According to a report from the United Nations University, the environmental impact of data centers is comparable to that of some of the world’s largest countries. The report predicts that water and energy usage, as well as pollution from data centers, will double in four years due to the growing use of artificial intelligence (AI).
In the past year, global data centers consumed 448 trillion watt-hours of electricity. This level of consumption is surpassed by only 10 countries globally. The electricity use resulted in approximately 208 million tons of carbon dioxide emissions, mirroring the emissions of Argentina. Additionally, the energy production required about 1.2 trillion gallons of water. By 2030, data centers are projected to use nearly 3% of the world’s electricity, which translates to 935 trillion watt-hours. If data centers were a country, they would rank as the sixth-highest energy consumer by that time, producing nearly 440 million tons of carbon dioxide.
“If you look at these numbers, we’re seeing scales comparable to nations,” said Kaveh Madani, co-author of the study and director of the United Nations University Institute for Water, Environment and Health in Canada.
The increasing demand for data centers is heavily driven by AI technologies. Currently, AI accounts for about 20% of the energy used by data centers, a figure that is expected to rise to 40% by 2030.
Fengqi You, a Cornell University energy engineering professor, highlighted the significance of the report due to its comprehensive approach in addressing carbon, water, land, and life-cycle impacts. Jean Su, director of the Energy Justice Program at the Center for Biological Diversity, emphasized the report as a groundbreaking global assessment of AI’s environmental impact.
Caleb Max, President of the National Artificial Intelligence Association, noted the industry’s pursuit of increased efficiency and its benefits to society in terms of improved safety, longevity, and food production.
Josh Levi, president of the Data Center Coalition, affirmed the industry’s commitment to responsible growth by engaging with policymakers, communities, and industry partners to adopt best practices.
Reducing Energy Use
Kaveh Madani pointed out ways individuals can mitigate the energy demands of AI. By being less verbose in queries, users can reduce AI energy consumption. The report highlights a potential 25% energy reduction by cutting query word count by 30%—enough to power about 700,000 people in Africa for a year.
AI processes like those used in ChatGPT are significantly more energy-intensive compared to simpler text classification tasks such as email filtering. More complex AI systems require substantial power for both training and operations. For example, training GPT-3 required around 1.3 billion watt-hours, while its subsequent version demanded between 50 and 70 billion watt-hours.
Efficiency Paradox
Madani identified a paradox where increased efficiency of machines can lead to higher overall energy use due to their more frequent deployment. While some companies rely on renewable energy, this may deplete clean electricity supplies, leading to increased use of non-renewable sources elsewhere. Transparency in data center operations remains a key challenge in managing energy use effectively.
The UN report, bolstered by its comprehensive data and global perspective, calls for greater transparency and responsible practices as the world navigates the growing energy demands of data centers and AI technologies.