Are AI Datacenters Raising Your Electric Bill?

The political story is that AI datacenters are driving up household electricity bills. SemiAnalysis argues the real culprit is market design. The numbers they lay out are specific. PJM's capacity auction price jumped from 29 to 270 dollars per MW-day, about a 9.3x increase, with roughly 16 billion dollars in capacity payments for the 2026/27 period, which works out to an estimated 25 to 30 dollars a month on a typical household bill. Their argument is that PJM's auction, and the simulation-based curve that sets the requirement, is extremely sensitive to forecast inputs. They point out PJM had to cut its own datacenter load forecast by 800 MW in 2024 and 1.1 GW in 2025, so the price shock ran ahead of actual demand. The comparison that carries the argument: ERCOT in Texas saw similar AI datacenter growth but only 11 to 17 percent forward-price increases, using real-time scarcity pricing instead of a capacity auction. The free portion of the piece covers the full argument and all the key numbers, with the paywall cutting before the winners-and-losers section.

Why it matters

If you work on datacenter siting, energy procurement, or policy, this reassigns blame with numbers: the same AI load produced very different bill outcomes under two market designs. The lever is auction structure, not whether the datacenters get built, which is where the 9.3x actually came from.