do not tolerate a line item that climbs while wages do not. Data centers are not the only reason regional prices move. Weather, fuel, aging plants, and slow transmission all play a part. None of that matters much at a kitchen table if the new load is the thing neighbors can point at.

The administration in Washington has tried to get ahead of that blame. A national policy framework for artificial intelligence asks Congress to speed federal permitting for the infrastructure, while also insisting residential customers should not pay higher electricity bills because of the new facilities. Alongside that, a ratepayer pledge pulled in major technology firms, including the largest cloud and model builders. The promise, in plain terms, is that those companies build, bring, or buy new generation and cover the upgrades their load requires, rather than socializing the cost through household rates.

That is the right political instinct. It is also a promise that has to survive contact with utility commissions, interconnection queues, and construction delays. A pledge signed in a capital is not the same thing as a transformer delivered to a county two years later. Voters will judge the pledge by the bill, not by the press release.

• New generation has to exist, not merely be announced, before the load arrives.
• Upgrade costs have to stay off the residential rate base in a way auditors can see.
• Local water, noise, and traffic impacts still need a separate answer. Power is not the only grievance.
• If any one campus breaks the promise, opponents will treat every campus as the same story.

I keep thinking of it as a receipt problem. The industry can talk about protected ratepayers all year. If a household in a contested state opens an envelope and the number is higher, the receipt wins. Campaigns know this. So do the firms writing the checks. The question is whether the engineering timeline and the political timeline can be forced onto the same calendar.

Permitting Speed Versus Neighborhood Patience

Streamlined permitting is the other half of the official push. From a competitiveness angle, the logic is straightforward. Compute capacity is a race. Rivals are not waiting on a county hearing. Delays compound. A project that slips two years is not a delayed project. It is a project that may lose the customer, the power allocation, or both.

From a neighborhood angle, streamlined sounds like steamrolled. People who already feel unconsulted hear “faster federal path” as “your objection will be noted and ignored.” That reaction is not always fair. Some objections are reflexive. Some are bargaining. Some are real constraints on water or substations that no slogan can engineer away. Sorting them takes time, and time is exactly what the race narrative says the country does not have.

Here is the tension a candidate will have to hold without dropping either plate. Go too slow, and the strategic argument collapses. Go too fast, and the local argument becomes a revolt that shows up in turnout, not just in comments on a planning portal. The 2028 field will be asked which plate they drop. Pretending there is no tradeoff is how you lose both rooms.

Jobs, Displacement, And The Story Each Side Will Tell

Employment is the other plain-language battlefield, and it cuts in two directions at once. Construction phases do hire. Electricians, pipefitters, concrete crews, and equipment operators will see real work. Once a campus is running, the permanent headcount is often modest next to the megawatts. Critics will put those two facts side by side and call the exchange a bad deal. Supporters will point at indirect jobs, supplier contracts, and productivity gains that show up far from the fence line.

Then there is the displacement story, which does not require a single data center in the county to feel personal. If a scheduling tool, a coding assistant, or a customer-service model trims a team, the worker does not experience that as “infrastructure.” They experience it as a smaller payroll. Polls that find a majority calling the technology bad for society are partly about this, not only about cooling towers. A candidate who talks only about server halls will sound like they missed the office.

Perhaps that is why the referendum frame fits. The public is being asked to accept a physical buildout whose benefits are diffuse and whose risks feel specific. Diffuse benefits lose to specific risks unless someone translates them into rent, wages, or a clinic that stays open. So far, that translation is thin.

How The Two Campaign Styles Might Sound

Imagine the pro-buildout speech, stripped of jargon. We will not hand the next industrial base to a competitor. The companies using the power will pay for the power. Your residential rate will be walled off. The jobs in the trades are real, and the productivity gains will show up in the businesses you already work for. Falling behind is the expensive option, not the cautious one.

Now the other speech. They want a factory that barely employs anyone, next to your house, on your grid, while the tools inside it are aimed at the work you do. They signed a pledge. Pledges do not pay bills. Slow it down until the numbers are honest.

Both can be delivered by serious people. Both will find an audience. The first needs evidence that accumulates before election day. The second needs only a few ugly local fights and a chart of rates. That asymmetry is what the software executive was pointing at when he said the case for speed is complicated and the case against it is simple. He was not being poetic. He was describing a media environment.

