Data Centers Have Not Raised Your Electricity Bill. A Causal Study Finds They May Have Lowered It.
As the AI infrastructure buildout draws a direct line between data center demand and higher electricity bills in popular coverage, a new causal study published on arxiv (2606.19777) finds the opposite held for most of the 2015-2024 period in the United States.
The paper’s headline result: a 10% rise in data center capacity was associated with approximately a 0.4% fall in average residential electricity prices, in states where the grid had room to grow. The finding covers the years before the current hyperscale AI wave fully hit the grid, which is a meaningful caveat.
The Mechanism
The explanation is not that data centers make electricity cheaper to produce. It is that electricity grids carry large fixed costs — transmission infrastructure, generation capacity, distribution networks — and those fixed costs can be spread across more kilowatt-hours when a large steady load uses the system more heavily.
Data centers are ideal for this calculation. They run 24/7 at consistent load, unlike residential demand that spikes in evenings and summers. When a new steady industrial customer joins the grid and new generation to serve it comes online at a lower marginal cost than the existing mix, the blended rate per kilowatt-hour falls.
Methodology
The researchers use US state-level data from 2015 to 2024 and apply instrumental variable estimation, using historical highway routes as an instrument for data center location. Highway access matters for data center siting because it affects construction logistics and supply chain access, but does not directly determine residential electricity rates through any other channel — which makes it useful for separating the data center effect from general economic trends.
The instrumental variable approach is necessary because data centers do not locate randomly. They cluster in states with cheap power and friendly policy, which creates a selection problem in simpler regression analyses. Using an instrument that predicts location but not rates through other channels isolates the causal effect.
The Caveat
The paper explicitly warns that the negative price effect holds only when the grid can keep building sufficient supply. The current AI infrastructure wave is different in scale and speed from what the 2015-2024 data covers. Several US grid operators are now projecting capacity shortfalls specifically driven by data center load growth, and some regional markets — PJM Interconnection, ERCOT in Texas — are already showing demand projections that outpace planned generation additions.
In that regime, the fixed-cost-spreading mechanism runs in reverse. A load that exceeds available supply forces expensive peaker plants onto the grid or drives capacity payments higher, both of which push rates up.
The paper’s finding is a useful historical baseline, not a projection. What it usefully refutes is the claim that data centers have been systematically raising residential electricity costs through 2024. They had not. Whether that holds through 2027 and 2028 depends on how fast the grid can build.