"CCTV Finance" reported that if a user asks 10 to 50 questions to ChatGPT, 500 ml of water may be consumed.
Because training and running large AI models usually requires tens of thousands of servers to provide computing resources, storage resources, and low-latency network connections, these servers will be deployed in clusters in physical locations called "data centers." When they When working together, highly concentrated heat is generated in a short period of time, so a large amount of water is required for cooling.
Take Microsoft and Google for example. In 2022, Microsoft used a total of about 1.7 billion gallons (about 6.4 billion liters) of water, which is equivalent to filling about 2,500 Olympic-sized swimming pools; while Google's data centers and offices used a total of 5.6 billion gallons (about 21.2 billion liters) of water, equivalent to 8,500 Olympic-sized swimming pools.
Data shows that the computing power required by AI is now expected to double every 100 days, and may increase more than a million times in the next five years. Researchers predict that by 2027, global AI demand may require 6.6 billion cubic meters of water, almost equivalent to the annual water withdrawals of Washington State in the United States.
Ann Skeeter, senior director of the Markkula Center for Applied Ethics at Santa Clara University, said: "I think regulators will start to act soon, and the European Union will increase relevant regulatory requirements next year to require companies to disclose their Water usage.”