A rough campaign balance, as it looks from here:
  Short warning beats long upside
  Local site beats national strategy
  This month's bill beats a 2032 productivity chart
  A broken pledge beats ten kept ones

Local Politics Will Not Stay Local

County boards do not usually set presidential agendas. This issue might be an exception, because the facilities cluster. A handful of regions with cheap power, available land, and workable interconnection will absorb a disproportionate share of campuses. Those regions are not randomly distributed across safe seats. Some sit in states that decide national elections. A zoning war in one of those counties becomes footage. Footage becomes a question at a debate. A debate question becomes a litmus test for donors and for door-knockers who do not agree with each other.

I have watched smaller infrastructure fights jump that ladder before. A pipeline. A warehouse belt. A transmission line through farmland. The pattern is familiar. What is less familiar is the speed of the capital behind this one, and the way the same companies are also the characters in the job-displacement story. That double role makes it harder to split the coalition. You cannot easily be the champion of the trades on Main Street and the champion of headcount cuts on the earnings call without someone noticing.

State utility regulators matter here as much as Congress. If a commission lets upgrade costs leak into residential rates, the federal pledge becomes a punchline. If a commission holds the line and projects stall, the competitiveness camp calls it sabotage. Either headline is usable by someone running for president. Neither headline is fully under a candidate’s control. That loss of control is what makes the issue dangerous for frontrunners who would rather talk about anything else.

The Language War Is Already Underway

Go back to the vocabulary complaint, because it is the part campaigns can actually work on before a single new turbine spins. Words are not decoration. They are the on-ramps. Right now the on-ramps point toward caution. A supporter who says “pause” has already accepted the critic’s frame, even if the next sentence tries to climb out of it.

What would a different frame even sound like? Not “unleash,” which has been worn out. Not “inevitability,” which feels like a threat. Something closer to upkeep. The grid is aging. Demand was going to rise anyway. These loads can fund the rebuild if the contracts are written tightly. The alternative is a quieter stagnation that still shows up as higher rates and fewer options. Whether that framing works is an open bet. It at least does not begin inside the other side’s dictionary.

Critics, for their part, will keep the short words, and they should not be underestimated for it. A movement that can say swarm, rogue, and pause has a chorus. Choruses win school-board races and, occasionally, bigger ones. Dismissing that chorus as ignorant is how industries lose permission. People can be wrong about a model and right about a substation. Both things fit in the same voter.

What A Candidate Cannot Fake

There is a temptation, in every cycle, to rent a position until the primary ends. This issue is a poor rental. The buildout has timelines measured in years. A promise to “review” projects reads as a stall. A promise to “accelerate everything” reads as a giveaway. Donors in the industry will hear the hedge. Homeowners will hear it too. The internet keeps both tapes.

What cannot be faked is a mechanism. A ratepayer wall that has a regulator, a contract, and a penalty. A siting rule that publishes water use and backup-power plans in language a neighbor can check. A labor agreement that names the trades and the duration, instead of gesturing at “jobs.” Voters have become good at smelling a gesture. They have had practice.

1. Pick one measurable promise about household bills, and publish the test for whether it held.
2. Separate construction employment from permanent employment so the numbers cannot be blended.
3. Say which local impacts are negotiable and which are not, before the hearing, not after the shouting.
4. Admit the displacement risk in offices and shops, or the other side will own that sentence entirely.
5. Tie any call for speed to a project that already kept its power off the residential rate base.

None of that is a slogan. That is the point. The slogan comes after the mechanism, or it does not survive the first hostile clip.

Investors Are Not Spectators In This Fight

Capital is already positioned for a long build. Chip suppliers, power-equipment makers, utilities with growth plans, construction firms, and the platforms themselves have told the market some version of the same story: demand is structural, delays are temporary, and the political noise is manageable. Political noise stops being noise when it changes the queue. A moratorium in one state, a rate case that assigns costs differently, a transmission project that loses a season, and the model that looked smooth in a deck develops a limp.

