The AI Trade Has a $23-Billion Electricity Bill, and Regulators Just Started Reading It
Every AI capex forecast assumes the power gets built. The people who pay for the grid have started asking why they are paying for it — and that is a constraint no model has priced.
- PJM's independent market monitor found roughly $23 billion in customer price increases attributable to expected data center power demand, lasting through at least the end of 2028.
- The finding appeared in the PJM State of the Market report for Q1 2026.
- PJM covers all or part of 14 mid-Atlantic and Midwest states — the largest US grid operator by population served.
- PJM's market watchdog described the shift as a 'massive wealth transfer' to technology companies.
- Washington State moved in August 2026 to prevent data centers from raising consumer electricity prices.
- A structural loophole exists: data centers can curtail consumption to avoid coincident-peak demand charges, shifting cost recovery to other ratepayers.
Every model of the AI buildout — every capex forecast, every hyperscaler revenue projection, every data center REIT valuation — rests on an assumption so basic it usually goes unstated: the electricity will be there, and someone will pay to deliver it.
PJM's own independent market monitor has now put a number on the second half of that assumption. Roughly $23 billion in customer price increases attributable to expected data center power demand, running through at least the end of 2028.
The monitor's own characterisation of the dynamic: a "massive wealth transfer" to technology companies.
That phrase came from inside the grid operator, not from an advocacy group. It is the kind of language that precedes regulation.
How a data center ends up on your power bill
The mechanism is unglamorous and matters enormously.
Grid costs are overwhelmingly fixed infrastructure. Generation capacity that must exist for peak demand. Transmission lines. Reliability reserves. Those costs are recovered through rates charged across the customer base, and the allocation between customer classes is set by state utility commissions.
When a 500-megawatt data center arrives — and dozens have — the grid must procure capacity and build transmission to serve it. That cost enters the rate base. The commission then decides how much the data center pays and how much everyone else pays.
Two structural problems follow.
The forecast may be wrong. Not every announced data center gets built, and those that do frequently draw less power than projected. But the capacity is procured on the forecast. If the load does not materialise, ratepayers have funded infrastructure serving no one.
Peak charges can be engineered around. Coincident-peak charges are designed to make the customers driving peak capacity needs pay for it. A large, sophisticated facility can predict peak hours and curtail draw specifically during them — reducing its assessed share of infrastructure costs while consuming nearly as much overall. That is legal, rational, and it shifts cost recovery onto customers who cannot do it.
The result, as PJM's monitor put it: residential ratepayers absorbing costs for infrastructure built to serve facilities that may never materialise or may underperform.
The politics have started moving
This is the part that changes the investment case, because it has left the technical literature and entered legislatures.
Washington State moved in August 2026 to prevent data centers from raising consumer electricity prices. It will not be the last. Utility rate cases are among the most politically visible proceedings a state government runs, and "your bill went up so a technology company could train models" is an unusually easy argument to make.
Technology companies have pledged to pay their fair share. The implementation mechanisms remain unclear, which in regulatory practice means they do not yet exist.
What this means for your portfolio
Regulated utilities have a genuine bull case here — with a ceiling. A regulated utility typically earns an authorised return on invested capital. More demand justifies more investment, which grows the earnings base. Load growth after two decades of flat electricity demand is a real change in the sector's outlook.
But the authorised return and the cost allocation are both set by commissions that answer to elected officials. That is the ceiling. Utilities are not a pure play on AI demand; they are a play on AI demand filtered through a political process.
Independent power producers have cleaner exposure and more volatility. IPPs selling into merchant markets capture higher power prices directly without the regulated return cap — and without its floor.
The second-order risk is to AI capex itself. If the grid cannot deliver power on the assumed schedule, or if cost allocation fights delay interconnection, the buildout slows. It does not stop. But a lot of valuation in the AI complex depends on when revenue arrives, not just whether it does.
Watch this as a risk factor, not a trade. The useful output of this story is not a stock to buy. It is a constraint to keep in mind when you read a capex forecast that treats power as a solved problem.
What to watch next
- State utility commission rate cases in PJM states, where cost allocation between data centers and residential customers is being decided in real time.
- Interconnection queue times. Lengthening queues are the clearest evidence power is a binding constraint.
- Special tariff classes for large loads. Several states are designing them; they are the mechanism that would resolve this without legislation.
- Whether hyperscalers move to behind-the-meter generation — building their own power, on site, and stepping outside the rate base entirely.
Frequently asked questions
Why do data centers raise electricity prices for households?
Grid costs are largely fixed infrastructure — generation capacity, transmission lines, reliability reserves — recovered through rates charged to all customers. When a large new load arrives and the grid must build capacity to serve it, the cost enters the rate base. How much of it the new load pays versus how much other customers pay is decided by state utility commissions, and residential ratepayers are rarely well represented in those proceedings.
What is a coincident peak charge?
A charge based on a customer's electricity draw at the moment of the system's highest demand. It is meant to make the customers driving peak capacity needs pay for that capacity. A large, sophisticated facility can reduce its draw during predicted peak hours specifically to lower the charge — legally reducing its share of infrastructure costs without reducing its overall consumption much.
Does this make utility stocks a good investment?
It cuts both ways, which is exactly the point. Load growth is genuinely good for regulated utilities, since earnings are typically a regulated return on invested capital and new demand justifies new investment. But regulated returns are set by commissions that answer to elected officials, and elected officials respond to constituents' power bills. Load growth is the bull case; political limits on cost recovery are the bear case.
Is this a real threat to AI capital spending?
It is a real constraint on the pace. Power availability, interconnection queues and permitting are already among the binding limits on data center buildout. Political resistance to cost allocation adds a fourth. None of these stop the buildout; all of them can slow it, and a slower buildout changes the timing assumptions in a lot of valuations.
Primary sources
Data & disclaimer: This article is for educational purposes only and is not financial, tax or investment advice. Figures reflect data available as of September 1, 2026, and conditions change — always confirm current pricing, rates and rules with the provider before you act. Written by Elizabeta Dimoska. See our editorial standards and disclosure.
Comments