The rapid growth of generative AI, like ChatGPT, comes with an unexpected consequence: a substantial environmental impact. A recent study by The Washington Post and researchers from the University of California, Riverside, has shed light on the significant water and electricity demands associated with even the most basic ChatGPT functions.
The study found that the amount of water required for ChatGPT to draft a 100-word email varies considerably depending on the location and proximity to OpenAI’s data centers. Regions with limited water resources and cheaper electricity often rely on electrically powered air conditioning, leading to higher water consumption. For example, generating a 100-word email in Texas requires an estimated 235 milliliters of water, while the same task in Washington state consumes a whopping 1,408 milliliters.
The increasing scale of AI data centers, driven by the demand for generative AI technologies, has pushed traditional air-cooling systems to their limits. To combat the heat generated by these powerful machines, many data centers have transitioned to liquid-cooling systems. These systems pump massive amounts of water past the server stacks to draw off thermal energy, which is then released into cooling towers. This shift to liquid cooling significantly increases water consumption.
ChatGPT’s electricity demands are equally alarming. The study found that drafting a 100-word email with ChatGPT consumes enough electricity to power over a dozen LED lightbulbs for an hour. If even a small fraction of Americans used ChatGPT to write this email once a week for a year, the total energy consumption would be equivalent to the entire annual power usage of all Washington, D.C. households combined.
This environmental impact is not unique to ChatGPT. Other AI models, such as Meta’s Llama 3.1, required 22 million liters of water for training. Google’s data centers in The Dalles, Oregon, were found to consume nearly a quarter of the town’s available water, according to court records. Meanwhile, xAI’s new Memphis supercluster demands a staggering 150MW of electricity, enough to power 30,000 homes.
This growing environmental footprint of AI is a serious concern, and it’s unlikely to improve anytime soon. As AI technology continues to advance and its applications expand, the demand for computational resources, including water and electricity, will only increase. Addressing this challenge requires innovative solutions and collaborative efforts to minimize the environmental impact of AI while harnessing its transformative potential.
