Data centers have been cutting your electricity bill for years and the AI buildout may end that

A new study by the Electric Power Research Institute and Watershed found that average U.S. residential electricity rates would have been 6% higher without data centers built between 2019 and 2024, as doubling a state's data center capacity was associated with a 3.5% decrease in rates. However, a projected $7 trillion AI buildout is eroding spare grid capacity, with Gartner reporting data center electricity consumption will grow 26% in 2026 to 132 gigawatts globally, and Dominion Energy proposing its first base-rate increase since 1992 in Virginia, where data center clusters have driven wholesale electricity cost increases of 267%.

A new study found average residential electricity rates would have been 6% higher without the data centers built between 2019 and 2024, but a $7 trillion AI buildout is approaching the point where that math reverses. The public narrative around data centers and your electricity bill has it almost exactly backwards. For the better part of a decade, sprawling server farms powering the internet quietly pushed residential rates down. That's the opposite of what most people assume. According to research published this month by the Electric Power Research Institute and Watershed, a doubling of a state's data center capacity was associated with roughly a 3.5% decrease in residential electricity rates. The mechanism is straightforward: data centers are enormous, predictable industrial loads that spread a grid's fixed costs, transmission infrastructure, substations, control systems, across a larger customer base. Everyone's bill gets a little smaller because a very large new customer joined. That's not the story most ratepayers have been told. More than 30 states introduced over 300 data center bills in 2026 alone, covering everything from moratoriums to mandatory cost-sharing agreements. Texas Governor Greg Abbott directed state regulators in June to ensure data centers pay for their own grid interconnection costs rather than passing them to residential customers , a reasonable precaution framed, somewhat misleadingly, as protection from a cost that the historical record shows hasn't materialized yet. The EPRI finding matters most as a baseline. Between 2015 and 2024, the average U.S. residential rate would have run about 6% higher without the data center build-out that happened over that period, according to EPRI researcher Asa Watten. North Dakota is a clean example: large industrial customers, data centers included, measurably lowered prices for households from 2019 to 2024. The key condition in every case is the same: spare grid capacity. When headroom exists, concentrated industrial demand is deflationary. The fixed costs get allocated across more customers rather than piled onto the existing base. That condition is eroding fast. Gartner reported in June that data center electricity consumption will grow 26% in 2026 alone, reaching 132 gigawatts globally, up from 104 gigawatts in 2025. By 2030 the figure is projected to hit 290 gigawatts. AI-optimized racks are driving the density problem as much as the volume problem: a traditional server rack draws between 5 and 15 kilowatts; an AI-optimized rack pulls anywhere from 30 to over 100 kilowatts. Near some U.S. data center clusters, wholesale electricity costs have already risen 267%, according to data cited by grid analysts this year. That's not the deflationary scenario. That's what happens when demand outruns supply. Dominion Energy, which serves a dense corridor of Virginia data centers, proposed its first base-rate increase since 1992 in February 2025, adding roughly $8.51 per month to residential bills in 2026. Virginia has more data center capacity than any other state. It also has, by some estimates, less available grid headroom than it did five years ago. The sequence is not a coincidence. What this means for site selection and policy For the industry itself, the EPRI research quietly reshapes the politics of permitting. The standard argument against new data centers has been that they burden ratepayers. In states with meaningful grid slack - parts of the Midwest, the Mountain West, rural areas with underused transmission - that argument is now empirically weak. A 3.5% rate reduction per doubling of capacity is a real number that local officials can take into a zoning hearing. Developers in those markets have a genuine case to make, and some are starting to make it. Texas is the more complicated picture. The state exempts qualifying data centers from sales taxes on servers, cooling equipment, and electricity itself - a package the Texas Comptroller estimates will exceed $1 billion in fiscal year 2025 and grow to roughly $1.75 billion annually by 2030. Governor Abbott has said he'll work with lawmakers in the 2027 legislative session to phase out those exemptions. The timing is notable. Texas pulled its tax incentive framework precisely as the national conversation shifted from "do data centers raise bills?" to "will they, once supply tightens?" The answer in Texas, given its grid history and the scale of planned builds, looks increasingly like yes. A PowerLines analysis of 51 U.S. investor-owned utilities published in April found planned capital expenditure of at least $1.4 trillion through 2030, more than a fifth higher than the same utilities projected just a year earlier. That capital will eventually show up in rate cases. The question is whether new data center load grows fast enough to keep spreading those fixed costs, or whether the build-out gets ahead of generation and turns into a straightforward cost transfer from tech capex to residential ratepayers. The honest answer, for most markets right now, is that the inflection point hasn't arrived yet. Not yet - but the direction is clear. In the highest-density corridors, Northern Virginia, parts of Texas, Phoenix, it's closer than the headlines about rate-lowering data centers suggest. The deflationary effect is real, well-documented, and already past tense in the places that built earliest and fastest. Also read: Baseten built the fastest GLM-5.2 API on earth and the playbook tells you where inference is heading https://startupfortune.com/baseten-built-the-fastest-glm-52-api-on-earth-and-the-playbook-tells-you-where-inference-is-heading/ • AMD delivered better stock returns than Nvidia in H1 2026 but the AI chip war is far from over https://startupfortune.com/amd-delivered-better-stock-returns-than-nvidia-in-h1-2026-but-the-ai-chip-war-is-far-from-over/ • Washington puts Bitcoin in the same program as Palantir and Anduril, treating it as geopolitical infrastructure https://startupfortune.com/washington-puts-bitcoin-in-the-same-program-as-palantir-and-anduril-treating-it-as-geopolitical-infrastructure/