Few human creations in the world get more hate than data centers. In the US, retaliatory action is now bipartisan, meaning everyone seems to be against them.

Nonetheless, a Reuters/Ipsos poll found that just 14% of Americans supported a data center in their community, compared with 36% who approved of US strikes against Iran.

Because it raises electricity bills.

Because it wastes water like there’s no tomorrow.

Because they are noisy..

Choose your preferred reason to hate; all are valid these days. Whether they are accurate or not is another thing.

But does this make sense? Do they raise prices? Do they consume a lot of water?

Will answer many of these questions today. The culprits are many: Auctions, closed loops, poor measurements, populism… this has it all.

This exercise has served as an eerie reminder to myself that decisions aren’t always based on hard data, and that populism, bad modeling, and ultimately outright lies matter just as much, or more, these days.

I don’t think I’m going to surprise you when I tell you data centers are deeply unpopular.

In a June 2026 Reuters/Ipsos survey of 4,531 Americans, only 33% approved of the current pace of data-center construction, while 64% disapproved.

When respondents were asked about their own communities, 57% opposed a local data center, and only 14% supported one.

Importantly, the resistance was bipartisan: approximately two-thirds of Democrats and half of Republicans opposed having one nearby. A separate Public First survey covering 15 countries found that only 26% of Americans supported expanding data-center infrastructure, the lowest result in the study and below Britain, France and Germany.

That sentiment has now escaped opinion polls. In July, organizers coordinated 142 anti-data-center demonstrations across 42 states. Texas hosted 18 events, Georgia 11, California eight, while Pennsylvania, Florida and Indiana hosted seven each. National attendance figures were unavailable, and some rallies were small, so this was not yet a mass movement.

But the political diversity of these actions is what should really worry AI incumbents: opposition is appearing in Republican, Democratic and swing states for broadly similar reasons, including electricity prices, water consumption, tax concessions and the secrecy surrounding local negotiations.

And, as you would’ve expected, the one cohort that owes allegiance to anything but holding to power, the politician, noticed.

But before we go deep, I want to make one thing clear from the get-go: moratoriums don’t delay overall progress, no matter how intuitive that sounds, because that capacity is built elsewhere.

At the federal level, Bernie Sanders and Alexandria Ocasio-Cortez, the latter of which is polling in prediction markets as the favorite for the 2028 presidential campaign, introduced legislation proposing a nationwide moratorium on new AI data centers until the nation adopts safeguards.

As expected, the bill has not become law. The Trump administration has chosen the opposite approach: continue building, but make developers “pay their own way.”

Amazon, Google, Meta, Microsoft, OpenAI, Oracle and xAI signed a voluntary White House pledge to procure additional electricity, finance the grid upgrades their projects require and pay for contracted infrastructure even if they ultimately use less power than expected. The two approaches are politically distant, but both begin from the same premise: ordinary customers should not underwrite the AI infrastructure boom.

The most consequential action, however, is happening at the state level. Pennsylvania Governor Josh Shapiro (another Democrat presidential hopeful) has removed data centers from the state’s accelerated permitting program, prohibited state agencies from signing nondisclosure agreements with developers, and conditioned favorable permit treatment and tax exemptions on stringent requirements.

Developers are expected to pay for the generation, transmission, and distribution infrastructure attributable to their projects, disclose energy and water requirements, and secure local approval. Projects that refuse those conditions are not formally prohibited, but they face a slower and less predictable permitting process.

In other words, it’s not a ‘NO’ as many read it in the press, but a ‘do it as we want, otherwise your permits will slow you down.

New York has gone further, although not quite as far as the word “moratorium” suggests. Governor Kathy Hochul’s July order temporarily freezes incomplete applications for discretionary state environmental permits covering new or expanded data centers capable of consuming at least 50 MW. Existing facilities, completed applications, local permits and qualifying research, education, manufacturing and medical facilities are not covered.

The state will use the pause to conduct an environmental review and consider upfront grid contributions, insurance against abandoned infrastructure and requirements to finance new generation or storage.

But here’s the surprise many didn’t expect: Republican states are acting similarly, although with different language.

Texas Governor Abbott signed SB 6, which requires large loads generally exceeding 75 MW to demonstrate control of their proposed site, disclose potentially duplicative connection requests, contribute to interconnection costs, and post financial security for transmission construction.

It also allows ERCOT (the non-profit entity that manages electricity transmission in Texas) to curtail those customers or require them to activate backup generation during emergencies.

