Your Power Bill Has an AI Line Item. Here Is Where It Hides.

There is an auction almost no household has heard of. Its clearing price rose eleven-fold in two years, data centres caused most of the increase, and it arrives on your bill as a number nobody itemises. The same squeeze is reshaping which companies earn from electricity and which quietly absorb it.

HeyTheo Research · Friday, October 2, 2026

Quick Read

• A capacity auction you have never seen set part of your bill. PJM's clearing price went from $28.92 per megawatt-day for 2024/25 to $269.92 for 2025/26 and $329.17 for 2026/27, an eleven-fold rise in two years that hit the market's price cap.

• Data centres caused 63% of one year's increase. That single auction pushed about $9.3 billion of capacity cost onto customers across the region, with roughly $1.4 billion more in the following year.

• Households are already seeing it. Monthly increases of about $21 in Washington DC, $18 in western Maryland and $16 in Ohio, with roughly half of the DC figure traced to capacity pricing alone.

• The supply response is blocked. One turbine maker alone has around 100 gigawatts under contract against annual output near 10 gigawatts, with lead times of five to seven years.

• The part nobody says out loud. Utilities underwrite generation at a 6% to 8% cost of capital. Hyperscalers can absorb a 10% to 20% premium per kilowatt. When a turbine slot opens, the data centre outbids the utility that serves your street.

For about twenty years, American electricity demand did almost nothing.

From the mid-2000s to the early 2020s, consumption was essentially flat. Efficiency gains cancelled out population and economic growth. Utilities planned for a world that did not grow, regulators approved capital plans that assumed the same, and very little new generation got built because very little was needed.

That assumption broke. Consumption has been rising since 2020 and is forecast to pass its all-time high. The growth is not evenly spread: the commercial sector, which is where data centres sit, is growing at roughly 2.6% a year, industrial at 2.1%, and residential at only 0.7%.

So the people using more are not the people in houses. But the pricing mechanism does not care who caused the demand.

The number on your bill that has no name

Source: PJM capacity auction results as reported in grid and consumer advocacy analyses, 2025 and 2026.

A capacity auction is not a market for electricity. It is a market for the promise to be available. Grid operators pay generators to guarantee they can switch on during the worst hour of the year, and that payment is recovered from everyone connected to the system.

Almost no residential customer has heard of it, and it does not appear as a line on a bill. It is folded into the supply or delivery charge.

Its price rose eleven-fold in two years and cleared at the market's cap.

Analysis of the 2025/26 auction attributed 63% of the increase to data centre demand, equal to roughly $9.3 billion recovered from customers across the region, with about $1.4 billion more coming from the following auction. In the territory with the densest data centre build, the long-range load forecast moved from roughly 5,700 megawatts of total growth projected in 2022 to more than 20,000 megawatts from data centres alone in the 2025 forecast.

Households have felt it already. Reported monthly increases run to about $21 in Washington DC, $18 in western Maryland and $16 in Ohio, with roughly half of the DC increase traced to capacity prices specifically.

Why the obvious fix is not available

When a price rises this fast, the textbook answer is that supply responds. Build more generation, the scarcity eases, the price falls.

That response is physically blocked, and the reason is a machine most investors never think about.

Large gas turbines are made at scale by three companies, which between them account for roughly two-thirds of global capacity under construction. One of them alone had about 100 gigawatts under contract by early 2026, split between firm backlog and reserved production slots, against annual manufacturing capacity near 10 gigawatts. That is close to a decade of output already spoken for, and management guided the figure higher still by year end.

Quoted lead times now run five to seven years. The window in which a new buyer can realistically take delivery moved out from 2029 to 2030 and beyond during the first half of 2026. Pricing on new orders in the first half of 2026 ran 10 to 20 percentage points higher per kilowatt than orders booked at the end of 2025.

One manufacturer has publicly pushed back on the idea that turbines are gating data centre construction, and that view deserves recording. But the order book is not in dispute, and neither are the lead times.

So the squeeze cannot be built away before the end of this decade. Whatever the price does between now and then is a rationing outcome, not a supply outcome.

The asymmetry nobody says out loud

Here is the part that changes how you read all of it.

A regulated utility underwrites new generation at a weighted average cost of capital of roughly 6% to 8%. It has to earn a return its regulator will approve, recovered from a customer base that has to agree to pay.

A hyperscaler building AI capacity does not face that test. The revenue it expects from the compute is large enough that it can absorb a 10% to 20% premium per kilowatt on the turbine and still clear its own hurdle. It reserves slots with non-refundable payments.

Roughly a fifth of that 100 gigawatt order book is tied explicitly to data centre load.

Put plainly: when a production slot becomes available, the data centre outbids the utility that serves your street. Not because of a policy choice, but because the two buyers are discounting the same machine at different rates. The household ends up paying for scarcity it did not create and cannot outbid.

And interest rates make this worse rather than better. A utility's cost of capital rises with the rate environment, and the 10-year Treasury sat at 5.29% on October 1, near a 24-year high. That widens the gap between what a utility can justify paying and what a hyperscaler can, and it also raises the return a regulator must allow on new utility investment, which is then recovered through rates.

