China is making significant strides in AI, especially considering that it, for the most part, was under severe sanctions from the US. However, as commendable as China’s development of its AI model is, its rapid expansion of data centers to support AI initiatives raises concerns about water consumption.
A recent report by China Water Risk revealed the staggering amount of water used by data centers in China. What’s more worrying, though, is that projections indicate a dramatic increase in water usage by 2030.
China’s data centers consume approximately 1300 billion liters of water a year, equivalent to the residential water use of about 26 million people in China. By 2030, however, this number could skyrocket to 3000 billion liters, significantly reducing water resources. To put that in perspective, the report claims that the entire country of South Korea won’t be using as much water as China’s data centers.
The water demand is primarily driven by the need to cool down the hardware used in training and maintaining AI models, which generate substantial heat.
China plans to triple the number of data centers by 2030 and deploy approximately 11 million racks.
It is not just China that wants to expand its AI models. The United States of America is also witnessing a significant surge in water consumption, thanks to the ongoing AI boom. Companies like Microsoft and Google have disclosed staggering figures regarding their water consumption in AI-related activities.
Microsoft and OpenAI consumed about 700,000 litres just for training GPT-3. In its 2023 Environmental Repo, Googlert revealed that it had used up an astronomical 21.1 billion liters of water in 2022.
The energy-intensive nature of AI chatbots, powered by specialized chips, further exacerbates the water consumption issue. For instance, if 100 million people were to ask ChatGPT just one question each, that would consume water equivalent to 20 Olympic swimming pools.
The report claimed that asking Google the same thing without relying on AI would consume about 20 times less water.
Experts are sounding the alarm over AI’s escalating energy and water demands. Some experts also predict that data centers will consume more electricity soon than several nations.
To mitigate these concerns, breakthroughs in energy-efficient chip technology and innovative approaches to AI model training and power consumption are urgently needed.
Addressing these challenges will require concerted efforts from industry leaders, policymakers, and technology innovators to ensure the sustainable development and deployment of AI technologies without further straining global water resources.