PJM's capacity price hit its legal ceiling. The grid still came up 6,831 megawatts short.

My power bill went up again.

I live in Pennsylvania. Same house, same thermostat, same appliances. But the bill for September was 22% higher than September a year ago. I called my utility. The woman on the phone said capacity costs had gone up. She said it twice, like that would help.

I looked it up. And what I found is one of the strangest structural problems I have seen in years.

PJM Interconnection runs the largest electric grid in America. It serves 67 million people across 13 states and Washington, D.C. Every year, PJM holds an auction where power plants bid to keep the lights on. Utilities pay the clearing price. That cost flows into your bill.

For the 2024/2025 delivery year, the auction cleared at $28.92 per megawatt-day. One year later, it cleared at $269.92. The year after that, it hit the FERC-imposed ceiling: $329.17 per megawatt-day. It has stayed at the cap for three straight auctions.

That is an eleven-fold increase in two years.

But that is not the strange part.

The strange part is that the grid fell short. PJM's most recent auction, for the 2028/2029 delivery year, came up 6,831 megawatts below its reliability target. That is enough power for roughly five million homes. It was the second year in a row the grid missed. PJM has never done that before.

So the price is at the legal maximum. And there still is not enough power. The money is not solving the problem.

The reason is data centers.

PJM's independent market monitor found that data centers drove 63% of the price increase in the 2025/2026 auction. That added $9.3 billion in capacity costs that every ratepayer across 13 states now absorbs. Over the last four auctions, data-center-driven capacity charges totaled $29.4 billion — 46% of the $63.6 billion in total charges.

I have no idea when this reverses. Goldman Sachs estimates U.S. data center power demand will grow from 31 gigawatts in 2025 to 41 gigawatts this year to 66 gigawatts in 2027. EPRI projects data centers will consume 9% to 17% of all U.S. electricity by 2030, up from about 4% today.

The hyperscalers — Amazon, Microsoft, Google, Meta — committed more than $300 billion in combined capital spending for 2026 on AI infrastructure. That money builds the demand. But the generation to feed it does not exist yet. PJM's interconnection queue holds 811 new projects representing 220 gigawatts of proposed capacity. Most are years from producing a single watt.

And here is the part that keeps me up at night. NERC, the body that watches grid reliability across North America, said in January that 13 of its 23 assessment areas face elevated or high resource adequacy risk over the next five years. Its director of reliability put it plainly: infrastructure is not being built fast enough to keep up with rising demand.

So demand is growing at the fastest pace in decades. Supply is short. Prices have hit their legal ceiling. And the companies that own the power plants — the ones everybody assumed would benefit — are losing money for shareholders. Constellation Energy, which runs the largest nuclear fleet in the country, is down 28% this year. Vistra, another major generator, is down about 15%.

The grid needs more power than it can buy. The people building the demand are spending hundreds of billions. The people who make the power are watching their stock prices fall. And the people who pay the electric bill — 67 million of them — are absorbing the cost of all of it.

That is this week's story.

When the price of keeping the lights on rises eleven-fold, somebody pays. In PJM's territory, that somebody is every household and business connected to the grid.

PJM's independent market monitor, Monitoring Analytics, found that data centers drove $29.4 billion of the $63.6 billion in total capacity charges over the last four base auctions. That is 46% of the total cost, absorbed by ratepayers who did not ask for a single GPU.

The average U.S. residential electricity rate is now 18.34 cents per kilowatt-hour as of June, up 5% year over year per the EIA. The average monthly bill sits at $142. But in PJM states like Ohio and Virginia, where data center clusters are densest, wholesale costs rose 50.3% in the first half of 2026 alone.

A Northern Virginia Technology Council analysis found that without data-center-generated tax revenue, the typical Loudoun County homeowner's property tax bill would need to rise by about $5,800 a year — 91% — to fund the same local services. The data centers pay taxes. But they also push power costs onto every neighbor.

“The price impacts on customers have been very large and are not reversible.” — Monitoring Analytics, PJM Market Monitor

PJM expected consumer bills to rise 1.5% to 5% from the 2026/2027 auction. But that estimate assumes the shortfall does not grow. It assumes no extreme weather events. And it assumes the cap holds.

I would not bet on all three.

And here is where the cost problem connects to a supply problem.

PJM reopened its interconnection queue earlier this year after a four-year backlog. The first reformed cycle drew 811 new generation projects representing 220 gigawatts of proposed capacity. Natural gas dominates the queue. So does solar. But even after PJM processed more than 60% of the backlog, the gap between approved projects and operating power plants remains wide.

The timeline is the problem. A natural gas plant takes four to six years to construct. A nuclear restart — like Constellation's $1.6 billion Three Mile Island project for Microsoft — is targeting 2028 at the earliest. Small modular reactors are further out. The demand is here now. The supply is on paper.

