Summary
Highlights
On July 2nd, the PJM grid, which serves 67 million people across 13 states and Washington D.C., experienced record-high electricity demand of 168,158 MW during a brutal heatwave. This surpassed the previous record from 2006, leading to a near collapse of the system.
To prevent a blackout, the US Department of Energy issued emergency orders. One order authorized PJM to force data centers and other large power users to disconnect and switch to backup generators. The second order waived pollution limits for 51 aging power plants, allowing 17 of them to run above normal limits for over 1,100 hours to keep the grid operational.
At the peak of demand, 18,363 megawatts of generating capacity were offline due to forced outages, representing a significant portion of PJM's available power. The grid survived by shedding load from large customers and heavily utilizing decades-old power plants, indicating a far more fragile situation than commonly reported.
The AI boom, with tech companies spending over $600 billion on data centers by 2026, is putting immense strain on an outdated electrical grid. Data centers are being built faster than the necessary power plants, substations, and transmission lines, leading to potential clashes where the grid operator must physically unplug facilities.
Regional utilities are significantly increasing borrowing to fund grid upgrades, with over $150 billion in new bonds expected this year. This rising debt, competing with other corporate and government borrowing, contributes to higher investment-grade yields. Ultimately, these costs will be passed on to consumers through higher electricity bills, regardless of personal AI usage.
Every new data center or significant weather event now introduces a grid reliability question. The AI boom, while appearing as a clean software story on Wall Street, has a physical ceiling in the form of the power grid, and its true cost will manifest in utility bills and bond portfolios before it appears in hyperscaler earnings.