The significant electricity consumption of data centers, many of which are powered by fossil fuels, leads to a substantial carbon footprint. The process of training a handful of artificial intelligence models can emit over 626,000 pounds of carbon dioxide. This is nearly five times the lifetime emissions of the average American car.

In a "mid-case" scenario, the power demand from data centers could result in an output of 400 TWh of CO₂-related emissions, which is comparable to the emissions of the entire country of South Korea. As the number of data centers grows, and if they continue to rely on non-renewable energy sources, these carbon emissions will trap the sun's heat in the atmosphere and contribute to increased global warming. While some thought leaders, like OpenAI's Sam Altman, are targeting solar energy as a solution, a large-scale shift to clean energy for data centers is still needed to mitigate this impact.

💨 Training GPT models has emitted over 626,000 pounds of CO₂ — that’s 5x the lifetime emissions of the average American car. (Hao, 2020)

💨 In the “mid-case” scenario, data centers will emit 400 TWh of CO₂-related output — comparable to the entire country of South Korea. (Enerdata, 2024)

💨 By 2030, data centers could emit 400 TWh of CO₂-related emissions—equivalent to the emissions of South Korea.

💨 The growth of AI-powered systems can lead to increased burning of fossil fuels.

💨 Shifting to renewable energy for data centers is crucial for mitigating this impact.

💨 Cloud computing infrastructure emissions are growing rapidly — much of the energy is not even used for computation but for cooling and backup. (Monserrate, 2022)

The voracious energy appetite of AI is a significant concern for the global energy grid. As the number of data centers grows, the demand for power will continue to increase. With much of this power coming from fossil fuels, the resulting carbon emissions trap heat and contribute to global warming. Some thought leaders are already exploring solutions, such as using solar energy to power these facilities. Building a sustainable future for AI requires a conscious effort to power our digital world with clean and renewable sources.

Sources:

• Jha, Rohan, Rishabh Jha, and Mazhar Islam. "Forecasting US Data Center CO₂ Emissions Using AI Models: Emissions Reduction Strategies and Policy Recommendations." Frontiers in Sustainability, vol. 5, 2024, 8 Jan. 2025.: https://doi.org/10.3389/frsus.2024.1507030

• Hao, Karen. "Training a single AI model can emit as much carbon as five cars in their lifetimes." MIT Technology Review, 6 June 2020.: https://www.technologyreview.com/2020/06/06/1002987/ai-carbon-footprint-cars-2020/

• The Economist (2024): https://www.economist.com/

• WSJ (2024): https://www.wsj.com/

• Statista (2023): https://www.frontiersin.org/journals/sustainability/articles/10.3389/frsus.2024.1507030/full#ref58

• MIT Technology Review: https://www.technologyreview.com/2019/06/06/239031/training-a-single-ai-model-can-emit-as-much-carbon-as-five-cars-in-their-lifetimes/

• Dig Watch Observatory: https://dig.watch/updates/how-much-water-and-energy-does-training-an-ai-model-require

• Google Environmental Report (2025): https://www.gstatic.com/gumdrop/sustainability/google-2024-environmental-report.pdf

Carbon: Assessing AI's Carbon Footprint

A silent plume: every prompt adds to the carbon cloud. Responsibly Prompt and adopt sustainable AI