Meanwhile, old plants keep closing. PJM has retired over 40,000 megawatts of generation since 2011, mostly coal. Some of that capacity has been replaced. Much of it has not. The grid is losing supply from the left while demand piles in from the right.

FERC rejected one shortcut last year. Talen Energy tried to sell 480 megawatts of behind-the-meter power from its Susquehanna nuclear plant directly to an adjacent Amazon data center. FERC said no — the arrangement pulled capacity off the grid without ensuring reliability for other customers. Amazon and Talen Energy later reworked the deal as a 1,920-megawatt front-of-the-meter agreement. It is the largest corporate nuclear power purchase agreement ever signed.

So the hyperscalers want the power. They are willing to pay. But the physical grid cannot deliver it fast enough. That is a bottleneck no amount of capital spending can fix in a quarter.

Meanwhile, the market is doing something I cannot explain.

Constellation Energy operates the largest nuclear fleet in the United States. In 2024, it signed a 20-year power purchase agreement with Microsoft for the Three Mile Island restart. It signed a separate 20-year deal with Google. These are the most valuable long-term contracts in the power business. Constellation's stock is down 28% year to date.

Vistra, which owns both nuclear and natural gas plants across PJM and ERCOT, guided for $6.8 billion to $7.6 billion in adjusted EBITDA this year. The stock is down about 15% year to date. It has drifted lower every quarter of 2026, even as fundamentals improved.

The S&P 500, for context, is up about 12% over the same period.

So the country needs more electricity than it can produce. The companies that produce it have pricing power, long-term contracts, and growing demand. And the market is selling them. Rising Treasury yields are part of the answer — higher discount rates weigh on capital-intensive businesses. But the disconnect between the physical reality and the stock price is wider than I have seen in this sector.

We are watching this closely.

Three more things worth keeping track of.

FINRA margin debt hit a record $1.50 trillion in June, up 49% from a year ago. The more telling number is underneath it: net free credit balances — the cash investors keep in their brokerage accounts minus what they owe — fell to negative $1.06 trillion, also a record. That means investors as a group have borrowed more than a trillion dollars more than they hold in cash. The last time margin debt set nominal records was late 2021. The S&P 500 peaked two months later and fell 25%. We are not calling a top. But the cushion is gone.

South Korea's household credit crossed 2,000 trillion won — about $1.41 trillion — for the first time in Q2 2026, driven by mortgage lending and stock-market borrowing. Average debt per borrower reached a record 97.90 million won ($73,000). May household loan growth rose from 2.1 trillion won to 6.9 trillion won, the fastest monthly increase since August 2024. South Korea's household-debt-to-GDP ratio sits among the highest in the developed world. Rate cuts could stabilize this. Rate hikes to defend the won could break it.

The S&P 500 equal-weight index sat about 5.5% below its 52-week high as of October 7, while the cap-weighted index was within 1% of its own. That gap tells you the average stock is not doing what the index says it is doing. Roughly 47% of S&P 500 index weight reports earnings in the five sessions from October 28 through November 3. If the mega-caps miss, the headline index catches down to where the average stock already sits. We'll see.

In 1996, California deregulated its electricity market.

The idea was simple. Let generators compete. Wholesale prices will fall. Consumers win. Governor Pete Wilson signed Assembly Bill 1890, and the spot market opened in April 1998.

For two years it worked. Then demand grew faster than supply. California had 45 gigawatts of installed capacity, but not enough of it ran on the days the state needed it most. Wholesale prices rose 800% between April and December 2000.

But retail prices stayed fixed. The state's two largest utilities — Pacific Gas & Electric and Southern California Edison — bought power at the market rate and sold it to customers at the regulated rate. The gap grew wider every month.

By early 2001, wholesale power cost more than $300 per megawatt-hour. The utilities sold it for a fraction of that.

On June 14, 2000, blackouts hit 97,000 customers in the San Francisco Bay Area. Rolling blackouts spread across the state through the winter. The grid operator called emergency after emergency.

On April 6, 2001, PG&E filed for Chapter 11. It owed billions. The company that kept the lights on for 13 million people could not pay for the electricity it delivered.

It took PG&E three years to exit bankruptcy. It took California a decade to unwind the contracts the state signed during the crisis to keep power flowing. Ratepayers absorbed the cost through surcharges that stayed on bills for years after the lights came back on.

The pattern was straightforward. Demand outgrew supply. Wholesale prices spiked. Retail prices stayed low. The gap broke the utilities in the middle.

I am not saying PJM is California in 2000. The market structure is different. The price cap is a negotiated collar, not a hard freeze on retail rates. And there is no evidence of the kind of market manipulation that Enron ran during the California crisis.

But the shape rhymes. Demand is growing faster than generation. The price mechanism is capped and not producing enough supply. The shortfall is growing. And 67 million people are on the other end of that wire. We'll see.