The Fight Over Data Centers Is Real. The Facts Are Missing.

Welcome to the first issue of the AIAF monthly. We build and run AI automation for a living, which means we spend our days inside the technology everyone else is arguing about from a distance. That vantage point comes with an obligation: to cut through the noise instead of adding to it. This month, we're starting with the biggest infrastructure story in the country — the war over data centers — because it is being fought almost entirely on hearsay, and hearsay is a terrible foundation for a decision this big.

Here is our position up front, so you know where we stand. The data centers have to be built. AI is a defining industry of this century, and the United States cannot lead it without massive amounts of compute, which means massive amounts of power, land, and cooling. That is not in serious dispute.

But the way the industry is going about it — dropping billion-dollar facilities next to neighborhoods and letting rumor fill the silence — is creating an avoidable backlash that could slow the whole enterprise down. This is, at its core, a public relations failure. And it is fixable.

Let's do the thing nobody else is doing. Let's look at the facts.

Part 1What's actually true

The power draw is enormous, and it is growing fast. U.S. data centers consumed about 183 terawatt-hours of electricity in 2024 — more than 4% of all electricity used in the country. By 2030 that's projected to reach roughly 426 terawatt-hours, a 133% jump in six years. The Electric Power Research Institute estimates data centers could consume anywhere from 9% up to 17% of total U.S. electricity by the end of the decade, depending on how fast AI scales. For perspective, a single large AI hyperscaler can use about as much electricity in a year as the entire nation of Pakistan.

The concentration is the real story. Nationally, 4% sounds manageable. But these facilities cluster. There are now more than 4,000 data centers in the U.S., and they pile into a handful of states — Virginia (643), Texas (395), and California (319) lead. In Virginia, data centers already consumed roughly 26% of the entire state's electricity in 2023. When a county goes from zero to hosting a 300-megawatt facility, the local grid feels all of it at once. That's why this fight is local even though the technology is national.

Water is a genuine concern, and the numbers are real. U.S. data centers directly consumed around 17 billion gallons of water in 2023, mostly for cooling. Hyperscale facilities alone are projected to use between 16 and 33 billion gallons a year by 2028. In a water-stressed county, that's not an abstraction — it's a line item the community can feel. Two caveats sharpen this rather than soften it. That 17-billion-gallon number is direct, on-site water; the indirect water consumed by the power plants generating all that electricity is several times larger. And here's the cost that should trouble climate-minded readers most: the round-the-clock demand is pushing utilities to keep aging coal and gas plants running past their planned retirement dates. Right now, the AI boom is propping up the fossil fleet — a real environmental cost that never makes the jobs-and-tax brochure.

And yes, they are pushing up some electricity bills — but not the way the viral posts claim. This is where the hearsay gets dangerous, so read carefully. You may have seen the claim that bills near data centers rose "267%." That figure is misleading: it refers to wholesale prices at specific grid nodes, not the bill a household actually pays. PolitiFact rated the popular version of that claim "Mostly False." Wholesale power is only 30–50% of a residential bill; the rest is transmission, distribution, and taxes.

Here's what's actually true. U.S. residential electricity prices rose about 42% nationally over five years, with steeper spikes in data-center-heavy regions — Washington, D.C. (94%), Maryland (74%), Maine (73%), New York (58%). In the PJM grid region, the cost of guaranteeing enough capacity jumped $9.3 billion — a 174% increase — for the 2025–26 year, driven substantially by data center demand. Carnegie Mellon researchers estimate data centers could raise the average American electricity bill about 8% by 2030, and more than 25% in the highest-demand markets. Utilities requested more than $29 billion in rate increases in just the first half of 2025 — double the year before, affecting some 40 million customers.

But here is the twist that makes this genuinely contested, and that almost no viral post will ever tell you: a landmark 2026 study from the Electric Power Research Institute found that from 2015 to 2024, data center load growth may have actually lowered electricity costs in many markets — by spreading the utility's huge fixed infrastructure costs across a much larger base of demand. A big new customer can pull the average cost per household down. The danger is that when the buildout gets large enough and fast enough, the cost of brand-new generation and transmission swamps that benefit and flips it the other way. Serious analysts genuinely disagree about where that tipping point is, and whether some regions have already crossed it. Anyone who tells you the answer is simple is selling something.

So the honest verdict on the economics of cost: data centers are demonstrably raising electricity costs in the regions that host them, the effect is real but smaller and more localized than the outrage suggests, and multiple factors — aging infrastructure, equipment costs — are tangled up in it. That nuance is exactly what gets lost when the conversation is left to viral screenshots.

Part 2Why communities are right to be angry (and where the industry blew it)

Now the other side of the ledger — because the benefits are just as real and just as under-communicated.

Data center construction is running at about a $59 billion annualized pace, roughly 8% of all private nonresidential construction in the country, up 23% year over year. A single large project can mean around 9,700 construction jobs during the build, $2.4 billion in economic output, and $84 million a year in state and local tax revenue. Virginia's own legislative watchdog found the industry supports roughly 74,000 jobs a year and contributes $9.1 billion annually to the state's economy — and in some jurisdictions, data centers pay up to 30% of local tax revenue. That is a school system. That is a fire department. That is a property-tax base that keeps a rural county solvent.

