A Thirst for Progress

An unlikely consequence of AI’s reign, water consumption is at critical levels.

As someone who has used AI extensively, whether it be for a curious search, math problem, or just a short, simple sentence to help me call out of work, I understand its convenience and appeal. Seeing firsthand how a task that could have taken hours can be reduced to minutes or seconds, it's easy to see AI as a gift from the tech gods, but all that convenience hides a growing environmental cost that most users, including myself, have never even thought to consider. The amount of energy and resources needed to upkeep AI’s fast pace development and large scale use is sapping the US of its natural water supplies at a rapid rate.

According to Sam Altman, the Ceo of Open AI, a single query uses about 0.34 watt-hours of electricity and roughly 0.000085 gallons of water, which translates to just about the amount of energy an oven would use in a second and a little less than a teaspoon of water. Alone, this may seem insignificant, but since AI has reached the power to operate on a global scale, billions of requests are sent through and processed every day. This amount boils down to around 8854 gallons of water an hour, which is enough to fill a standard-sized swimming pool, or 35 bathtubs to the brim.

Behind these responses, there are data facilities scattered all over the globe, numbering up to 11,800 total and more than 5,400 located in the United States, according to Brightlio, a company that assists and monitors other datacenters. These centers run 24/7 and rely on water cooling systems to keep their servers running, which can consume up to 3 to 5 million gallons of water per day, causing strain on local water supplies. States like Oregon, Virginia, and Arizona are the most pressured of the bunch, and data centers have already begun competing with local communities for their dwindling water supply. Google's data center in Dalles, Oregon, contributes to a quarter of the city's water usage and is affecting the livelihood of agricultural workers.

The demand for these data centers is only growing as AI continues to advance. According to a report from Lawrence Berkeley National Laboratory, they found that data centers’ water use has increased 21.2 billion liters from 2014 to 66 billion liters in 2023. Though they are trying to find ways to improve efficiency, their total water consumption will continue to scale up with time and further strain on water deprived areas.

Total global water scarcity is becoming more and more prevalent, and the United Nations now estimates that more than 2 billion people live in droughted areas, and that 4 billion people experience water scarcity at least once a month. By 2030, there may be a 56 percent gap between freshwater supply and demand, and AI’s severe use is a leading factor in that statistic.

Companies are aware of this challenge, and companies like Nortek are developing solutions to reduce water usage by transitioning to water evaporation techniques, which involve recycling water into a pad, and blowing small amounts of evaporated water through a fan to cool the air blowing on the system. However, efficiency won't solve this problem, with the demand for coolant increases as AI’s booming industry continuously grows. The water used for the cooling process is oftentimes evaporated or discharged, meaning it is not fully recoverable. In a world where 1.8 billion people already live in water scarce areas, this begins to raise concerns on the weight of morals in the pursuit of innovative technology.

SpaceX’s concept of orbital data centers could represent a major shift in the development and sustainability of computing infrastructure. By sending data centers out to space, their systems could harness near constant solar radiation, reducing the amount of energy needed from data grids on earth, and therefore reducing the amount of coolant needed. In the vacuum of space, heat can simply be dissipated through radiation, meaning that water cooling systems are no longer needed. This solution however, is only theoretical right now, and shows how far we are from overcoming the global or even planetary consequences that AI is bringing.

The development of AI has brought with it some undeniable benefits and is only continuing to develop more for our convenience. But there is an argument to be had about the physical impact it has on environments and people's livelihoods. If the sustainability of AI depends on these increasingly scarce resources, then it is our responsibility as users and developers to make sure that convenience doesn't come at a cost to remote areas and eventually ourselves.