Environmental Impact of Data Centers and AI

Data centers significantly impact the environment, rivaling some of the world’s largest countries in their resource usage. A report by the United Nations University highlights the environmental consequences of data centers as artificial intelligence (AI) expands. In 2021, the global electricity consumption by data centers reached 448 trillion watt-hours. This level of consumption exceeds that of all but ten countries globally. It also resulted in 208 million tons of carbon dioxide emissions, comparable to the emissions of Argentina. The energy production involved used approximately 1.2 trillion gallons of water.

By 2030, projections suggest data centers will use nearly 3% of the world’s electricity, amounting to 935 trillion watt-hours. If data centers were classified as a country, they would rank sixth in power consumption by then. This would result in nearly 440 million tons of carbon dioxide emissions. The report did not delve into the extensive water usage required to cool these centers.

Kaveh Madani, a co-author and water scientist, compared the scale of data centers to nations, emphasizing the substantial demand they impose. Presently, AI accounts for about 20% of data centers’ energy consumption. This figure is expected to double by 2030. The study marks the first global examination of some of these impacts.

“The general public should be concerned, but not panicked,” said Fengqi You, a Cornell University energy engineering professor, on the significance of the report’s insights.

Jean Su of the Center for Biological Diversity emphasized the importance of the report as it is the first comprehensive examination by the UN on the environmental impacts of AI. Caleb Max, President of the National Artificial Intelligence Association, highlighted the efficiency improvements in the AI industry and the benefits AI brings to society. Meanwhile, Josh Levi, President of the Data Center Coalition, stated the industry’s commitment to responsible practices.

Guidance to Reduce Energy Use

Reducing the energy consumed by AI involves making precise and brief queries, according to Madani. The report suggests cutting word use in AI requests by 30% can reduce energy use by 25%. This saving is equivalent to the annual electricity usage of approximately 700,000 people in Africa. Simple practices like this can have a considerable environmental impact.

AI operations, particularly sophisticated queries such as those used in ChatGPT, consume significantly more energy than simpler processes like email spam filters. Training large AI models like GPT-3 used around 1.3 billion watt-hours, but newer versions require even more power.

Challenges with Data Transparency

The study also encountered difficulties due to limited transparency from companies about their data centers’ energy usage. Miriam Aczel, a co-author and researcher, noted that most AI energy consumption arises during operational requests. Currently, GPT handles 2.5 billion prompts daily. Despite technology becoming more efficient, overall power usage continues to rise due to increased demand.

Even though renewable energy use is promoted for operating data centers, the overall clean energy supply is depleted, resulting in less clean energy being available elsewhere. The report highlights the urgent need for increased transparency in operations to manage environmental impacts effectively.