Global data centers, driven significantly by the growth of artificial intelligence, consumed 448 trillion watt-hours of electricity last year, an amount exceeding the usage of all but 10 countries worldwide. This energy consumption resulted in approximately 208 million tons of carbon dioxide emissions, comparable to Argentina's yearly output, and utilized about 1.2 trillion gallons of water. A United Nations University report predicts that the environmental footprint of these data centers will double within the next four years.
By 2030, data centers are projected to account for nearly 3% of the world's electricity use, reaching 935 trillion watt-hours. If considered a country, data centers would rank as the sixth-highest in power consumption, producing an estimated 440 million tons of carbon dioxide. The report highlights that AI currently accounts for around 20% of data center energy use, a figure expected to rise to 40% by 2030. Training more complex AI models, such as the progression from GPT-3 (1.3 billion watt-hours) to its successor (50 to 70 billion watt-hours), requires significantly more energy.
Operational requests for AI, rather than training, are the primary driver of energy consumption, making up about 90% of AI's power usage. For instance, a typical ChatGPT-style query is 200 times more energy-intensive than basic text classification. AI-generated images and video demand even greater energy. Despite this, transparency regarding data center energy and water consumption remains a significant challenge, with many companies not disclosing key details. Researchers like Kaveh Madani and Miriam Aczel emphasize the difficulty in managing what is not disclosed and point out that making queries more concise can reduce energy usage; cutting word count by 30% can save 25% of the energy, equivalent to the annual electricity use of 700,000 people in Africa.
The increasing demand for AI has led to a construction boom for data centers, and major tech companies like Microsoft are forecasting substantial increases in their water usage. Microsoft, for example, initially projected its annual water needs to more than triple to 28 billion liters by 2030, up from 7.9 billion liters in 2020. While they later updated this to 18 billion liters for 2030, a 150% increase from 2020, even these revised figures exclude many new data center deals. In the US, water needs for data centers are expected to surge from about 60 billion liters in 2022 to between 150 billion and 275 billion liters by 2028. This growing demand for resources transforms tech giants into resource-intensive industrial businesses, complicating their sustainability efforts and prompting concerns from local communities about water scarcity, as seen in cases with Amazon, Google, and Meta.
