In the past five years, the United States began experiencing a rapid increase in electricity demand, fueled primarily by data centers for artificial intelligence. A single data center can use the amount of electricity consumed by a city of approximately 80,000 people. Most data center companies seek electricity from the same utilities that provide electricity to retail and commercial customers (including all of us). And these utilities are building massive amounts of transmission and generation to meet data centers’ growing demands.
In many cases, though, utilities are passing on the costs of new data center infrastructure to all of us, rather than ensuring that data centers pay for the costs that they are imposing. The Union of Concerned Scientists estimates that within just one regional electric transmission grid called PJM, in 2024 alone, all utility customers paid $4.4 billion in costs incurred by utilities in seven states to build new transmission for data centers. And PJM, which runs auctions to ensure that there is enough generation infrastructure to meet demand, has overseen auctions that generated utility and generator revenues in the billions of dollars — nearly all of which funded new generation for data centers.
In our new report, we explore several policy pathways to stop this cost shifting.
First, data center companies must more transparently disclose where they are planning to locate data centers so that grid planners and utilities do not “double count” the same data center proposed in many different locations. (It is common for data center companies to simultaneously consider numerous locations for one data center.)
Second, states and utilities should mandate or strongly incentivize a practice called “demand response” for data centers. Demand response is when an electricity customer reduces electricity usage during periods of peak demand. This allows utilities to avoid blackouts, ensuring that there is enough generation to meet demand even when demand is high. A 2025 report by Tyler Norris, Dahlia Patino-Echeverri, and Tim Profeti of Duke University concludes that the grid could accommodate most proposed data centers without building new generating units if data centers curtailed their electricity usage just a few hours or days out of the year, during the highest demand periods.
Beyond more transparency and mandates or incentives for demand response, we argue for utility rates and terms of service that cause data centers to directly pay for the costs of new infrastructure. These tools include mechanisms such as upfront “payments in aid of construction” for new transmission lines or generation, and “exit fees” for data centers that cause a utility to incur costs for new generation and transmission and then leave the utility’s service area relatively quickly, causing the utility’s remaining customers to foot the bill.
Among other potential policy pathways, we also argue for ending secret contracts between utilities and data centers for rates and terms of service, reducing state and local grants of tax incentives and other perks to data centers without more proof of localized benefits, and tracking data on utilities’ disconnection of customers who cannot afford to pay their utility bills — a phenomenon that is likely to get worse as retail rates rise, in part due to rising demand from data centers.
Artificial intelligence currently drives a large chunk of U.S. economic growth, but we should not overlook the serious impact it is having on average consumers. We suggest ways to meaningfully address these impacts.
