How we check a data-center claim.

The chart tests whether states with a larger modeled 2024 data-center electricity share also saw faster residential price growth from 2014 to 2024. This descriptive cross-section cannot trace facility opening dates, model price causes, or estimate the bill effect of a proposed project.

It uses EPRI’s 2024 low, medium, and high modeled state energy estimates; final EIA Form 861 state retail sales; and exact EIA residential average prices for 2014 and 2024. Each plotted point uses EPRI’s medium estimate. We calculated each share from those records.

share (%) = EPRI modeled 2024 data-center TWh
            ÷ EIA 2024 total retail-sales TWh × 100

price growth (%) = (EIA 2024 residential ¢/kWh
                  ÷ EIA 2014 residential ¢/kWh − 1) × 100

Pearson r

−0.124

R²

0.015

95% interval for r

−0.389 to +0.160

Scope

50 states · descriptive · not causal

EPRI’s modeled values include cryptocurrency mining. They are not metered census data. Comparing a 2024 endpoint share with a ten-year outcome cannot reveal timing or local tariff effects. This simple state comparison found no faster 2014–2024 residential price growth in states with a higher modeled 2024 share.

A digital action can trigger computation, storage, and network traffic. Its marginal energy changes with the service, workload, model, device, network, hardware, utilization, cooling, and power supply. Shared infrastructure also uses energy while waiting for or protecting against demand.

One person's action cannot be assigned a universal energy value or mapped to a new campus. Capacity planning responds to aggregate and peak use across many people and organizations. Dividing average system energy by actions produces an allocation, which may differ from the energy saved by skipping one action.

individual request → shared digital service
          → compute, storage, and network activity
          → aggregate and peak demand → capacity planning

This was a targeted review of primary government health, environmental, wildlife, permitting, and local assessment sources available through August 12, 2026. Advocacy claims and company fact sheets helped identify questions. Causal conclusions rely on the primary record.

Each claim was tested as a chain: identify a facility source, measure exposure at a receptor, establish dose and duration, match the exposure to applicable outcome evidence, and evaluate alternative causes. General evidence about diesel exhaust, environmental noise, habitat change, or artificial light may justify investigating a pathway. It cannot by itself attribute a local diagnosis or death.

source → receptor exposure → dose and duration
       → applicable outcome evidence → causal assessment

The review did not locate a credible primary source directly attributing cancer or livestock deaths to a data center. Stronger direct evidence could change that conclusion. This was a targeted review, not a registered systematic review.

Virginia's legislative auditor estimated about 7,200 acres occupied by operating data centers in 2024. The site acreage was not parcel-matched to former farm use. Dividing it by USDA's 2022 “land in farms” total can show statewide scale; it cannot say those acres were farmland or assign a cause to changes in the agricultural statistic.

7,200 data-center acres ÷ 7,309,687 Virginia acres in farms
  = 0.0985% scale comparison

7,200 data-center acres ÷ 488,292-acre net 2017–2022 decline
  = 1.4745% scale comparison; no loss attribution

“Land in farms” includes more than planted cropland and can change with farm definitions and reporting. Those limits prevent this comparison from attributing farmland loss to data centers. A project decision needs parcel-level former use, disturbed and impervious acreage, soil capability, habitat, utility corridors, and a local denominator from the same geography and period.