Projected emissions from US AI data centres have sparked widespread attention with comparisons to millions of cars, but a new analysis clarifies these figures and underscores how energy choices and regulations could dramatically alter the future carbon footprint of this growing sector.
- Emissions projected between 24 and 44 million tonnes CO2-eq annually by 2030
- Water use could equal household consumption of 6-10 million Americans
- Energy sourcing and regulation can cut emissions by up to 73%
What happened
A Cornell engineering team published a study in Nature Sustainability projecting the environmental impact of planned US AI data centres. Their research estimates these data centres could emit between 24 and 44 million tonnes of CO2-equivalent annually by 2030. This range accounts for differing growth scenarios, from moderate to aggressive expansion, and includes a water usage estimate matching the needs of millions of households.
The study garnered widespread media attention and alarm, especially through headlines suggesting data centre emissions would equal those of 24 million cars. However, this figure conflates tonnes of CO2 emissions with numbers of vehicles, leading to an overstated comparison. The authors clarify that the correct equivalence is between 5 to 10 million cars, not 24 million.
Why it matters
The environmental footprint of AI data centres is a growing concern as demand for computational power expands rapidly. This study illustrates how server efficiency, energy source, and geographic location heavily influence emissions and water consumption. For example, deploying servers in regions with cleaner grids or abundant water can reduce impacts significantly.
Policy decisions and corporate practices also play a decisive role. The study shows that best-in-class interventions such as strategic site selection and green energy procurement could cut emissions by as much as 73% and water use by 86%. Conversely, reliance on fossil fuels, like fracked gas, risks locking in high pollution levels well into the next decade.
What to watch next
As the AI industry and data centre operators move forward with expansion plans, regulatory frameworks and energy market developments will be key to realizing lower emissions scenarios. European regulatory efforts to enforce data-centre reporting and efficiency standards serve as a potential model for mitigating impacts.
Additionally, ongoing negotiations about new gas plants and power purchase agreements in the US will help determine whether the worst-case emissions projections come to pass. The difference between planned versus actual built capacity remains a critical factor, as many proposed facilities may not materialize amid power, permit, or market constraints.