AI Needs a BYOG Model to Keep Power Bills Down

PJM is the name given to the largest regional US electricity grid, stretching from Chicago to the Chesapeake. It is also synonymous with the AI-power surge, being home to Virginia’s ‘datacenter alley’ and seeing recent big increases in bills tied to concerns that the electricity needs of artificial intelligence will swamp supply.

PJM has just undergone a week of whiplash, beginning with a cut to its projections of electricity demand — which you might think would ease those concerns — and ending with that reliable indicator of bubbling angst, a White House plan, followed by what looks like an actual plan from the grid operator itself.

If the lowered projections suggested the AI boom has been deferred, the new proposals set forth by President Donald Trump and several state governors, plus PJM’s own set, demonstrate the political storm around that boom has merely gathered strength. A dash by Big Tech to commandeer the existing grid for its own ends seems to be giving way to a new era where it must bring its own generation.

The AI-influenced bull market in electricity, and the stocks of companies generating it, owes much to recent increases in PJM’s peak demand forecasts after years of flatlining. The latest projections, however, pulled back — kind of. Forecasts for peak demand through 2032 were lowered slightly, while those further out were raised, and by much more.

BB Pushing back

The AI-power boom is haunted by three interrelated ‘E’s. The first two are efficiency and exuberance. More efficient datacenters would undercut bullish demand forecasts, while excess enthusiasm tees up excess supply — for which not just investors but also households can end up paying. This has happened before, in the early 2000s, and a more recent warning arrived last year with the DeepSeek freakout. Moreover, analysts at CreditSights calculate that US grid-connection requests are more than double the consensus forecast for actual datacenter demand, suggesting a bubble is forming.

PJM’s cut would appear to reinforce such concerns. but the truth is that no-one knows the scope and timing of AI’s impact for sure.

Meanwhile, there are clear signs of tight supply and demand. After all, PJM’s near-term cuts to peak demand forecasts, maxing out at about 4.5 gigawatts, are less than the shortfall of about 6.5 gigawatts that featured in the last capacity auction — when prices hit their maximum allowed level. (This auction sets a price paid to power plants for committing to have capacity available during a certain time period, like an insurance premium). In addition, electricity prices in the PJM region are set to continue rising this year and next, according to the Department of Energy’s own short-term energy outlook, also out last week.