Solmar CapitalSolmar Capital
Coverage
ERCOTPJMMISOSPPWECCSoutheastSouthwestERCOTPJMMISOSPPWECCSoutheastSouthwestERCOTPJMMISOSPPWECCSoutheastSouthwestERCOTPJMMISOSPPWECCSoutheastSouthwest
Day-ahead power

Hyperscale data center power demand rivals state-level consumption

Solmar Insights

This year, hyperscale data centers in the United States consumed nearly as much electricity as the entire state of Michigan, according to a new peer-reviewed study from Harvard. The research estimates the sector used 82 terawatt-hours (TWh) over a 12-month period as the artificial intelligence demand surge accelerated, with energy-related emissions equaling Denmark’s annual total.

Key figures

82 TWh annual consumption
25.7 million metric tons CO2 emissions
1.8% of total U.S. electricity generation
54% of power from fossil fuels

Sharp growth driven by AI

The Harvard study marks the first facility-level accounting of U.S. data center energy and emissions since the widespread adoption of AI, revealing a two- to threefold increase in greenhouse gas output compared to 2018 models. Most of this growth has been attributed to the heavy computational loads and 24/7 operation required by modern hyperscale and AI workloads. The emissions estimate for the sector has risen to 25.7 million metric tons of carbon dioxide for the period between May 2024 and April 2025, up significantly since the last comprehensive review.

The increase in emissions is a direct result of the rising electricity draw across major data center campuses, especially as firms race to build out new sites to accommodate AI infrastructure. This highlights the challenge in decarbonizing the digital infrastructure sector, where rapid expansion can outpace efforts to procure clean energy supply.

Regional energy mixes and emission hotspots

More than half (54%) of data center consumption is currently supplied by fossil fuel sources, according to the study, with substantial geographic variation. Four states, Virginia, Iowa, Ohio, and Oregon, concentrate much of this energy use. The emissions footprint is particularly high in regions where the grid is more carbon intensive, such as Ohio, where 80% of electricity is fossil-generated and 20% still derives from coal plants.

Grid composition directly impacts the net emissions attributable to each data center. In regions like Loudoun County, Virginia, large campuses are penalized by the regional electricity mix. The analysis underscores the difficulty of reaching meaningful reductions in overall sector emissions without shifts in procurement strategy or grid decarbonization at scale in high-growth areas.

Policy implications and guidelines for siting

The Harvard research team developed Never Just a Building, a data center explorer tool, to map and compare environmental and public health impacts at potential development sites. This resource allows users, from developers to community members, to evaluate proposed data centers based on projected emissions and localized air quality effects, scoring sites on a spectrum from 0 (best) to 100 (worst).

One illustrative scenario showed a data campus in Loudoun County scoring 76 due to the fossil-reliant PJM grid. If powered round-the-clock by clean energy, the score improves to 54, and modeled pollution plumes shrink considerably. The creators emphasize that the tool provides unbiased, facility-level data to inform both public and private decision-making processes before projects are greenlit.

Investor, developer, and regulatory outlook

The study’s release and associated explorer tool are likely to influence both investor and policymaker approaches to data center siting and procurement. Transparency in reporting energy consumption and carbon intensity at the facility level is becoming central in planning, especially as localities weigh economic benefits against environmental costs.

Large cloud and hyperscale operators face growing scrutiny, and the research supports calls for investment in new clean power supply to support digital infrastructure. As capital allocation decisions are increasingly tied to environmental metrics, site selection in lower-carbon regions or with additional clean energy procurement may tilt future project economics and approval timelines.

Market context and next steps

This updated emissions and energy use profile is a critical input for modeling future power demand and infrastructure requirements across PJM, MISO, and the Western Interconnect. With hyperscale facilities now representing nearly 2% of U.S. load, utilities and ISO planners are likely to revisit assumptions in generation planning, grid upgrades, and interconnection queues.

Regulatory responses may include more stringent reporting for new AI and cloud campuses, urban air quality rules, or clean energy mandates for bulk buyers. The study’s authors intend further work on the health impacts associated with regional pollution attributable to this growing load, which may in turn shape national and local permitting criteria and policy debate.

What this means for buyers

Data center capacity and power procurement in PJM, MISO, and key states such as Virginia and Ohio are directly impacted. The study found hyperscale centers used 82 TWh and sourced 54% of power from fossil fuels, escalating emissions exposure and regulatory risks. Buyers should quantify grid mix impact, review local air quality scoring, and prioritize 24/7 clean energy procurement in siting and contracting.

Reporting via the original publisher

Share the Post:
Solmar Platform

Origination starts with a scored field.

Pre-screened energy and digital infrastructure projects, scored for readiness and searched against your criteria. Buyers select confidentially and sellers accept or decline before any introduction.