A causal study finds data-centre growth pushed the average U.S. residential electricity rate down between 2019 and 2024. But in the grid serving Virginia's "Data Center Alley," the effect runs the other way — and researchers warn the national picture could flip.
Data centres now use an estimated 4.5% of U.S. electricity, more than double their share five years earlier5, concentrated overwhelmingly in Virginia, Texas and a handful of other states1. Two-thirds of Americans say they oppose a data centre being sited near them5, and the assumption behind that opposition is straightforward: more demand should mean higher bills.
The state-level numbers don't cooperate. An analysis of EIA data by the Institute for Energy Research (IER), a fossil-fuel-industry-funded think tank with an interest in shifting blame for high electricity costs onto policy rather than fuel or grid spending, found the correlation between a state's number of data centres and its current electricity price is statistically indistinguishable from zero2. The ten states with the most data centres — Virginia, Texas and California among them — averaged 14.46¢/kWh in 2025, against 14.39¢/kWh for every other state: a gap of under half a percent2.
Run the same test on price growth rather than price levels, and the pattern goes further than "no relationship" — it runs backwards. States where electricity sales grew fastest from 2015 to 2025, a group led by Texas and other data-centre states, saw roughly half the price increase of states where sales grew slowest or fell2. That direction is not just an industry-funded finding: E3's review cites a 2026 Lawrence Berkeley National Laboratory analysis reaching a similar conclusion outside the mid-Atlantic grid that covers Virginia3, and LBNL's own DOE-funded retail-price research programme independently covers this exact relationship6. The causal paper discussed below reports the same negative correlation in its own descriptive data, before attempting to control for it5.
A negative correlation like this doesn't mean data centres lower prices; it more plausibly means developers chase states that already have spare grid capacity and low costs, and that avoiding the states already under cost pressure. California, which had by far the steepest residential price rise of any state in this period — nearly 40% — got there mostly through wildfire-hardening spending on its grid, not data-centre load, and has one of the lowest concentrations of data-centre capacity of any large state5. Virginia, the state with the single largest concentration — more than a fifth of its electricity goes to data centres — has had price rises close to the national average5. Comparing states this way conflates the effect of data centres with everything else that makes a state's grid expensive or cheap.
A working paper from researchers at EPRI, the utility-funded Electric Power Research Institute, tries to isolate the effect properly. Using the length of each state's 1947 Interstate Highway system as an instrument — fibre-optic cable, and so data centres, tend to run alongside the old highway corridors, a link plausibly unrelated to today's electricity prices once population and GDP are controlled for — the authors estimate that a doubling of a state's data-centre capacity causes average residential prices to fall by roughly 3.5%5. Applied to the actual growth in data-centre capacity between 2019 and 2024, that implies the average residential customer's rate fell by about 6% because of data centres, not despite them5.
The mechanism is ordinary utility economics rather than anything data-centre-specific: a large share of a power bill recovers fixed costs — the wires, substations and plant already built — and spreading those costs over more kilowatt-hours of durable new demand lowers the average cost per unit, provided the grid isn't already straining to meet it.
An hourly dispatch model built by Federal Reserve Bank of Dallas researchers, simulating wholesale markets across the continental U.S. with and without observed data-centre load, finds a real but sharply uneven effect. Nationally, existing data centres have added an estimated 2–6% to wholesale electricity prices since 20214. Broken down by grid, almost all of that effect sits in a handful of constrained corridors.
Virginia's grid — home to the densest cluster of data centres in the country — saw wholesale prices pushed up by roughly $5 per megawatt-hour, over a 10% increase4. Texas, which hosts the second-largest concentration, saw under $1, which the researchers attribute to the state's aggressive build-out of new generation capacity keeping pace with new load4. California's effect was smaller still, under $0.75, because relatively little data-centre capacity sits on its grid at all4.
The same researchers separately estimate that PJM, the grid operator covering Virginia and much of the mid-Atlantic, attributed roughly 50% of a recent jump in its capacity-market auction price to data-centre load growth — with the remainder coming from power-plant retirements, market-design changes and reduced credit given to fossil generators3. That analysis was commissioned by the Data Center Coalition, the industry's own trade group, though it was independently authored and its regional finding is directionally consistent with the Dallas Fed's separate wholesale-market model.
The causal estimate that data centres have, on net, lowered the average U.S. bill rests on a single working paper, not yet peer-reviewed, written by researchers at a utility-funded institute; its preferred specification's statistical instrument is only borderline strong by conventional standards, even though a weak-instrument-robust test still rejects a price-raising effect. The Institute for Energy Research's correlation findings, though corroborated in direction by independent research, come from a fossil-fuel-funded organisation with a stated interest in the conclusion. The Data Center Coalition-funded regional breakdown of PJM's price rise is independently authored but industry-commissioned. Whether the fixed-cost-spreading benefit persists depends on generation supply keeping pace with data-centre demand through the rest of this decade — a genuinely open question given current turbine and permitting bottlenecks, and one none of these studies can settle.
Wholesale dispatch models and PJM's own capacity-auction pricing show a measurable, non-trivial price effect concentrated in the grids where data centres cluster most densely — and that effect is a plausible major driver of the double-digit residential increases some Northeastern states reported in 2025.34
Once the fixed-cost-spreading benefit of durable new demand is properly isolated from confounding factors like siting selection, the best available causal estimate finds data centres have on average lowered residential rates — and cross-state correlations point the same direction.25
The apparent contradiction dissolves once the two effects are named separately. One is the "operating margin": data centres are large, steady loads that push grid operators to dispatch more expensive generators, which raises wholesale prices in the short run — the effect the Dallas Fed and PJM analyses capture, concentrated wherever the local grid is already tight. The other is the "build margin": durable new demand spreads a power system's large fixed costs over more units sold, which lowers average prices over the longer run — the effect the EPRI causal paper isolates, and the one that currently appears to be winning nationally.
On the balance of the strongest currently available evidence, data centres have not driven up the typical American's electricity bill nationally, and may have modestly lowered it. But that finding is younger, thinner, and more interested-party-adjacent than its confident framing suggests, and it explicitly does not hold in the specific grids — chiefly the mid-Atlantic corridor serving Virginia — where data-centre growth has outpaced new supply. Whether the national picture holds past the next few years depends on whether the build margin can keep winning as the buildout accelerates.