The Grid Dividend

The problem

The first bill most Americans will receive from artificial intelligence is not a subscription. It is their electricity bill.

In the PJM Interconnection — the grid serving some 65 million people across thirteen states from Illinois to New Jersey — the capacity auction price went from $28.92 per megawatt-day for 2024–25 to $329.17 for the following period. That is an increase of about 833 percent, and it hit the price cap the Federal Energy Regulatory Commission had approved. Then it hit the cap again the next year.

Capacity payments are not an obscure line. They are what the grid pays generators to guarantee they can produce power when demand peaks, and they flow through to retail bills. PJM's own market monitor attributed 40 percent of capacity costs in the most recent auction to data centers. Analysis of the 2025/2026 auction attributed 63 percent of the price increase — about $9.3 billion — to data center demand. Across the last four auctions, the figure is roughly $29.4 billion added to electricity costs.

For a household, the projected effect in PJM territory is on the order of $70 a month by 2028. Pepco customers in Washington, DC are looking at about $21 a month, western Maryland around $18, Ohio around $16 — and those are the near-term numbers, before the buildout finishes.

Here is the structure of the injustice, stated plainly. A hyperscaler decides to site a facility. Its demand raises the clearing price for everyone in the market, because capacity prices are set at the margin and paid by all load. The retired machinist in Ohio who has never used a chatbot pays more so that a data center three counties away can train a model. He was not consulted, he receives nothing, and the mechanism by which he is charged is invisible to him.

This is not an argument against data centers. It is an argument about who pays for them, and it is the single most concrete way the AI economy is currently reaching ordinary American households.

The proposal

Large new loads pay the cost they cause, not the average. And the households in the region that hosts them receive a dividend for carrying the load.

Two halves, and both are necessary. The first is cost causation, which is orthodox utility regulation applied honestly. The second is the part nobody proposes: if a region's grid is going to be built out to serve an industry, the people who live on that grid should hold a stake in it rather than merely a bill.

How it would work

1. A large-load tariff class. Any new interconnection above a threshold — on the order of 100 megawatts — enters its own rate class. It pays the marginal cost of the capacity, transmission, and generation its arrival requires, rather than blending into the residential and commercial pool. This is not a penalty. It is the principle utility regulators already claim to follow, applied to a load large enough that the fiction of averaging breaks down.

1. Bring your own generation, or pay for someone to build it. Large loads must contract for firm new capacity matched to their demand, or fund it. The grid's problem is not that data centers use power; it is that they arrive faster than generation can be built, and the scarcity that creates is what the capacity price is measuring.

1. The Grid Dividend. The difference between what large loads pay under cost causation and what they would have paid under blended rates is not kept by the utility and is not swept into general revenue. It is returned to households in the affected region as a visible annual credit, paid on Dividend Day (№ 017), with a share deposited into the American Permanent Fund (№ 001) as principal.

1. Publish the attribution. PJM's market monitor already estimates data center contribution to capacity costs. Require every regional transmission organization to publish that attribution quarterly, in a standard format. The reason the machinist does not know why his bill went up is that nobody is required to tell him.

1. Speed as the incentive. Large loads that bring their own firm generation online, or that can curtail on demand during peak hours, pay materially less. The goal is not to keep data centers out. It is to make the fast, clean, curtailable ones cheaper than the ones that free-ride on everyone else's reliability.

The numbers

The transfer that already happened is the argument. $29.4 billion in added electricity costs across four PJM auctions is roughly $450 for every person in the PJM footprint, paid through bills most of them did not read closely, for infrastructure they do not own.

Run the counterfactual under this proposal. If even half of the $9.3 billion attributable to data centers in a single auction had been charged to cost-causers and returned, that is $4.65 billion to roughly 25 million PJM households: about $185 per household, once, from one auction. Applied across the four auctions, the household dividend approaches $600.

That is a real number and it is larger than most of the energy rebates Congress debates. And it requires no appropriation, no new tax, and no new agency. It requires charging a cost to the entity that caused it.

Against that, the honest counterweight: data centers pay substantial property taxes, and in some counties they have rebuilt school budgets. We are not claiming they are parasites. We are claiming that a benefit concentrated in one county and a cost spread across thirteen states is a bad arrangement, and that the fix is to align the two.

The honest objections

"Cost causation for large loads will push data centers to other regions or offshore." This is the strongest objection and we will not wave it off. Capital is mobile and states compete for these facilities with real subsidies. Three responses. Interconnection queues in the attractive regions are years long, which means demand currently exceeds the supply of grid access — a seller's market weakens the threat. Federal rather than state-level rules limit the race to the bottom within the country. And if a facility genuinely will not come unless other people pay its power costs, that is information about whether the project was ever worth what its boosters claimed.

"You are taxing the industry America most needs to win." We are pricing an input it currently underpays. Note what this proposal does not do: no cap on data center construction, no siting veto, no restriction on model training, no windfall levy on profits. A firm that brings its own generation pays little or nothing extra. That is a strange shape for an attack on an industry.

"Utility rate design is not a dividend program. This will get captured." A fair worry, and the reason the return must be a visible line item on the bill or a dated payment rather than a rate adjustment nobody perceives. Invisible benefits get captured precisely because nobody can see them being taken. The whole logic of № 017 applies here.

"Blaming data centers for electricity prices is too simple." It is, and the fuller picture includes retiring generation, transmission underbuild, gas prices, weather, and a decade of underinvestment. Data centers did not create every problem in the capacity market; they arrived into one and made it much more expensive. The 40 percent and 63 percent figures come from the market monitor and independent analysis, not from us, and we would revise them publicly if better estimates appeared.

"Regional dividends are unfair to people outside the host region." The costs are regional, so the returns should be too. Where a load raises prices across a whole interconnection, the attribution should follow the price effect rather than the county line. That is harder to compute and it is the right target.

Sources

• PJM Interconnection capacity auction results — clearing price of $329.17/MW-day against $28.92/MW-day in the prior period, at the FERC-approved cap (pjm.com)
• PJM Independent Market Monitor — data centers accounting for roughly 40% of capacity costs (utilitydive.com)
• Analyses attributing about 63% of the auction price increase, roughly $9.3B, and about $29.4B across four auctions, to data center demand (ieefa.org)
• Citizens Utility Board, data center reform and sustained high capacity prices (citizensutilityboard.org)
• Household bill impact projections for PJM territory (thehill.com)
• Proposals № 001 (The American Dividend), № 017 (Dividend Day)