A report from United Nations University highlights the significant environmental footprint of data centers. These facilities rival some of the world’s largest countries in water and energy usage. As artificial intelligence (AI) usage grows, their environmental impact is expected to double within four years.
Energy and Carbon Footprint
Last year, data centers globally consumed 448 trillion watt-hours of electricity, surpassing all countries except for ten. This energy use resulted in around 208 million tons of carbon dioxide emissions, comparable to Argentina’s total. Additionally, producing this energy required approximately 1.2 trillion gallons of water.
By 2030, data centers could account for nearly 3% of global electricity consumption, using 935 trillion watt-hours. If data centers were a nation, this would rank them sixth in power use, leading to nearly 440 million tons of carbon dioxide emissions. The study focused on energy usage and did not deeply explore the extensive water requirements for cooling these centers.
The demand is enormous,said Kaveh Madani, study co-author and water scientist.
The Role of Artificial Intelligence
AI significantly contributes to this growth. Currently, AI accounts for roughly 20% of data centers’ energy consumption, which could increase to 40% by 2030. This report is crucial due to its holistic approach, considering carbon, water, and land impacts in one framework.
Fengqi You, a Cornell University professor, noted the report brings necessary transparency to an issue often hidden by partial disclosures. While AI offers societal benefits, such as increased efficiency and improved safety, its environmental impact cannot be ignored.
Industry Response and Efforts
Industry leaders acknowledge the importance of addressing the environmental effects of data centers. Caleb Max, President of the National Artificial Intelligence Association, emphasized AI’s benefits. Josh Levi, president of the Data Center Coalition, highlighted the sector’s commitment to responsible growth and collaboration with policymakers.
Reducing Energy Use
There are practical ways to curb energy use. Madani noted that shortening AI queries could reduce energy consumption substantially. Reducing word count in requests by 30% might cut energy use by 25%, saving electricity equivalent to what 700,000 people in Africa consume annually.
AI systems require significant energy, especially for training sophisticated models. For instance, training GPT-3 consumed about 1.3 billion watt-hours, while the next version demanded 50 to 70 billion watt-hours. Operational requests, however, account for 90% of power use.
The Efficiency Paradox
Increased efficiency does not always lead to reduced energy consumption. As technology becomes more efficient, usage usually increases, causing overall energy use to rise. Kaveh Madani pointed out that while some firms promote using renewable energy, this reduces the availability of clean electricity elsewhere.
A challenge in the study was the lack of transparency among some companies regarding data center consumption. Without this data, managing environmental impacts becomes difficult.
Leave a Reply