EPRI Study Finds Data Centers Lowered U.S. Power Rates Through 2024
An EPRI working paper released June 18 estimates that doubling U.S. data center capacity reduced residential retail electricity prices by 3.5% from 2015 to 2024; growth from 2019 to 2024 lowered average residential rates by roughly 6%. The authors attribute the result to fixed-cost spreading and cheaper new capacity, but warn that supply bottlenecks or data-center demand that fails to materialize could reverse the effect during the much larger AI buildout.
An Electric Power Research Institute working paper posted June 18 estimates that U.S. data centers modestly lowered average retail electricity rates between 2015 and 2024. The result is historical, not a promise that the current AI infrastructure boom will make power cheaper.
What the study found
Researchers Asa Watten, John Bistline and Geoffrey Blanford estimate that a doubling of data center capacity reduced residential retail electricity prices by 3.5% for a fixed level of demand. Because the average customer lived in a state where capacity rose about 160% from 2019 to 2024, their model implies an average residential-rate reduction of roughly 6% over that period.
The authors argue that electricity pricing does not behave like a simple commodity market. Utilities recover large fixed costs for generation, transmission and distribution across the kilowatt-hours they sell. Durable new demand can spread those costs across more sales, while new generating capacity may be cheaper than older assets it replaces.
To address the possibility that data centers simply chose states with lower power prices, the paper uses the length of the 1947 interstate-highway plan in each state as an instrumental variable. The reasoning is that later fiber routes followed highway corridors, influencing data center location without being designed around modern electricity prices. EPRI says the results remain consistent across customer classes and several robustness checks.
Why the AI buildout could differ
The authors explicitly limit the finding to a period of comparatively modest expansion. The much larger, faster buildout of hyperscale AI facilities could encounter transformer and turbine shortages, permitting delays, fuel-price effects or other supply constraints that make new capacity more expensive.
Demand durability is another key risk. If utilities build generation and grid infrastructure for projected AI load that never arrives, those fixed costs must be recovered from fewer kilowatt-hours, reversing the mechanism described in the historical data.
For infrastructure planners, the useful conclusion is narrower than the headline debate: large computing loads do not automatically raise or lower household rates. The outcome depends on whether demand is durable, capacity arrives at an affordable cost, and tariffs allocate new infrastructure costs without shifting them to other customers.
Key Points
- 1The working paper estimates that doubling data center capacity reduced residential retail electricity prices by 3.5% from 2015 to 2024.
- 2Its model implies data center growth from 2019 to 2024 lowered average residential rates by roughly 6% through fixed-cost spreading and cheaper new capacity.
- 3The authors warn that supply constraints or AI demand that fails to materialize could reverse the historical price effect.
Scoring Rationale
The study offers a quantified causal estimate on a major AI-infrastructure cost question and clearly states the conditions under which the historical result may not carry into the current buildout. Its practical value is substantial for capacity and rate-design decisions, while the preprint status and extrapolation limits keep the score below a definitive policy or market-changing result.
Sources
Primary source and supporting public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems

