Global Impact of Data Centers and AI on the Environment

The environmental impact of data centers now rivals some of the world’s largest countries. A United Nations University report reveals that water and energy use as well as pollution from these centers will double soon, driven by artificial intelligence growth. Last year, data centers globally consumed 448 trillion watt-hours of electricity, exceeding all but 10 countries in consumption. These operations generated around 208 million tons of carbon dioxide, akin to Argentina’s output, and required approximately 1.2 trillion gallons of water. This information highlights AI’s environmental consequences.

By 2030, data centers will use nearly 3% of global electricity. Their power consumption could reach 935 trillion watt-hours. In this scenario, data centers would rank sixth in electricity use worldwide, producing nearly 440 million tons of carbon dioxide. While the focus of the study is on energy consumption, the significant water used to cool data centers couldn’t be ignored. Kaveh Madani, co-author and water scientist, notes the demand’s enormous scale.

AI significantly boosts data center growth. Currently, AI accounts for around 20% of data centers’ energy usage and may rise to 40% by 2030. These findings are crucial because of the United Nations’ role in compiling carbon, water, and environmental justice data. Fengqi You, an energy engineering professor at Cornell University, emphasizes the necessity for public awareness.

AI industry leaders, like National Artificial Intelligence Association President Caleb Max, highlight AI’s efficiency and numerous benefits. Josh Levi, president of the Data Center Coalition, stresses a commitment to responsible growth and collaboration with policymakers and communities.

“AI impacts are tangible and grounded in infrastructure and energy use,” explained Madani. Although AI may seem cleaner than cars or furnaces, its unseen pollution remains significant. Being less verbose in AI queries can substantially decrease energy use, cutting word count by 30% saves about the same energy as what 700,000 people in Africa use annually. Madani advises precision and brevity in AI interactions.

AI complexity correlates directly with energy consumption. GPT-3 required around 1.3 billion watt-hours to train, while its successor used 50 to 70 billion watt-hours, illustrating AI’s escalating demands. Despite efficiency advances, increased AI use often results in greater overall power consumption. Miriam Aczel, a report co-author, notes that operational requests constitute most of AI’s energy requirements, with GPT handling 2.5 billion prompts daily.

Even as technology becomes efficient, its usage often increases; therefore, aggregate energy consumption rises. Some claim renewable energy use, but this often shifts the clean energy use elsewhere. A challenge in the study was the lack of transparency from companies regarding data centers and AI’s consumption. Without disclosure, these impacts cannot be managed effectively.

The Associated Press funds its environmental coverage through private support, yet maintains editorial independence, ensuring comprehensive climate and environmental reporting.