According to the Washington Post, a new study from the University of California, Riverside, reveals thatGenerative AI The staggering environmental costs of AI, especially its consumption of water. Studies have found that AI requires large amounts of water to cool servers, even when just generating text. This, coupled with the enormous strain it puts on the power grid, further emphasizes the environmental challenges of AI development.
The study notes that AI's water usage varies by state and data center location, but overall, water consumption is inversely proportional to electricity costs.Texas has the lowest water consumption, requiring only 235 milliliters of water to generate a 100-word email, compared to a whopping 1,408 milliliters in Washington, the equivalent of three 16.9-ounce bottles of mineral water.
In addition, data centers themselves are large consumers of utilities, which pushes up utility costs for residents in the areas where they are located. For example.Meta used 22 million liters of water to train its LLaMA-3 model.That's equivalent to the amount of water needed to grow 4,439 pounds (IT note: about 2013.5 kilograms) of rice, or the amount of water used by 164 Americans in a year. That's no small amount, especially considering that it's only for the GPT-4 Very light use scenarios for the target audience.
The cost of electricity from GPT-4 is also very high; if one in ten U.S. commuters used GPT-4 once a week, over the course of a year, their electricity demand would be equivalent to 20 days of electricity consumption for the entire Washington, D.C., area (approximately 671,803 people).
Representatives from OpenAI, Meta, Google, and Microsoft responded, mostly just reaffirming their commitment to working to reduce environmental demand, without giving specific courses of action. Microsoft representative Craig Cincotta said the company is committed to developing cooling methods that do not consume water at all, but did not provide specifics.