A few years and China’s data centers will require as much water as the entire population of South Korea

Data centers already require an incredible amount of water, and will only increase as time goes on.

According to a new report by the Hong Kong-based non-profit organization China Water Risk, China may soon use 1,298 billion liters of water for its data centers, which is equivalent to the water consumption of roughly 26 million people. South China Morning Post. This number is expected to rise to 2,998 billion liters by 2030, equivalent to the water needs of the entire population of South Korea.

Training and maintaining AI models is notoriously energy-intensive and requires a lot of water. This is because a huge amount of heat is produced in data centers, and in order to avoid overheating, water is used to cool the hardware – a dizzying amount of water.

The advance of artificial intelligence naturally results in increased energy and water consumption not only in China. According to experts, Microsoft used 700,000 liters of water to train GPT-3, a large language model developed in the USA (with which ChatGPT also works), which is enough to cool an entire nuclear reactor. US-based Google also admitted in its 2023 environmental report that it alone used an astronomical amount of 20 billion liters of water in 2022.

It is worth considering whether it is really worth using chatbot services, as they require significantly more energy than traditional online information search methods. According to a report by China Water Risk, 100 million users chatting with ChatGPT uses the equivalent of 20 swimming pools of water, compared to a simple Google search that shows “only” 1 swimming pool of water for 100 million users.