ChatGPT's Rising Resource Demands: Why They're Becoming Unmanageable

The rapid expansion of generative AI has significantly increased its demand for water and electricity. A recent study by The Washington Post, alongside researchers from the University of California, Riverside, reveals the substantial resources that OpenAI's ChatGPT requires to carry out even basic tasks.

When it comes to water consumption, the amount needed for ChatGPT to draft a 100-word email varies by location and proximity to OpenAI’s data centers. In regions with limited water availability and lower electricity costs, data centers often rely more on electrically powered air conditioning systems. For instance, in Texas, generating a single 100-word email requires about 235 milliliters of water. In contrast, the same email crafted in Washington demands a staggering 1,408 milliliters, nearly one and a half liters.

As generative AI technology has driven the growth of data centers, these facilities have become larger and more densely packed. Air-based cooling systems struggle to maintain optimal temperatures, prompting many AI data centers to transition to liquid cooling. This approach circulates large volumes of water around server stacks to absorb thermal energy before sending it to cooling towers where the heat is released.

The electrical demands of ChatGPT are equally significant. According to The Washington Post, composing a 100-word email with ChatGPT requires enough electricity to power over a dozen LED lightbulbs for an hour. If just 10% of the U.S. population used ChatGPT once a week for a year, the total energy consumption would equal that of all households in Washington, D.C. for 20 days, impacting approximately 670,000 residents.

This challenge is unlikely to resolve quickly and may exacerbate over time. For example, Meta required 22 million liters of water to train its latest Llama 3.1 models. Google’s data centers in The Dalles, Oregon, consume nearly a quarter of the town's water supply, according to court documents. Meanwhile, xAI's Memphis supercluster is already tapping into 150MW of electricity—enough to support approximately 30,000 homes—demanding resources from the local utility, Memphis Light, Gas and Water.