That does not mean the buildout reverses. It means the path gets lumpier, and lumpy paths reprice the companies that assumed a straight one. Firms that truly self-supply power look different, in a backlash year, from firms that hoped the grid would stretch. Utilities that can show residential customers a clean separation look different from utilities that cannot. Equipment makers still sell into both worlds, but their customers’ timing shifts, and timing is margin.

I would watch three practical markers between now and the next inaugural, more than I would watch any single speech. First, whether announced generation actually reaches commercial operation on a schedule that matches new load. Second, whether residential rate cases in heavy-build regions stay calm or turn into campaigns of their own. Third, whether local election results in host counties start rejecting incumbents who approved campuses. Those markers are dull. They are also harder to spin than a podcast quote.

A Note On Rivalry And The National Frame

Supporters will keep reaching for the geopolitical argument, and they are not wrong that capacity has a strategic dimension. A country that cannot train or run advanced systems at scale accepts a dependency. Dependencies get expensive. The trouble is that strategic arguments rarely move a voter who is angry about a specific street. They move donors, generals, and a slice of the professional class. They do not, by themselves, move the 71 percent.

The bridge, if there is one, runs through visible public goods. A rebuilt substation that also hardens the town against outages. A training program that is not a press release. A tax contribution that shows up in a school budget the same year the trucks arrive. Without that bridge, the national frame floats above the local fight and never lands. With it, a candidate can say the rivalry part without sounding like they work for the campus.

Is that enough to flip a skeptical majority before 2028? Maybe not. It might be enough to keep the majority from hardening into a ban. Politics often rewards the option that prevents the worst outcome rather than the option that delivers the brochure. A buildout that survives in a narrower, better-contracted form would still be a buildout. A buildout that races ahead of its permissions might win the quarter and lose the decade.

The Cultural Unease Under The Rates

It would be a mistake to treat this as only a utility story. The same polling that rejects nearby campuses also finds a majority uneasy about the technology as a social fact. People worry about students who cannot write without a prompt, about fraud that sounds like their bank, about creative work that gets copied into a training set, about a customer-service loop that never reaches a person. Those worries do not require a zoning map. They travel by group chat.

When unease is already ambient, a physical project becomes the place to put it. The data center is visible. The model is not. Opponents understand that substitution instinctively. They will staple the ambient fear to the fence line, and some of the staples will hold even if the power contract is clean. A campaign that answers only the bill, and ignores the broader flinch, will sound evasive to people who wanted both answers.

None of this requires panic about science fiction. Ordinary institutional trust is low enough already. A technology that arrives inside institutions people do not trust inherits the distrust. That is an unromantic explanation. It fits the numbers better than a theory about voters failing to understand progress.

What “Referendum” Does And Does Not Mean

Calling 2028 an AI election can be overread. Presidential races are bundles. Inflation memories, borders, courts, wars, and the simple question of whether the incumbent mood feels tired will all be on the ballot. Artificial intelligence will not erase those. It may color them. A candidate’s answer on data centers becomes a proxy for whose side they are on when a large industry wants something from a small place.

That proxy is powerful because it is concrete. Voters who cannot adjudicate a foreign crisis can adjudicate a proposed campus on the edge of town. They will import that judgment into the bigger choice. This is how single issues colonize multi-issue elections. Not by replacing the other issues. By giving people a felt example of how power listens.

So the referendum, if it arrives, will be uneven. In some states it will be a headline. In others it will be a background hum that changes which anecdote a volunteer repeats. Either way, candidates who have not practiced the answer will improvise one, and improvisation on this subject tends to please nobody for long.

A Plausible Path That Is Neither Boom Nor Ban

There is a version of the next two years that does not match either camp’s poster. Projects that bring their own generation, or that contract for it firmly, move. Projects that lean on everyone else’s grid wait. Host communities that negotiate water, noise, and tax terms early sign. Communities that are handed a finished plan revolt. Federal permitting gets faster for corridors that are already studied, and stays slow where the study was wishful. Household rates in a few regions spike for older reasons, and the new load gets the blame anyway.

In that muddle, the election becomes a referendum less on the existence of the technology than on the terms. Who pays. Who decides. Who is allowed to say no. Those are old questions wearing a new building. Old questions are what elections are good at. New buildings are what make them urgent.

