Harvard Finds AI Data Centers Burn 48% Dirtier Than the U.S. Grid
A new Harvard-led study maps 403 U.S. hyperscale data centers and finds their carbon intensity runs 48% above the national grid average, driven by where they are built.
- A Harvard-led study mapped 403 U.S. hyperscale data centers and estimated their total electricity use at 68-99 TWh annually.
- The sector's carbon intensity is 545 gCO2/kWh, roughly 48% above the U.S. national grid average of 370 gCO2/kWh.
- About 54% of electricity attributed to these facilities comes from fossil fuels, with only 25% from renewables.
- Virginia hosts 142 of the 403 facilities and sits in the PJM grid, one of the more carbon-intensive balancing authorities in the country.
- The study uses EPA eGRID2023 plant-level data for a bottom-up, facility-level attributional emissions framework, more rigorous than prior top-down estimates.
- No code is released, but the methodology is reproducible using public EPA eGRID data; the key input is the proprietary 403-facility dataset.
The AI buildout has a carbon problem that goes beyond raw electricity consumption. A new paper from researchers at Harvard's T.H. Chan School of Public Health, the University of Pisa, and several other institutions delivers the most granular accounting yet of U.S. hyperscale data center emissions, and the findings reveal a structural issue: these facilities are not just energy-hungry, they are disproportionately powered by fossil fuels because of where they are built.
The rapid proliferation of hyperscale data centers in the US, mainly driven by AI adoption, has raised concerns about the industry's environmental footprint. The team compiled facility-level information on 403 U.S. hyperscale data centers operating between May 2024 and April 2025, estimating their electricity consumption, electricity sources, and attributable CO2 emissions.
The numbers are bigger than the headline suggests
Across different facility-load scenarios, these hyperscale data centers consumed approximately 68 to 99 TWh of electricity and were associated with about 37 to 54 million metric tons of CO2. Under the central scenario, that demand corresponded to roughly 1.8% of total U.S. electricity consumption, with about 54% of attributed generation supplied by fossil-fuel sources. The electricity-weighted average carbon intensity landed at approximately 545 gCO2/kWh, about 48% above the contemporaneous U.S. national grid average of 370 gCO2/kWh.
That 48% gap is the headline finding, and it demands explanation. The national grid is already a mix of coal, gas, nuclear, and renewables. So why are hyperscale facilities so much dirtier than average? The answer is geography.
Location is the hidden carbon variable
Virginia had more hyperscale data centers (142) than any other state, followed by Oregon (56) and Ohio (38). Virginia's dominance is not accidental: land is cheap, fiber connectivity is dense, and the PJM grid interconnection historically offered reliable power. But PJM is one of the more carbon-intensive grids in the country.
This story is for Pro members
You've reached the end of the free preview. Upgrade to AlphaSignal Pro to read the full article - and everything else behind the paywall.
Comments
