Industry

US data centers consume 82 terawatt-hours annually, rivaling mid-sized nations' electricity use

A Harvard study tracking 403 hyperscale facilities finds fossil fuels power over half of US AI infrastructure, producing emissions equivalent to Denmark's annual carbon footprint.

Last verified:

AI infrastructure’s energy demand now rivals entire nations

US hyperscale data centers—the massive facilities developers are racing to build primarily to support generative AI—consumed approximately 82 terawatt-hours of electricity annually, according to research published in the journal Nature Sustainability. This volume represents 1.8% of total US electricity consumption and exceeds the annual electricity use of roughly 148 countries, including Austria, Israel, and Venezuela. The carbon footprint is equally striking: According to The Verge, the electricity consumed by these facilities produced 25.7 million metric tons of CO2 emissions annually, comparable to Denmark’s entire annual carbon footprint and exceeding the total emissions of approximately 113 nations.

Satellite imagery and EPA records: Harvard’s methodology for 403 facilities

Researchers led by Gianluca Guidi, a postdoctoral fellow in the Department of Biostatistics at Harvard T.H. Chan School of Public Health, used satellite imagery to identify 403 hyperscale data centers across the US. Rather than measuring consumption directly, the team estimated power consumption based on electrical capacity data. According to The Verge, they then used Environmental Protection Agency records to approximate how much carbon pollution each facility would have produced if connected to the local power grid between May 2024 and April 2025. This methodology preserves uncertainty inherent in estimation-based approaches while providing the first systematic view of AI infrastructure’s distributed energy footprint.

Fossil fuels power 54% of hyperscale demand

A critical finding from the Nature Sustainability paper is the persistent reliance on fossil fuels. According to The Verge, approximately 54% of the electricity consumed by US hyperscale data centers likely came from gas and coal. This dependency undercuts the narrative that AI’s environmental impact is a fixed property of the technology itself. Guidi emphasized this point in comments to The Verge: “The same computing infrastructure can have dramatically different consequences depending on the electricity system supporting it. I actually find that encouraging, because it means these impacts aren’t inevitable. We have choices about where we build and how we power this infrastructure.”

Why This Matters

The Harvard study reframes AI’s environmental cost as a policy and infrastructure choice rather than a technological inevitability. Teams evaluating data center locations now have quantified evidence that electricity grid composition directly determines emissions per compute unit. For policymakers, the findings suggest that expanding renewable energy capacity in regions hosting data centers—or mandating renewable procurement by hyperscalers—could reduce AI-driven emissions without sacrificing compute throughput. The 25.7 million metric ton annual carbon footprint is substantial enough to shift procurement decisions at cloud providers and enterprise AI builders, particularly those facing Scope 3 emissions reporting requirements or renewable energy commitments. If the broader trend of data center clustering in fossil-heavy regions continues unchecked, annual emissions could exceed current estimates significantly by 2028–2030.

Frequently Asked Questions

How did researchers measure the power consumption of 403 data centers?

Harvard researchers used satellite imagery to identify facilities, then estimated power consumption based on electrical capacity data and EPA grid records for the May 2024–April 2025 period. Direct metering data was not available, so estimates carry inherent uncertainty.

Why does the electricity source matter for the same computing infrastructure?

According to Harvard researcher Gianluca Guidi, 'the same computing infrastructure can have dramatically different consequences depending on the electricity system supporting it.' A data center powered by renewable energy produces lower emissions than one relying on fossil fuels, even at identical compute capacity.

How do US data center emissions compare globally?

According to The Verge, the 25.7 million metric tons of CO2 from US hyperscale facilities exceeds the total carbon emissions of approximately 113 nations. The electricity consumption of 82 terawatt-hours exceeds the annual use of about 148 countries individually.

#data-centers #climate #energy #ai-infrastructure #emissions