The rapid expansion of artificial-intelligence data centers is putting new pressure on electricity systems, with households and manufacturers potentially absorbing part of the cost when utilities build generation, transmission lines and other infrastructure to serve the facilities. The effect is not uniform: it depends on where data centers are built, how much new power they require and whether regulators require the companies to pay the full cost of connecting to the grid.
A modeling study published in May examined projected data-center and cryptocurrency-mining demand through 2030 across 26 interconnected U.S. It estimated that electricity costs could rise by 6% to 29% nationally, with increases reaching 57% in some regions. The sharpest modeled effects were in Virginia, eastern North Carolina, Pennsylvania, Maryland, Delaware, New Jersey, west Texas, Ohio, West Virginia and New York. The study measured demand-weighted wholesale electricity prices, so its projections do not translate directly into the same percentage increase for every household bill.
The model also projected that power-sector carbon-dioxide emissions could be as much as 28% higher in 2030 than in a scenario without data-center growth. The outcome depended partly on fuel prices and the location of new facilities. Concentrating development in areas such as Virginia produced larger regional increases, while spreading facilities across the country reduced the size of local spikes without eliminating the national increase.
The difference between wholesale prices and residential bills is important. Wholesale power is only one part of a household’s bill, alongside transmission, distribution, taxes and other charges. Those increases have multiple causes, including equipment costs, grid investments and clean-energy requirements, but data-center demand has been identified as a major driver in some high-growth regions.
Manufacturers are facing a similar problem. Capacity charges help pay for power supplies reserved for periods of peak demand and can represent a larger share of manufacturers’ bills than residential bills. In the same region, data centers can use as much electricity as a mid-sized town, and their projects can be built faster than the generation needed to serve them.
Some utilities and policymakers are responding by proposing special rates, advance payments or financial guarantees for very large electricity users. California’s current structure generally gives transmission-connected data centers lower rates than customers using the distribution network, partly because they avoid many distribution costs. A state policy review proposed a new very-large-load tariff requiring stronger credit protections, upfront infrastructure payments and rates that cover transmission, distribution and reliability costs.
Technology companies have also made voluntary commitments to finance energy and grid improvements for their facilities, but those commitments do not determine how regulators will allocate costs across customers. The immediate policy question is whether data centers will pay for the additional capacity and infrastructure they trigger, or whether some of those expenses will continue to flow through rates paid by households and other businesses. The answer will vary by utility territory and will shape the cost of the next phase of data-center construction.