Article: AI is Drinking our Water
Writer: Graceana Hardy
It’s hard to find a part of the internet that hasn’t been affected by the expansion of AI. Google automatically completes an AI search when you use it, YouTube writes video summaries using AI, and apps like Instagram and Snapchat have rolled out their own AI chatbots. But climate scientists are now investigating whether that expansion has leaked into the real world, and what it could mean for climate change.
Image from Pexels
We often think of the internet and its apps as this non-tangible thing, something that only exists on our phones and computers. In reality, the World Wide Web would not function if not for physical infrastructure like data centers, satellites, and undersea cables. Data centers in particular are important for the average user's needs. Without them, you couldn’t send emails, store photos, or use any social media or streaming platform.
Data centers take the form of massive warehouses, each one running 24 hours a day, managing an average of 2000 to 5000 servers inside. The United States alone has 5,390 data centers. Whenever you ask ChatGPT a question, it sends it to one of its data centers, where its servers run thousands of calculations to determine the best answer. Like your computer when you have too many tabs open, those servers heat up and require water to cool them. The Washington Post worked with researchers from the University of California to understand how much freshwater water ChatGPT 4, the most recent model of ChatGPT, uses to write a 100-word email.
According to The Washington Post, asking ChatGPT-4 to write a one-hundred-word email consumes 519 milliliters of fresh water, a little more than your standard plastic bottle’s worth. If 1 out of ten working Americans (roughly 16 million people) do this once a week, they will use 435,235,476 liters of water. That is equivalent to the water consumed by all Rhode Island households for 1.5 days.
Some centers use large air conditioners to cool their systems. These require electricity to run. If you ask ChatGPT-4 to write the same email and it sends it over to a data center using electricity to cool its servers, then it will use up enough energy to power 14 LED light bulbs for 1 hour.
Image from Pexels
Why is this a problem? Electrical cooling systems raise the power bill of the residents of the town it's in. This doesn’t take into account the massive amount of energy used training these AI models either. Places experiencing water shortages are especially affected by the depleted water these data centers leave behind. Data from OEHHA reports that California has been slowly drying up since 1895 and has been in a drought since 2021. This has resulted in a lack of water to fight the fires they are currently experiencing. Furthermore, according to The Washington Post, Google’s data centers in The Dalles, a town near Portland Oregon, were using nearly a quarter of all the water available to the town.
So what can be done to help? On a personal scale, you can start by reducing your own use of ChatGPT and other similar models. According to the UN, “More than 190 countries have adopted a series of non-binding recommendations on the ethical use of AI, which covers the environment.” This, along with suggestions from Stack Infrastructure to recycle the water used by these data centers for industrial use, indicates that the discussion of how to tackle AI’s climate issue has started.