It's only getting worse.

ThirstGPT

China is doubling down on AI, focusing its efforts on opening vast new data centers — which consume a staggering amount of water.

According to a recent report by Hong Kong-based non-profit China Water Risk, the country could soon be consuming around 343 billion gallons of water in its data centers, or the equivalent of the residential water use of 26 million people.

By 2030, that number could rise to a whopping 792 billion gallons — enough to cover the needs of the entire population of South Korea, as the South China Morning Post reports.

Down the Drain

Training and maintaining AI models is an infamously energy-demanding task that generates a huge amount of heat. To keep data centers from overheating, companies use water to cool down the hardware.

According to China Water Risk, China could triple the number of data centers by 2030, reaching roughly 11 million data center racks that house servers and other equipment.

And it's not just China. The AI boom is already leading to an astronomical amount of water being used elsewhere, including in the US. Last year, researchers found that just in training GPT-3 alone, OpenAI partner Microsoft consumed a whopping 185,000 gallons of water, which is enough to cool an entire nuclear reactor. Google also admitted in its 2023 Environmental Report that it had used up an astronomical 5.6 billion gallons of water in 2022.

AI chatbots are vastly more energy-intensive than conventional methods of finding information online, relying on specialized chips that consume huge amounts of electricity.

Per China Water Risk's report, if 100 million users were to chat with OpenAI's ChatGPT, it would consume the equivalent of 20 Olympic swimming pools. Doing the same via simple Google searches would only "consume one swimming pool."

All that usage could have devastating effects on parts of the world where water resources are already extremely scarce. Last month, The Atlantic reported that Microsoft was trying to cover up the water usage of its data center in the Arizona desert.

Keep Cool and Carry On

And the AI hype cycle isn't quite over yet. Experts are worried about the further rise of energy and water usage.

Arm Holdings Plc CEO Rene Haas told Bloomberg this week that he expects the world's data centers to use more electricity than India, the world's most populous nation, by the end of the decade.

"We are still incredibly in the early days in terms of the capabilities," he said, ringing the alarm bells over AI eclipsing our global energy demands.

To Haas, one meaningful step forward is to find new ways to train and power these AI models with more energy-efficient chips.

"There needs to be broad breakthroughs," he told Bloomberg. "Any piece of efficiency matters."

More on AI and water: Microsoft Is Draining Arizona's Water for its AI


Share This Article