The policy tool aimed at inflation makes this particular component of inflation more expensive. Electricity in the Northeast rose 4.1% over the twelve months to August 2026 and energy services rose 5.5%, against an all-items regional figure of 3.9%.

Who earns this and who absorbs it

One commodity, two opposite effects, and both sides are investable.

The earners. Regulated utilities grow earnings by investing capital in the system and recovering an allowed return on it. A step change in required investment is, mechanically, a step change in the asset base they earn on. Independent power producers with existing capacity sit on assets whose scarcity value has repriced sharply upward.

The absorbers. Anyone who uses a lot of electricity and cannot pass the cost on. Heavy manufacturing, industrial processing, cold chain and logistics, and the consumer businesses whose customers are already paying $16 to $21 more a month and have less left over.

The ambiguous middle. Data centre operators and the companies building AI capacity are paying the premium knowingly. Whether that shows up as a margin problem depends entirely on whether the compute revenue arrives on schedule, which is the same open question our earlier note on the cost of the AI buildout set out.

The useful discipline here is not to pick a winner from a headline. It is to find out which companies you already hold have an electricity cost line big enough to matter, and whether they have said anything about it.

The rules HeyTheo tracks

• Basket: hold the regulated utilities and the heavy electricity users as two sides of one position. The same capacity price is revenue on one side and cost of goods sold on the other.

• Triggers: flag each capacity auction result, utility rate case filings and approved allowed returns, quarterly utility capital expenditure guidance, and any earnings call where a company names power costs as a margin factor.

• Money flow: watch whether utility sector strength holds when long rates rise. Utilities are rate-sensitive and capital-hungry at once, and those two pull in opposite directions.

• Ask Theo: three questions worth running on your own watchlist rather than reading someone else's conclusion:

  • Which of my holdings have the highest energy cost relative to revenue?

  • Show me the capital expenditure trend and interest expense for this utility over eight quarters.

  • Which companies I follow mentioned electricity or power costs on the last earnings call?

• Check the rule behind any trigger before acting on it. You trade through your own broker; HeyTheo helps you decide.

The Meter Nobody Reads

The story most people have heard is that AI needs a lot of electricity. That is true and it is not very useful.

The useful version is narrower. A market for standby capacity that no household has heard of repriced eleven-fold in two years. Most of that increase traces to one category of customer. The money is collected from everyone. The supply response that would normally end it is sold out until the next decade, because three manufacturers make the machine and the buyers with the best economics have already reserved the output.

And the one policy lever being pulled right now, higher interest rates, widens the gap between who can afford to build generation and who cannot.

None of that is a forecast. It has already happened, it is in auction results and order books, and it is arriving on bills in the single digits of dollars per month and in the margin lines of companies that have not drawn attention to it yet.

The question worth carrying is not whether electricity gets more expensive. It is which of the things you own quietly pays for it.

FAQs

What is a capacity auction and why does it affect my bill?

It is a market where a grid operator buys promises of availability rather than electricity itself. Generators are paid to guarantee they can run during the highest demand hours of the year, and that cost is recovered from everyone connected to the grid. It does not appear as its own line on a residential bill, which is why an eleven-fold price increase can happen without most customers knowing the market exists.

Why can't utilities just build more power plants?

They can order them, but they cannot get them quickly. Large gas turbines are built at scale by three manufacturers, and one of them alone has roughly a decade of production already committed. Quoted lead times run five to seven years. A utility that decides today to add gas generation is planning for the 2030s, which means the current tightness cannot be built away within this rate cycle.

How are data centres able to outbid regulated utilities?

Cost of capital. A utility underwrites generation at roughly a 6% to 8% weighted average cost of capital and has to justify the spend to a regulator. A hyperscaler expecting AI compute revenue can absorb a 10% to 20% premium per kilowatt and still meet its own return threshold, and can reserve manufacturing slots with non-refundable payments. The same machine is simply worth more to one buyer than the other.

Does a rising interest rate environment help or hurt this?

It makes it worse on both sides. Higher rates raise a utility's cost of capital, which widens the gap against buyers who are less rate-sensitive, and they raise the return a regulator has to allow on new utility investment, which is recovered through customer rates. A policy tool aimed at reducing inflation increases the cost of this specific component of it.

Is this a reason to buy utility shares?

It is a reason to understand how utilities earn, which is by investing capital and recovering an approved return on it, so a large increase in required investment expands the base they earn on. That is a mechanism, not a recommendation, and it comes with real offsets: utilities carry heavy debt, so a high rate environment raises their financing costs, and regulators can decline to approve what companies propose. The work is name by name, not sector-wide.

Sources

Sources: US Energy Information Administration, PJM capacity auction results, IEEFA, Citizens Utility Board, Bureau of Labor Statistics, GE Vernova disclosures, Manufacturing Magazine, TechCrunch, CNBC.

Disclaimer

Disclaimer: HeyTheo is a research and education platform, not an investment adviser or broker-dealer. Nothing here is advice to buy, sell, or hold any security. You trade through your own broker; HeyTheo helps you decide. Backtested results are hypothetical and do not guarantee future returns. References to governments, officials, or policies are for market context only and are not political endorsements. All investing involves risk, including loss of principal. Data is as of the dates noted.