PJM's capacity price rose 1,038% in two years. Data centers caused 63% of it.

I got my electricity bill last month. It was $22 higher than the same month a year ago. I hadn't changed anything. Same house. Same thermostat. Same two window units running too hard in September.

So I looked into it.

The answer has nothing to do with natural gas prices or summer heat or the usual explanations. It has to do with a market most people have never heard of… and the companies driving it aren't utilities. They're tech companies. Building data centers. For AI.

PJM Interconnection runs the largest wholesale electricity grid in North America. It covers 13 states and Washington, D.C. It serves 67 million people. Every year, PJM holds a capacity auction. Generators bid to keep the lights on for the coming year. The clearing price sets what everyone pays.

Two years ago, the clearing price was $28.92 per megawatt-day.

This year, it hit $329.17 per megawatt-day. That's an increase of more than 1,000%.

It hit the maximum price allowed by federal regulators. And it wasn't enough. The most recent auction, for 2028–2029, came in 6,831 megawatts short of what the grid needs. That was the third year in a row at the cap. PJM has never fallen short like this before.

PJM's own analysis found that data centers caused 63% of the price increase. The driver wasn't weather or coal plant retirements. It was data centers.

The NRDC estimates this will cost ratepayers $100 billion to $163 billion through 2033. The average family in PJM territory faces a $70-per-month increase by 2028.

In Virginia, the state that runs 70% of the world's internet traffic through Loudoun County, regulators already approved a $16-per-month increase for residential customers for this year. Most of the grid upgrades are for data centers.

But the data center companies aren't paying for it. You are.

That's the story this week. The AI boom is now a line item on your electricity bill. And the grid can't keep up.

The companies driving that grid strain are spending at a pace I've never seen in any industry.

Microsoft, Meta, Amazon, and Alphabet told investors they will spend roughly $700 billion on capital expenditures in 2026. That's nearly double what they spent in 2025. Bank of America expects the number to pass $1 trillion in 2027.

But spending money and getting power are two different things.

About 12 gigawatts of data center capacity was planned for 2026 across the United States. Nearly half of it—roughly 7 gigawatts—has been delayed or canceled. The reason is the same everywhere: no power. Utility interconnection queues in Virginia, Texas, and Georgia are measured in years, not months. Projects that got their land and permits in 2024 are sitting idle, waiting for a power agreement.

Gartner forecasts worldwide data center power demand will reach 132 gigawatts in 2026. BloombergNEF sees it hitting 106 gigawatts in the U.S. alone by 2035. The grid was not built for this.

They committed $700 billion. The grid can deliver maybe half.

Meanwhile, the Philadelphia Semiconductor Index fell for eight straight sessions through October 16—its longest losing streak since 2008. The VanEck Semiconductor ETF dropped 10%. And Vistra, the power company that spent the last two years riding the data center trade, fell 9.8% in a single day on October 22 after Meta's earnings reminded investors that AI spending and AI profitability are not the same thing.

The money is moving fast. The electrons are not.

And here's where the strain moves from the grid to the balance sheet.

CoreWeave is a cloud computing company that rents Nvidia GPUs to AI developers. It had roughly $52 billion in debt and lease liabilities on its balance sheet at the end of the second quarter. That's up from $30 billion at the end of 2025. The company added $22 billion in obligations in six months.

The bond market noticed. CoreWeave's October 2031 bonds fell to 79 cents on the dollar. Yields on those bonds briefly passed 16%. Its five-year credit default swap spread hit a record 1,195 basis points. In plain terms: insuring $1 million of CoreWeave's debt costs $119,500 a year. That's distress territory.

The stock fell more than 25% in October. It hit an intraday low of $59.51 on October 15.

CoreWeave says it has $103.7 billion in remaining performance obligations—contracts from customers like Microsoft. So the revenue is coming. But the debt is here now. And the data centers that debt was supposed to build need power that the grid can't yet provide.

Barclays called the comparison to a broader AI debt crisis overblown. I have no idea if they're right. But when a company carries $52 billion in debt and its bonds trade at 79 cents, I pay attention.

Three more things worth keeping track of this week.

Money market fund inflows fell off a cliff. In the first three quarters of 2026, investors added $158 billion to money market funds. In all of 2025, they added $823 billion. In 2024, $840 billion. Total money fund assets reached $7.98 trillion by early September—still a record—but the stream of new cash has slowed to a trickle. Three-month Treasury bill yields rose nearly 10 basis points above overnight index swaps earlier this month, after hitting the widest gap since September 2024 the week before. Barclays expects $225 billion in new bill issuance in October alone. If the cash doesn't show up, bill yields keep climbing. And if bill yields keep climbing, it puts pressure on every rate in the short end.

Investment-grade corporate bond issuance is on pace for a record year. Morgan Stanley projects issuance could pass $2 trillion in 2026, driven by AI-related capital spending, refinancing of older debt, and acquisition financing. On the first Monday of 2026, companies sold $37 billion in high-grade bonds. The rush makes sense: lock in rates before they go higher. But the scale of new supply means the bond market has to absorb more paper than ever, right as the buyers are slowing down. That's the kind of mismatch I track.

U.S. banks are still sitting on $326.7 billion in unrealized losses on their bond portfolios, per the FDIC's second-quarter report. That number rose $1.6 billion from the prior quarter. It's down from the peak, but it represents 5.5% of amortized cost and it's not going away as long as rates stay here. I covered the bank angle in prior issues. It hasn't improved. We'll see.

In September 1996, California's governor signed Assembly Bill 1890. It deregulated the state's electricity market. The idea was simple: let generators compete, and prices would fall.

The bill froze retail rates for residential customers. Utilities like Pacific Gas & Electric and Southern California Edison had to buy power on the new wholesale market. But they couldn't raise prices to cover the cost. If wholesale prices stayed low, the utilities would make money on the spread. Everyone wins.

Wholesale prices did not stay low.

Between April and December 2000, wholesale electricity prices in California rose roughly 800%. The state's total electricity costs jumped from $7 billion in 1999 to $27 billion in 2000, per the California Independent System Operator. Enron and other energy traders gamed the market, creating artificial shortages. Rolling blackouts hit millions of people. California had 45 gigawatts of installed capacity. Demand was only 28 gigawatts. The shortage was manufactured.

PG&E spent $9 billion more buying electricity than it collected in rates. On April 6, 2001, it filed for bankruptcy.

The frozen retail rates meant utilities absorbed the entire gap between what power cost and what customers paid. PG&E spent $9 billion more buying electricity than it collected. It ran out of money. On April 6, 2001, it filed for Chapter 11. Southern California Edison came within days of following.

The total excess cost to California consumers and taxpayers reached $40 billion to $45 billion, per the Public Policy Institute of California. Governor Gray Davis was recalled from office in October 2003. The electricity crisis was a major reason.

I keep thinking about the structure of that story.

A deregulated market. A surge in demand from a source the system wasn't designed for. A pricing mechanism that passes costs to the wrong people. And a gap between what power costs and what the people paying for it expect.

In California, the demand surge came from market manipulation. In PJM today, it comes from data centers. The cause is different. But the structure is the same: wholesale costs are rising at a pace the retail side of the system was never built to absorb.

Virginia legislators are now writing bills to shift costs from households to data centers. States are proposing moratoriums on new construction. PJM is running an emergency backstop auction it has never needed before.

I don't know how this ends. I do know that the last time a U.S. electricity market saw prices rise 800% while the cost allocation system stayed frozen, the largest utility in the state went bankrupt and the governor lost his job.

We'll see.