Governor Greg Abbott subsequently ordered an audit before any data-center connection could proceed. ERCOT was considering 474 GW of proposed new load, more than five times Texas’s record peak demand, and approximately 90% of those requests were attributed to data centers. That does not mean Texas will actually add 427 GW of data-center demand; it’s a projection and a really poor one, likely, which, as we’ll see later, matters a lot.

Florida is another example, and the most comprehensive in my view. Utilities must create special tariffs for customers expecting peak demand of at least 50 MW, ensuring that they bear the full cost of serving them rather than shifting it onto other ratepayers.

The law permits minimum-demand charges, financial guarantees, long-term contracts, early-termination fees, and take-or-pay provisions. Under take-or-pay, a data center commits to paying for a minimum amount of electrical capacity whether or not it ultimately uses it.

If a utility expands its generation, transmission, or connection infrastructure for a promised data center that is never built or uses far less power than expected, the customer remains responsible for the contracted capacity. This protects ordinary ratepayers from bearing the cost of stranded infrastructure. It also preserves emergency curtailment and adds special water-permitting requirements.

But why so much focus on ensuring these projected loads materialize? Well, as you’re about to find out, this alone can raise electricity prices.

Europe is moving in the same direction, but with less political theatre. EU rules require data centers with at least 500 kW of installed IT power to report electricity consumption—that’s literally less than what a single Rubin Ultra NVIDIA server will require next year, which goes to prove how outdated Europe is in all these matters—water consumption, renewable-energy use and efficiency indicators to a common European database.

So, is this as bad as the press and social media make it out to be? No, things are more nuanced. But it surely doesn’t help that the politicians make it a bigger deal than it is when they announce their bills.

Thus, our first insight is that calling this a global revolt against data centers would be premature. Politicians are taking positions, but most of them are surprisingly logical for the most part, just blown out of proportion by the press (and by them, too, in order to score political points).

But do data centers really push up prices? Let’s see.

The rationale behind price increases is dead simple: if demand exceeds supply, prices rise. What is less intuitive is that electricity prices can rise because demand is expected to rise, even if it doesn’t.

But how does that make sense?

And the answer is the US's largest grid operator, PJM, and its potential $12 billion mistake.

PJM (which stands for Pennsylvania, Jersey, and Maryland, although it serves 13 states and 67 million Americans today) is the largest grid operator in the US.

Unlike ERCOT (Texas), which doesn’t auction capacity, PJM pays electricity suppliers for two things: capacity and electricity.

But what is a capacity market?

Years before demand is actually created, PJM holds a “capacity auction” based on predicted demand. In layman’s terms, PJM takes in demand requests, puts them into a proprietary forecasting model, predicts that demand in 2028 will be, say, 158 GW, and then holds a capacity auction where suppliers bid prices to be available as supply.

This lets suppliers look two years or more ahead and gives them a clear incentive to build new supply or guarantee current supply is available then. In exchange, and here’s the key, PJM agrees to pay a daily capacity payment simply for being available, even if that supply source doesn’t even participate in that demand.

I know that sounds confusing, so let me clarify. For example, say the agreed price on that 2028 auction is $300/MW-day. If I agree to participate as supply that year for 2 GW, I’ll get paid $300×2,000 = $600,000 per day for the entire 2028 year. This value is paid daily, even if I don’t contribute any electricity that day. Conversely, if I fail to appear, I face very heavy penalties.

Therefore, citizens in the PJM area, one-fifth of America’s population, not only have to pay for electricity at the price agreed on the daily electricity auction, but also pay for those $300/MW-day x number of MWs available.

This is quite a bit of money. If the agreed capacity supply was 158 GW and the agreed price was $300/MW-day, that is $17 billion every year PJM ratepayers (i.e., you if you live in that area) have to pay apart from electricity.

At this point, you can easily tell what is going on:

Given this, some of the regulations we’re seeing in the region (e.g., Gov. Shapiro’s) make much more sense: data center operations shouldn’t just freely add 10 GW of power to demand curve projections; they need to show there’s a strong chance the data center will be built.

Being an accelerationist (i.e., zero regulation) when you’re a billionaire is very easy, but many Americans do notice electricity bills rising. This disconnect between Silicon Valley elites and reality proves once again that the AI industry has some of the worst PR faces and lack of self-awareness we’ve ever seen, on both sides of the aisle: those who want so much regulation that only they survive, (e.g., the regulatory-capture, power-hoarding bros like Anthropic and to some extent OpenAI), and the zero-regulation bros, who show debatable empathy for the average American. But I digress.

Worryingly, it’s not only that many data center projects are being proposed, which shouldn’t necessarily increase prices, but Americans also face another enemy: poor demand projection.

But how?