Enough to Fill 2,500 Olympic-Sized Pools – Rolling Stone

Charlie Neibergall/AP

Artificial Intelligence has its fans and detractors — with people sounding the alarm on the social implications and other dangers that were evident long before ChatGPT existed while others praise the revolutionary technology for its potential for immortality and more debate in between. While the conversation about the societal costs of AI continue, the environmental costs are also tallying up, and one vital resource is being heavily tapped to produce the technology: water.

In Microsoft’s latest environmental report, the company revealed that its global water consumption spiked by more than a third from 2021 to 2022, the equivalent of which could fill more than 2,500 Olympic-sized swimming pools at nearly 1.7 billion gallons, as The Associated Press reports.

Meanwhile, in Google’s environmental report, the company was up in water use by about 20 percent.

In order to cool the supercomputer that helped teach its AI systems, Microsoft-backed OpenAI used water from the Raccoon and Des Moines rivers in Iowa where ChatGPT was built. The company is not alone in its race for generative AI, with OpenAI and Google also acknowledging that the demand for AI increases water consumption and requires expensive semiconductors.

The computing necessary to build a large language model that analyzes patterns of human-written text takes massive amounts of energy and creates a lot of heat. To cool things down, data centers have to pump in water. In Microsoft’s case, it disclosed that when Iowa’s temperature exceeds 85 degrees Fahrenheit, that is when the supercomputer will draw water so it can run properly and vent heat out. Much of the year, the weather is below that temperature so no water is drawn.

Shaolei Ren, a researcher at the University of California, Riverside is working to tabulate the environmental impact of generative AI tools, such as ChatGPT.

Ren told AP, “Most people are not aware of the resource usage underlying ChatGPT. If you’re not aware of the resource usage, then there’s no way that we can help conserve the resources.”

In a paper that is expected to be published later this year, Ren and his team estimate that Chat GPT chugs around 500 milliliters of water every time it’s asked a series of between 5 to 50 prompts or questions.

In a statement to AP, Microsoft said it was investing in research to measure AI’s energy and carbon footprint. “We will continue to monitor our emissions, accelerate progress while increasing our use of clean energy to power data centers, purchasing renewable energy, and other efforts to meet our sustainability goals of being carbon negative, water positive and zero waste by 2030,” the company’s statement said.

OpenAI said in a statement: “We recognize training large models can be energy and water-intensive” and said it will work to improve efficiencies.