I suspect the winning message, if there is one, will sound almost boring. Pay for your power. Publish the rate impact. Hire locally for the years you actually hire. Leave people a real veto where the impact is irreversible, and a real timetable where it is not. Boring is not how technology culture likes to talk. Boring is how permission is sometimes kept.

Permission is the scarce resource. Capital is not.

Questions Worth Asking Before The Primaries Harden

A few questions cut through the speeches. If a campus misses its generation schedule, who eats the cost, and is that written down? If residential rates rise for unrelated reasons in a host region, how will the public be shown the difference? Which trades are contracted, for how many months, and what happens to those workers when the shell is finished? What water source is reserved in a dry year? Who can halt a phase, and on what evidence?

Candidates who can answer those without a binder have done the work. Candidates who pivot to competitiveness every time are telling you the local file is empty. Voters can hear an empty file. They have heard them before, about stadiums, about distribution centers, about plants that were supposed to anchor a town and did not.

The industry side has its own questions, and they deserve airtime too. If every community says not here, where does the capacity go, and what does the country do without it? If permitting returns to a decade-long default, which workloads simply leave? If the grid is not reinforced by someone, what fails first in a heat wave that has nothing to do with a model? A serious debate holds both sets of questions. A referendum often refuses to. That refusal is the risk the podcast warning was really about.

Why The Skeptical Mood May Stick

Moods fade when experience contradicts them. Moods stick when experience rhymes. If the next several campuses open quietly, pay for their upgrades, and sit beside towns that still recognize themselves, the 71 percent can soften. If a few open loudly, with rate surprises and thin hiring, the number can harden into identity. Identity is what you get when a local fight becomes a way of saying who the economy is for.

There is also a generational wrinkle that polls flatten. Younger workers use the tools and still fear the labor market those tools imply. Older homeowners may never use the tools and still fear the substation. A coalition that odd can still vote together if the ballot offers them a shared no. Building a shared yes is harder. It requires benefits that arrive on different schedules for different people, and campaigns are bad at staggered gratitude.

So the skeptical mood has a structural advantage heading into the cycle. It does not need coordination. It needs examples. Examples are cheaper to produce than proof of a national upside. That is an unpleasant arithmetic for anyone whose plans assume public patience.

Reading The Next Two Years Without The Hype

Strip the hype and the panic, and the situation is legible. A capital-intensive industry is trying to add a new class of industrial load during a period of tight power equipment, slow transmission, and low institutional trust. Political leaders want the capacity and do not want the bill. Communities want the tax base and do not want the burden. Workers want the tools’ productivity and do not want the tools’ leverage used against them. Those wants do not reconcile themselves.

Elections are one of the places the contradiction gets forced into a choice. Not a clean choice. A branded one. The brand, if the software executive is right, will be artificial intelligence, even when the argument on the doorstep is about a transformer and a rate case. Brands are how bundled grievances travel. This one already has the luggage packed.

Between now and that vote, the useful posture is specific. Ask who pays. Ask what was delivered versus announced. Ask which town actually consented, and on what terms. The national story will keep changing adjectives. The invoices will not.

A Closing Read On The Referendum Claim

Will 2028 be entirely about this? Almost certainly not. Will a candidate be able to reach the White House without a credible answer on data centers, power bills, and the social unease attached to the tools? That is looking less likely by the month. The polling is too lopsided to file under elite anxiety. The infrastructure is too physical to hide inside an app. The language of caution is too well rehearsed to be talked over with a competitiveness paragraph.

The executive who called it a referendum was describing a messaging mismatch as much as a policy one. Supporters are carrying a long argument into a short-argument contest. They can still win that contest, but not by insisting the public has misunderstood the future. The public has understood the part that touches the house. The rest of the case has to meet people there, in contracts and in bills, or it will meet them later, in a voting booth that has already made up its mind.

I do not think the outcome is written. I do think the side that keeps speaking in strategy, while the other side speaks in envelopes and street names, is volunteering for a brutal cycle. Brutal, here, does not mean chaotic for its own sake. It means a large, expensive, nationally framed project will be judged by people who never asked for it to be built next door. That judgment is what a referendum feels like, whether or not the word appears on the ballot.