AI news

US hyperscale data centers now consume more power than most nations

A new study reveals US hyperscale data centers consumed 82 terawatt-hours of electricity in one year, emitting CO2 comparable to Denmark.

A small server rack on a US map showing varying pollution levels by region.
Illustration generated for this article

A recent study published in Nature Sustainability reveals that hyperscale data centers in the United States consumed approximately 82 terawatt-hours of electricity between May 2024 and April 2025. This massive energy demand, driven largely by the expansion of generative AI infrastructure, resulted in 25.7 million metric tons of carbon dioxide emissions. The research highlights that fossil fuels remain the primary power source for these facilities, creating an environmental footprint comparable to that of entire nations.

What happened

Researchers from Harvard T.H. Chan School of Public Health identified 403 hyperscale data centers across the US using satellite imagery. These facilities, which span millions of square feet, are being built at a rapid pace to support the computational needs of modern AI models. By estimating power consumption based on electrical capacity and cross-referencing with Environmental Protection Agency data, the team calculated the carbon pollution associated with each site if connected to the local grid during the study period.

The scale of consumption is staggering. The 82 terawatt-hours used by these 403 facilities represents about 1.8 percent of all electricity consumed in the United States during that year. John Sterman, director of the MIT System Dynamics group, noted that this amount exceeds the annual individual electricity consumption of 148 countries, including Austria, Israel, and Venezuela. Only the largest economies, such as the US and China themselves, consume more individually than this subset of data centers.

The resulting greenhouse gas emissions are equally significant. The study estimates that these data centers produced 25.7 million metric tons of CO2, a figure comparable to the annual carbon footprint of Denmark. Sterman pointed out that this emission level exceeds the total carbon output of approximately 113 nations. The primary driver for this high carbon intensity is the energy mix; roughly 54 percent of the electricity consumed by these hyperscale facilities likely originated from natural gas and coal sources.

How it works

The environmental impact of a data center is not determined solely by its hardware efficiency but heavily depends on the local electricity grid it connects to. The study demonstrates that location is a critical variable in calculating carbon footprints. For instance, generating electricity in Wyoming, which relies heavily on coal, results in roughly seven times more carbon emissions per kilowatt-hour than in Oregon, where hydropower is abundant. Consequently, two data centers consuming identical amounts of power can have vastly different environmental consequences based purely on their geographic placement.

This dependency on local infrastructure creates a risk of long-term carbon lock-in. Fengqi You, a professor at Cornell University, explains that power plants and supporting infrastructure built today to meet surging data center demand will operate for decades. Decisions made now regarding where to build and how to power these facilities will shape emissions profiles well into the future. The current trend of relying on existing fossil-fuel-heavy grids exacerbates this issue, as new gas turbines and infrastructure are sometimes installed directly at facilities to bypass grid constraints, further entrenching fossil fuel use.

Key details

  • The study analyzed 403 hyperscale data centers in the US between May 2024 and April 2025.
  • These facilities consumed an estimated 82 terawatt-hours of electricity, equaling 1.8 percent of total US usage.
  • Carbon emissions totaled 25.7 million metric tons, comparable to the annual footprint of Denmark.
  • Approximately 54 percent of the electricity used by these centers came from coal and natural gas.
  • Data centers in coal-reliant regions like Wyoming have seven times the carbon intensity of those in hydropower-rich areas like Oregon.
  • Current estimates are limited by a lack of transparent data disclosure from data center operators.

Why it matters

For software engineers and technical leaders, this study underscores that the environmental cost of AI is not an inherent, unchangeable property of the technology itself. As Gianluca Guidi, the lead author, notes, the same computing infrastructure can have dramatically different consequences depending on the electricity system supporting it. This means that architectural decisions, cloud provider selection, and deployment locations have tangible impacts on sustainability metrics. Ignoring these factors contributes to irreversible carbon emissions, as highlighted by Sterman, who warns that once CO2 is emitted, it cannot be undone.

Furthermore, the lack of transparency from data center operators poses a challenge for accurate reporting and accountability. Researchers had to rely on satellite imagery and estimations because direct data was unavailable. This opacity makes it difficult for companies to verify the green claims of their infrastructure providers. With bipartisan bills in Congress potentially mandating greater disclosures, technical leaders should anticipate stricter regulatory requirements and increased scrutiny on the energy sources powering their applications.

What you can do

  • Evaluate cloud providers based on their specific regional energy mixes rather than just global sustainability pledges.
  • Prioritize deploying workloads in regions with high renewable energy penetration, such as areas with significant hydro, wind, or solar capacity.
  • Monitor emerging legislation regarding data center energy disclosure to ensure compliance and prepare for reporting requirements.
  • Optimize model efficiency to reduce computational load, thereby lowering the absolute electricity demand regardless of the energy source.
  • Advocate for internal policies that require carbon impact assessments for large-scale infrastructure deployments.
  • Stay informed about independent tracking tools, such as the website being developed by Guidi’s team, to better estimate the footprint of specific facilities.
  • #ai infrastructure
  • #sustainability
  • #data centers
  • #energy consumption
  • #carbon emissions

Tools from the Bytechap store

$89

DocBento

Self-hosted document management that reads every scan and answers with page citations.

Live demo

Keep reading

All stories