Introduction

The rapid growth of data centres in the US is raising concerns about wide-ranging impacts on the energy system and environmental conditions in local communities. Data centres are electricity-intensive, and the surge in demand is straining regional electricity grids, complicating long-term planning, and potentially increasing electricity prices. At the local level, data centre development generates tensions over land use, water consumption, noise, and air quality. This paper provides an assessment of electricity use by data centres and quantifies the social costs from pollution emitted during electricity production.

Methods and data

The study uses facility-level data for approximately 2,800 operational data centres in 2025, sourced from S&P Global Insights. Electricity consumption is estimated using net utilised power, power usage efficiency (PUE), and a load factor set to 55% in the default scenario. The AP4 integrated assessment model connects emissions of local air pollutants to public health consequences and social costs. Greenhouse gas (GHG) emissions are valued using the social cost of carbon (SCC), with a default value of approximately $190/ton CO₂. Gross external damages (GED) combine monetary damages from local air pollution and GHGs.

Results

Air quality impacts

Data centres consumed approximately 250 TWh of electricity — between 5 and 6% of total US generation (p.10). Areas of elevated PM₂.₅ occur in Texas, Virginia, the Carolinas, Pennsylvania, Ohio, Wyoming, Nebraska, Iowa, and North Dakota (p.10–11). The highest air pollution damages are in large cities including Houston, Dallas, San Antonio, Chicago, Pittsburgh, Detroit, Columbus, Cleveland, Los Angeles, Phoenix, Las Vegas, and Boston (p.11).

Total GED

The total GED is estimated at $24.6 billion, with a sensitivity range of $10 billion to $33 billion (p.10). GED from GHGs comprises 80% of the total, though this share is sensitive to modelling assumptions (p.10). Facilities in just 10 states contribute 70% of total GED, with Texas and Virginia accounting for 30% (p.10). Commercial electricity prices in Virginia and Texas are under $0.10/kWh, well below the national average of $0.13/kWh (p.11).

GED, GDP, and personal income

Nationally, GED from all data centres amounts to $25 billion, while GDP for data processing, hosting, and related services was $544 billion in 2024 — meaning GED comprises just 5% of GDP (p.12). Significant heterogeneity exists across states. In North Dakota, GED was over 5 times larger than GDP from IT data hosting services (p.12). In Wyoming, GED was about 2 times larger than GDP (p.12). Data centres in California contribute 37% of national GDP and just 2% of the GED (p.12). In North Dakota, GED is 15 times larger than personal income from data hosting facilities (p.13).

Planned facilities

Total planned capacity is estimated to consume 210 TWh with an associated GED of $21 billion (p.13). Coupled with operational facilities, this suggests increases in electricity load and GED of up to 85% in the near term (p.13). Virginia exhibits the largest expected GED from planned data centres of just under $7 billion — 60% more than the GED from operational data centres in Virginia (p.13–14). GED from planned data centres in Illinois, Pennsylvania, and Arizona may double (p.14).

Data centre types

Of the 2,816 operational facilities, 432 are hyperscale, contributing 41% of the GED and consuming 40% of electricity (p.14). Despite numbering only 121, cryptocurrency mining centres produce one quarter of the GED while consuming 20% of total power (p.14).

AI

IEA (2025) reports that roughly 15% of all energy use at data centres is due to AI activities, suggesting 38 TWh and $3.7 billion in GED are attributable to AI (p.15). If AI boosts GDP by 0.5%, the GED constitute just 2.4% of increased GDP (p.15).

Conclusions

The GED, totalling $24 billion, comprise just 5% of the GDP contributed by data storage and hosting facilities. GED are highly concentrated, with 30% stemming from facilities in Virginia and Texas, where lower-than-average commercial electricity prices likely play a central role in facility location decisions. Planned facilities may increase electricity load and associated GED by up to 85% in the near term. Unless the effect of AI on GDP is exceedingly small, the GED from AI-based energy use at data centres appears to be very small relative to the possible benefits of the technology.