Utility rate structures, federal transmission rules, and regional grid markets decide who pays for the power grid expansion needed by large AI data centers—and Harvard's Electricity Law Initiative says households are ending up on the hook.
When a utility builds a new substation or transmission line to serve a hyperscale data center, the bill for that build doesn't automatically land on the data center. Under the rules most US utilities operate under, the cost flows into the rate base (the pool of infrastructure spending regulators allow a utility to recover from customers) and is spread across all ratepayers through a process Harvard's Electricity Law Initiative (ELI) calls cost socialization. For a residential customer in Louisiana or Virginia, that means a line item on the monthly bill with no label and no connection to whether they use ChatGPT.
A March 2025 brief from Harvard ELI calls this the mechanism by which "Big Tech" data centers are paid for by ordinary ratepayers. A follow-up CleanLaw explainer from the same project is more direct: the people paying for AI's electricity are probably you.
Ari Peskoe directs the Harvard ELI project and has testified on utility cost-allocation questions before state public utility commissions. In a video interview and a Utility Dive op-ed, he argues that the real story behind climbing US electric bills is not that AI uses a lot of power, but that utilities and grid planners are socializing the cost of serving data centers onto ordinary ratepayers through three specific channels.
The first is transmission. Under FERC's Order 1000, the federal rule that governs how regional grid planners pick and pay for new high-voltage lines, the cost of a line built to serve a data center cluster is allocated to the utilities whose customers benefit. In practice, Harvard ELI argues, those costs spread beyond the immediate industrial customers because transmission pricing averages across the regional grid. A residential customer in PJM, the regional grid operator covering 13 mid-Atlantic and Midwest states, can end up paying a share of a line built specifically for a Virginia data center campus.
The second is capacity markets. These are forward auctions, run by regional grid operators like PJM, in which power plants are paid to be available to meet projected peak demand several years out. A Union of Concerned Scientists analysis found billions of dollars in data-center-driven capacity costs flowing to PJM customers without line-item disclosure. Peskoe's op-ed calls out an "outdated FERC policy" that prevents those costs from being assigned to the load causing them.
The third is rate-class allocation at the state level. State public utility commissions decide which customer class, residential, commercial, or industrial, pays which slice of the build. When a hyperscaler negotiates a special tariff (a rate schedule with custom pricing for a specific large customer) with below-cost rates in exchange for siting in the state, Harvard Law Today reports, the fixed costs the hyperscaler would otherwise pay get redistributed to remaining customers through higher base rates.
PBS NewsHour and the Christian Science Monitor reported in August 2025 that data-center load growth is now a structural driver of US retail electricity prices, alongside aging-grid and fuel-cost factors. Harvard Magazine, citing the same Harvard ELI work, framed the question in July 2025 as whether AI compute buildout is putting upward pressure on retail electricity costs. Latitude Media's Open Circuit podcast treats the cost-allocation question as a live grid-policy debate rather than fringe speculation.
The abstraction turned concrete this year in Louisiana, where a single data center facility under construction is expected to draw several times the load of the city of New Orleans. The transmission build for that campus illustrates exactly the cost-allocation mechanism ELI has named.
The fights over who pays are happening in three venues. State public utility commissions in Louisiana, Virginia, Georgia, Ohio, and Texas have live rate cases involving data-center load, per the Harvard ELI brief. FERC's transmission-cost-allocation rule is the lever Peskoe's op-ed targets as most likely to change the math. PJM's capacity-market rules face scrutiny over whether data-center load should be procured separately from retail ratepayer load.
The next concrete milestone is the FERC proceeding on transmission cost allocation, expected to advance in 2026. Whether that rule changes will determine whether the next hyperscale data center pays its own grid buildout, or whether the cost lands on the residential bill with no label.