Data Centers Add Little to Home Power Costs, Study Finds

Data centers accounted for about 4% of the increase in residential electricity prices between 2020 and 2025 at utilities examined in a new working paper, suggesting their contribution to rising household power costs has been smaller than the public debate often implies.

The researchers estimate that new facilities added about $1.70 per megawatt-hour to residential electricity prices at the average utility studied. That amounts to 0.17 cents per kilowatt-hour, or roughly $1.70 a month for a household using 1,000 kilowatt-hours, holding consumption constant.

The effect was larger for industrial customers. Data center additions increased their average electricity price by about $5.40 per megawatt-hour, accounting for nearly 16% of the observed increase over the period.

The findings come from *The Cost of Intelligence: Impacts of AI Data Centers on US Electricity Prices*, by Vishrut Rana and Serguei Netessine of the University of Pennsylvania’s Wharton School and Christian Kaps of Harvard Business School. The working paper was posted on SSRN in October and reported by The Washington Post on Friday.

The estimates concern the contribution of additional data centers to price increases at the utilities studied. They do not mean household electricity prices rose just 4%, nor do they measure the costs of the entire data center industry.

The estimated residential effect was equivalent to about 1.4% of the average 2020 residential electricity price. Most of the subsequent increase remained unexplained by data center additions in the analysis.

Who builds—and where—matters

The researchers linked facility locations, opening dates and power capacity with federal electricity-sales, revenue and utility-cost records. Their main sample covered 385 large data centers added across 42 utilities, with 41 utilities included in the residential-price analysis.

For each utility receiving new facilities, they constructed a comparison from a weighted combination of other utilities, matching four years of preceding outcomes. They then tracked how prices diverged as additional data center capacity accumulated.

That approach attempts to address a central complication: developers choose where to build. Utilities hosting data centers already had lower electricity prices before the recent expansion, making a simple comparison with other utilities potentially misleading.

The residential results also differed by developer. Facilities owned by Amazon, Google, Meta and Microsoft showed no statistically discernible residential price effect in the preferred specification. The positive effect was concentrated among the remaining developers, which the paper groups as third-party providers.

That finding does not establish that Big Tech facilities impose no costs on households. The estimated difference between the two developer groups was itself statistically weak, and the researchers could not separate the influence of location choices from electricity procurement and infrastructure-financing arrangements.

Utilities’ experience also mattered. Systems with a larger existing industrial customer base generally showed smaller residential spillovers. Utilities operating at higher average utilization also tended to have smaller effects, suggesting that apparently unused capacity does not necessarily translate into infrastructure suited to a large, continuous load.

Following the electricity bill

The findings speak directly to the questions examined in TheEnergyMag Explainer Series: how new electricity demand changes grid costs, and how those costs reach household bills.

As Part 2 explains, a large customer can help spread the fixed costs of existing power plants and infrastructure across more electricity sales. But where new demand requires additional generation, transmission or substations, the benefit depends on whether the customer’s payments cover the added expense.

The paper’s framework reaches a similar distinction. More electricity sales can reduce average costs even when a utility’s total spending increases. Household outcomes also depend on how regulators allocate those costs among customer groups.

The results add to a research record that remains mixed. A separate working paper examining data center entry from 2010 through 2024, estimated a 2.1% increase in residential retail rates. Differences in samples, periods and methods mean that estimate is not directly comparable with the new paper’s 4% share of price growth.

Historical averages have limits

The new study ends in 2025 and does not forecast the consequences of the next wave of AI infrastructure. Its sample is restricted to utilities providing bundled electricity and delivery service, and it measures data center expansion rather than isolating electricity used for AI.

Its preferred residential estimate also meets a 10% statistical-significance threshold, but not the more commonly used 5% threshold. The authors acknowledge that unobserved utility-specific shocks and the limited number of utilities could affect their conclusions.

There are signs within the paper that recent developments may impose greater pressure. Restricting the analysis to additions from 2022 onward produces a larger residential price effect for an equivalent increase in capacity relative to utility peak demand.

Regional supply constraints also remain material. On July 14, PJM reported that its auction for the 2028–2029 delivery year cleared at the regulatory price ceiling while committed resources fell 6,831 megawatts short of its reliability requirement. Capacity payments compensate resources for being available and form one component of electricity costs. The shortfall indicated thinner reserves, rather than inevitable blackouts.

For regulators weighing new projects, the paper points toward examining the receiving utility and its cost-recovery arrangements. A modest historical average leaves open the question facing each community: whether the next facility’s payments will cover the cost of serving it.