I have sat in two county meetings about the same kind of project and watched them go opposite ways. Same size build, same tax abatement structure, roughly the same utility. One room was fine with it. The other one was not, and the reason had nothing to do with how much electricity the facility would use.

It came down to a single line in the interconnection agreement about who was paying for the substation.

The number everyone quotes

Start with the part that is well established. Berkeley Lab's 2024 United States Data Center Energy Usage Report, commissioned by the Department of Energy, put data center electricity consumption at 176 terawatt hours in 2023, which is 4.4 percent of total United States electricity. That is up from 1.9 percent in 2018 and 58 terawatt hours in 2014. The same report projects between 6.7 and 12 percent by 2028.

Those are real numbers from a credible source and the growth is genuine. What does not follow from them is the claim usually attached, which is that this consumption is what is showing up on residential bills.

The Energy Information Administration expects nominal average United States electricity prices to rise 13 percent between 2022 and 2025, and it attributes that to utilities increasing capital investment to replace or upgrade aging generation and delivery infrastructure. Data centers do not appear in that explanation. Distribution spending, not generation, is where the money went.

That gap between the two stories is worth understanding, because the real mechanism is more consequential than the one people argue about.

What a utility bill actually is

A residential electricity bill is not primarily a bill for electricity. It is mostly a bill for the poles, wires, substations and transformers that deliver it, plus a regulated return on the capital that built them. The energy itself is a minority of the total in most jurisdictions.

Which changes the question entirely. It is not how much a data center consumes. It is who pays for the build required to serve it.

The cost that moves onto your bill is not the electricity. It is the build, and where that lands is a regulatory choice rather than a physical fact.

Once you see it that way, both halves of the public argument turn out to be right, in different places, for the same reason.

Both directions are real

New load can genuinely lower rates. Fixed costs get spread across more kilowatt hours, and the Columbia Center on Global Energy Policy's literature review found that between 2019 and 2024 states with faster load growth generally experienced smaller price increases or even real declines. A California analysis found a typical 100 megawatt hyperscale facility generated roughly 3.4 million dollars of surplus value in 2025 against the marginal cost of serving it, projected to reach 6.1 million by 2030. PG&E's own analysis suggests every gigawatt of new data center load could reduce average household bills by 1 to 2 percent.

New load can also raise rates, in the same country, in the same year. The same review found that ratepayers across seven of thirteen PJM states were charged 4.3 billion dollars in 2024 for transmission projects required to connect data centers. Proximity to data centers has been associated with local marginal price increases of up to 20 percent.

Neither finding is wrong. They describe two different cost allocation regimes, and the physical facts underneath them are almost identical.

There is a structural reason the second outcome keeps happening. A regulated utility earns a guaranteed return on capital investment. That creates a pull toward building more infrastructure and recovering it broadly rather than assigning it narrowly, which is not corruption, just an incentive working as designed. Distribution capital spending rose 160 percent between 2003 and 2023.

Two counties, one build

Take a company siting a 60 megawatt facility with two viable locations. The energy price per kilowatt hour is close enough to identical that it does not decide anything.

In the first county, the utility socialises the interconnection and transmission upgrade across the ratepayer base. On the spreadsheet this is the cheaper site, sometimes dramatically so, and it will win a first-round site selection almost every time.

In the second, a beneficiary-pays structure applies. The load finances its own generation and transmission upgrades, accepts curtailment during grid stress, and sits on a separate rate schedule. Year one costs more. Sometimes a great deal more.

Now run both forward five years. The first county's residential customers see a line item they did not vote for attached to a facility that employs forty people. That produces a rate case, a moratorium, a special tariff applied to the class you are in, or all three. The company that chose the cheap site is now negotiating from inside a political problem, and the retroactive tariff is not something it can model in advance.

The second county produced no backlash, because nobody was asked to pay for something that was not theirs. The higher year-one number bought a stable ten-year one.

The objection that this is not your problem

The reasonable pushback is that all of this is regulatory, decided in state commission proceedings, and there is nothing here for a technology leader to act on. Utilities and public utility commissions will resolve it, and enterprises will pay whatever the tariff says.

That holds right up until you notice where those proceedings are and who is not in the room.

If your organisation is building or leasing significant compute, this is likely the largest single cost exposure in the program and the least negotiated. It is being settled in dockets that are open, scheduled, and almost never attended by the load. The terms set there outlast every vendor contract you will sign.

If you are not the load, it still lands on you, because your facilities and colocation costs are set by the same tariffs. And the direction of travel is clear enough to plan against. Beneficiary-pays structures are spreading, and they change siting economics in ways that make some current infrastructure roadmaps look considerably worse than they did when they were approved.

Questions worth asking

Three of them, and none requires an energy specialist to answer.

For any compute build or lease in your plan, ask who is financing the interconnection and transmission upgrade, and whether that answer is written down or assumed. Ask whether the sites under consideration sit in jurisdictions with beneficiary-pays rules already adopted, proposed, or absent. And ask what your cost looks like if the jurisdiction you chose adopts one after you break ground.

If nobody on the team can answer the first question, the project has an unpriced exposure larger than most of the line items that did get priced.

What this changes

Not much about how much power you use. Efficiency work is still worth doing and it is not the lever here.

What changes is that siting decisions stop being a comparison of energy prices and start being a comparison of cost allocation regimes, which is where the actual money is. It also changes the conversation with leadership and the board from a sustainability question, which invites a gesture, into a cost and continuity question, which invites a decision.

And it changes how you read the headlines. When a community fights a data center, they are almost never objecting to the electricity. They are objecting to being handed the invoice for the wires.

Your bill is going up because someone decided who pays for the build. That decision is still being made, in public, on a published calendar.