The White House says data centers will pay for their own power, not you. The promise binds no one, and the two words holding it up already leak.

A pledge is not a tariff. That is the whole thing to hold onto this week, because the White House expanded one and described it as if it were the other. The Ratepayer Protection Pledge, first signed in the spring by Amazon, Google, Meta, Microsoft, OpenAI, Oracle and xAI, now includes more than two hundred additional utilities, data-center developers, cooperatives and states, and by the administration's own count covers eighty percent of the electricity delivered to American homes and businesses and protects two hundred and sixty-three million people. Those are the numbers in the announcement. They are large, they are round, and they decide nothing. A pledge builds no megawatt and binds no utility. It is a promise to behave, and the question I always end up asking about a promise is the same question I ask about a battery or a reactor: what does it actually deliver, on what date, and who pays if it does not.

So price it. Read what the pledge says the companies will do, and then read the two words that decide whether it means anything.

What the pledge promises

The commitment is straightforward and, on its face, reasonable. Signatories agree that when a new data center arrives, the company will build, bring, or buy the new generation needed to meet its own demand and pay the full cost of it, rather than drawing down power everyone else was already counting on. They agree to pay for the new delivery infrastructure — the transmission lines, the substation upgrades — required to serve the facility. The stated goal is that the household down the road does not see its bill rise because a hyperscaler plugged a campus into the local grid. If that is what happened in practice, it would be a genuinely good outcome, and I would say so.

But the mechanism that makes power bills rise does not work the way the pledge imagines it. When a utility builds a new transmission line or upgrades a substation, the cost of that infrastructure is not usually assigned to one customer. It is spread — socialized, in the regulator's word — across everyone on the grid, because the wires are shared and the accounting treats them as shared. That default is the thing that quietly moves money from households to industrial loads. A pledge to pay "the full cost" only reverses that default if the full cost can be cleanly assigned to the data center in the first place. And here the language does something specific: it limits the promise to the costs directly attributable to the facility. Two words, and they are carrying the entire pledge.

A pledge builds no megawatt and binds no utility. It is a promise to behave, and the question is the same one I ask about a battery: what does it deliver, on what date, and who pays if it does not.

The two words that leak

"Directly attributable" sounds like a tightening. It is a loophole, and an old one. A great deal of what a data center costs the grid is not direct. It is a wholesale price that ticks up across an entire region because a few gigawatts of new demand tightened the market for everyone. It is a transmission upgrade that was needed to serve the data center but also, conveniently, strengthens the network for other customers, and so gets shared. It is the reserve capacity a grid operator has to hold to keep the lights on when a hungry, always-on load spikes. None of that is neatly, directly attributable to one company's fence line, which means under the pledge's own wording none of it has to be paid by the company. It falls back to the default. It socializes. The households the pledge is named for keep paying for it, and the pledge remains, technically, honored.

This is not a hypothetical leak. It is already measured. Across the thirteen states of the PJM grid, ratepayers were charged about four and a half billion dollars for transmission upgrades tied to data-center growth between 2022 and 2024 — costs that arrived on ordinary bills under exactly the socialization default the pledge does not change. In Virginia, the densest data-center market on earth, residents have watched monthly bills climb; one household's power bill ran to two hundred and eighty-one dollars in January, roughly triple its normal level. That money moved before any pledge, under the rules that are still in force, and a voluntary commitment layered on top of those rules does not claw a dollar of it back. The four and a half billion is the number that matters. The eighty percent and the two hundred and sixty-three million are the numbers in the press release.

• The press numbers: covers eighty percent of US power, protects two hundred and sixty-three million people, more than two hundred new signatories.
• The number that decides your bill: about four and a half billion dollars in transmission costs already socialized across PJM ratepayers to serve data centers, 2022 to 2024 — under rules the pledge leaves untouched.
• The words that decide the rest: 'directly attributable,' which excludes the shared, indirect and market-wide costs that make up much of a data center's true burden on the grid.

A promise with no enforcement

Then there is the nature of the instrument. The pledge is voluntary. There is no regulator empowered to check whether a signatory actually paid its full share, no penalty for a company that signs and then lets the usual socialization proceed, no filing a household could point to. Analysts who study utility rates have been blunt about it: a voluntary pledge to protect ratepayers carries, in enforcement terms, roughly the weight of a New Year's resolution. The Brookings Institution, reviewing it, concluded the plainest possible thing — that a pledge to protect ratepayers needs enforcement to protect ratepayers. Without a binding tariff behind it, the document is a statement of intent, and intent does not appear on a bill.

The clearest evidence that the pledge is not sufficient is that its own backers are trying to replace it. Congress is now moving to codify ratepayer protections into law, which is the tell: you do not write a statute to enforce a promise that was already working. And where states have tried to convert the sentiment into something binding, the carve-outs have followed the money. Tennessee enacted a version of the protection and left a gas-turbine loophole wide enough that the communities nearest the data centers can still absorb the pollution, the noise, and a share of the cost. The lesson travels, and I have watched it travel: the gap between the promised protection and the delivered invoice is not an American peculiarity. It is what happens anywhere the people who benefit from cheap industrial power are not the people who pay to firm it up. Socialized cost is regressive by design. It lands hardest on the households with the least room to absorb it, and the least standing to contest it.

The invoice is the honest document

None of this means data centers cannot be made to pay their own way. They can. The tools exist, and they are boring, which is how you know they would work: a binding tariff class for very large loads, so the biggest customers are billed for the capacity they actually require; a requirement that new demand bring its own firm generation before it connects, not as a pledge but as a condition of service; cost-allocation rules that assign shared upgrades to the load that triggered them instead of defaulting to socialize. Some states are already writing these. They are unglamorous, enforceable, and specific, and not one of them is a pledge. That is the point. The thing that protects a ratepayer is a rule with a number and a penalty, not a signature at a summit.

So measure this announcement the way you would measure any energy claim: not by the size of the promise but by the megawatts and the dollars it actually moves. By that test the Ratepayer Protection Pledge moves neither. It generates no power, it reallocates no cost that was already socialized, and it binds no one to do either. The demand is real and it is coming fast — American data centers are on track to consume a rising share of the grid every year this decade — and the bill for firming that demand is real too. The only open question is whose name is on it. The pledge answers that question with eighty percent and two hundred and sixty-three million and a phrase, 'directly attributable,' engineered to leak. The invoice answers it with four and a half billion dollars, and the invoice is the honest document. Watch the tariffs, not the pledges. The tariffs are where anyone actually decides who pays.