On July 14, PJM Interconnection — the largest grid operator in the U.S., serving roughly 67 million people from Illinois to Virginia and Washington, D.C. — announced its capacity auction for 2028-29 cleared at $325 per megawatt-day, the maximum allowed under its price cap, even as supply fell about 6.8 gigawatts short of what the grid needs to stay reliable.

It’s the clearest sign yet that America’s AI buildout has a bill attached, and increasingly, ordinary ratepayers are the ones paying it.

Moody’s Ratings, in a July 22 sector report, didn’t mince words about why: “the current system lacks adequate mechanisms to ensure that the cost of building new supply is borne by the new entrants and instead socializes new build costs across all customers.”

In plainer terms, when a data center activates hundreds of megawatts of new AI computing load, the cost of building power plants to serve it doesn’t land on a hyperscaler’s bill — it gets spread across every household and small business on the grid.

What makes the finding notable is the source. Moody’s exists to price risk dispassionately for bond investors, not to editorialize about fairness. Yet the agency explicitly flagged that “other power markets in the U.S. require new large-load customers to secure power under direct supply contracts that recoup cost of new generation and associated infrastructure over time,” implying PJM’s rules are, by comparison, out of step.

The numbers behind the squeeze

Total new generation capacity that cleared this auction fell to just 525 megawatts, roughly half of what cleared six months earlier, according to Syso Technologies’ auction analysis, despite PJM seeing a peak electricity demand record high of 168.2 gigawatts on July 2, nearly 3 GW above a record set almost two decades earlier.

Data centers are the primary driver of that demand growth, according to PJM’s own market monitor, Monitoring Analytics, which found that of the $16.4 billion in total capacity charges from this auction, about $6.3 billion is directly attributable to data center demand — and $29.4 billion over the last four auctions combined, per The Hill.

Without the price cap in place, PJM’s simulation shows the region’s unconstrained capacity price would have hit $554.72/MW-day — and $776.69/MW-day in the Chicago-area zone served by Exelon’s ComEd utility — underscoring that scarcity stress extends well beyond any single state. This is playing out well beyond spreadsheets. Consumer Reports has documented individual cases, including an Ohio resident whose bill hit $281 this past January.

Why utilities say one thing and bills say another

But utilities are promising investors and regulators that ratepayers won’t be stuck with AI’s tab, even as watchdogs point to cases shifting exactly those costs onto residential bills. Harvard Law’s Electricity Law Initiative has identified two distinct cost channels: utilities spreading new infrastructure costs across all ratepayers, and market-based capacity prices rising simply because supply hasn’t kept pace with data center-driven demand. Both mechanisms are visible in PJM’s July auction.

Separately, Reuters reported that households and businesses in PJM territory could face rate hikes of up to 60% over the next five years as Big Tech’s data center buildout accelerates, citing projections by the global consulting and technology services firm ICF.

The backlash has reached multiple state legislatures, where lawmakers are moving against what consumer advocates call outsized utility profits layered on top of rising bills. Meanwhile, PJM is asking the Federal Energy Regulatory Commission for permission to hold an emergency “backstop” capacity auction in September — an unusual step that amounts to an acknowledgment that the normal market isn’t producing enough new power fast enough.

Some analysts have proposed requiring hyperscaler data center operators to sign long-term contracts for new generation, an idea explored in Latitude Media’s coverage of a Mid-Atlantic governors’ proposal, though critics warn it may not close the gap fast enough and risks distorting the broader market. But adoption remains uneven, and PJM still lacks the direct-contract requirement that other markets already use.

For now, Moody’s technical language and the swelling monthly bills of PJM’s tens of millions of residential customers are describing the same phenomenon from opposite ends of the transaction: someone is paying for the AI revolution’s power hunger, and increasingly, it isn’t the companies building it.

For this story, Fortune journalists used generative AI as a research tool. An editor verified the accuracy of the information before publishing.