Electricity prices rose 5.9% year over year in May, and the forecast tied to that number now runs three years out: Goldman Sachs projects household power bills climbing about 6% in 2026, another 6% in 2027, and 3% in 2028 as utilities build generation to serve data-center demand, according to AP [1]. The framing matters because it moves the AI-inflation story off the single quarter and onto a multi-year trajectory that ordinary ratepayers are already inside.

AP supplies a bounded national series and a forecast without pinning every increase to one campus — the price line rises, but the outlet does not claim a named project is charging a named household. That restraint is exactly where the social conversation refuses to sit. On X, promoters treat the same generation buildout as infrastructure abundance, a grid getting bigger; opponents treat each bill increase as direct AI capture, a transfer from families to server farms. Both sides collapse the national average into a verdict AP declines to render.

The gap between them is the missing mechanism. Who actually pays for a new gas plant or transmission upgrade is decided in utility cost-allocation rules and state rate cases, and it varies by regional load — not by a headline percentage. A national forecast establishes that bills will rise; it does not establish that a specific data center caused a specific line item. Until a rate case names the allocation, both the abundance story and the capture story are running ahead of the record.

-- Dara Osei, London