And here's where we hold ourselves to the same honesty we're asking of the industry, because the critics have a real point: those benefits are gross, not net. States hand data centers enormous tax breaks to win them. Virginia's data center sales-tax exemption alone cost the state an estimated $1.9 billion in fiscal 2025 — more than half of all the economic-development incentive spending in the entire state. Whether any given project is actually a win for local taxpayers depends entirely on whether the jobs and tax base it adds outweigh the breaks it's handed. Sometimes they clearly do. Sometimes they don't. Every community owes it to itself to run that specific math before it signs — and the industry should be the one putting the honest numbers on the table.

Here's the problem: almost nobody in those communities knows any of this. They know the noise. They know the diesel backup generators. They know the transmission lines and the water draw and the rumor that their bill is about to double. When a family stands up at a town hall in Lysander, New York — where 350-plus residents turned out to oppose a 300-megawatt facility and won a six-month moratorium — they are not being irrational. They are filling an information vacuum the industry left wide open.

And the vacuum is spreading. As of mid-2026, 14 states have considered data center moratoriums, and local pauses have landed in Groton, Connecticut; Peculiar, Missouri; Bangor, Maine; and a growing list of New York towns. The concerns are consistent: noise, diesel emissions, water, property values, and electricity rates. Every one of those is a legitimate thing to worry about. And every one of them is something the industry could have gotten ahead of with honest engagement — showing up early, sharing the real numbers, funding the local benefits, and treating residents as partners instead of obstacles.

That is the PR failure in one sentence: the companies building the future of computing let the story be written by everyone except themselves. You cannot win a fight you refuse to show up for.

Part 3What actually solves this

Diagnosis is easy. Here's where we think the real solutions are — some already happening, some worth pushing for.

1. Make the companies pay their own way — and let that drive innovation.

The single biggest driver of community anger is the fear that ordinary ratepayers are subsidizing billion-dollar tech infrastructure. The fix is to make sure they aren't — and there's now a real precedent. In late 2025, Ohio regulators approved a landmark AEP Ohio tariff requiring large data centers (loads over 25 MW) to pay for at least 85% of the capacity they reserve for up to 12 years, whether they use it or not, with exit fees and financial guarantees. The stated purpose, in the commission's own words, is to ensure infrastructure costs are "borne by those incurring them, and not shifted to other ratepayers." That's the model. It's spreading, and it should.

There's a more radical version of this idea worth putting on the table. On the All-In podcast, investor David Friedberg floated allocating the grid's cost differently: raise the price on industrial and commercial electricity — he suggested by roughly 50% — and in exchange make residential electricity effectively free up to a monthly cap. (Worth noting: the number that circulates is sometimes remembered as a "20% tax," but Friedberg's actual framing was closer to a 50% increase on industrial and commercial use.) His argument isn't just about fairness. It's that if industry were forced to bear the true cost of power, the smartest, best-capitalized companies on earth would race to build cheaper, cleaner, more abundant generation — and the whole grid would get stronger as a result. Friedberg has made the broader point bluntly: if America were adding a terawatt of new electricity capacity a year, our energy anxieties would largely disappear.

The skeptics have a fair rebuttal, and we'll give it to you straight: transmission and distribution make up roughly half of a residential bill, and no amount of generation subsidy zeroes those out; industrial users also tend to pass higher input costs right back to consumers in the price of goods. So "free residential power" is more provocation than shovel-ready policy. But provocations can still point at something true. You don't have to adopt the full version to see the logic: the entity that creates the demand should carry the cost of meeting it, and pricing it that way turns the world's most resourceful firms into an engine for grid abundance instead of a drain on it. That is a solution worth debating seriously rather than dismissing.

2. Site them intelligently — with real setbacks, not slogans.

Distance from homes is the community's biggest lever, and it's a reasonable one. But it has to be calibrated to reality. An instinct we hear a lot — "keep them 20 miles from any neighborhood" — sounds right but would effectively ban data centers anywhere near the metro areas whose grids and fiber they actually need; you'd be siting national infrastructure in the middle of nowhere and then wondering why it can't connect. The serious policy conversation is happening in fractions of a mile. Advocacy groups and a growing number of ordinances converge on roughly a half-mile (2,640-foot) setback from residential property lines to manage noise, low-frequency hum, and air quality from generators — while cities like Tempe, Arizona have started with 500 feet (widely regarded as too little). Half a mile, backed by hard noise and emissions limits, is the number worth fighting for. It protects neighborhoods without pretending the facilities can be exiled to the horizon.

3. Fix the public relations — for real, not with a press release.

This is the one the industry controls entirely, and it's the one it's failing worst. Winning community consent isn't spin; it's substance communicated honestly:

• Show up before the zoning fight, not during it. Bring the real numbers — jobs, tax base, the specific grid-cost protections — to the town hall while trust can still be built.
• Fund visible local benefits. A share of that 30%-of-local-tax-revenue should land as things residents can see: lower county taxes, upgraded water systems, community energy projects.
• Publish the water and power commitments and let people hold you to them. Secrecy is what breeds the 267% rumors. Transparency is what kills them.
• Tell the national story too. Every one of these facilities is a piece of American AI leadership. That case is never made to the people being asked to host it.