How Tech Giants Contribute To Global Water Shortages

About two-thirds of the world’s population experiences severe water shortages for at least one month each year. By 2030, the situation is expected to worsen, with almost half of the world’s population facing severe water stress.

This prediction was made in a report published a few years ago by the United Nations Environment Programme. To avoid this fate, the report said, water use must be “decoupled” from economic growth by developing policies and technologies to reduce or maintain consumption without compromising performance.

The authors mentioned a few water-intensive sectors, such as agriculture. What they did not imagine, in 2016, is they should have added another source of consumption: artificial intelligence.

So far, researchers and developers have mainly focused on reducing the carbon footprint of AI models. However, a crucial aspect that has often been overlooked is their water footprint. In “Making AI Less “Thirsty”: Uncovering and Addressing the Secret Water Footprint of AI Models,” a new paper that has yet to be peer-reviewed, researchers from the University of Colorado Riverside and the University of Texas Arlington fill this gap by shedding light on the significant water consumption associated with training and deploying AI models in data centers.

Using public data sources, they estimate that “training GPT-3 in Microsoft’s state-of-the-art US data centers can directly consume 700,000 liters of clean freshwater”, which they calculate could be used to produce 370 BMW cars or 320 Tesla electric vehicles.

Furthermore, ChatGPT ‘drinks’ the equivalent of a 500ml bottle of water for a simple conversation of 20-50 questions and answers. Which may not seem like much… until you consider that the chatbot has more than 100 million active users, each of whom engages in multiple conversations.

And it’s not just Microsoft: when it comes to water consumption, Google is second to none. In 2021, its data centers in the US alone will consume 12.7 billion liters of freshwater for on-site cooling, about 90% of which will be potable water.

Overall, the combined water footprint of US data centers operations was estimated at 626 billion liters in 2014. To give credit where credit is due, it’s not as if the big tech companies are doing nothing to tackle the problem. Many of them, such as Amazon, Meta, Google and Microsoft, have pledged to become “water positive” by 2030 – meaning they will refill more water than they consume.

Unfortunately, as the study’s authors point out, there’s often a trade-off between carbon efficiency and water efficiency.

This is due to the fact that current approaches to achieving sustainable AI predominantly center on engineering solutions, such as enhancing the efficiency of data center cooling towers. While these supply-side solutions conserve water, they fail to address the demand-side management aspects tied to the timing and location of AI model training and use.

“For example, AI model developers may want to train their models during the noon time when solar energy is more abundant, but this is also the hottest time of the day that leads to the worst water efficiency,” researchers write, using LaMDA’s training in sun-drenched Nevada as an example.

In other words, using renewable energy can sometimes get in the way of saving water.

The challenge, then, is to find a way to balance carbon and water efficiency, which will require new and holistic approaches to sustainable AI.

One potential solution lies in exploiting what researchers call the ‘spatio-temporal diversity’ of water use efficiency. In short, by scheduling AI model training and inference in different places and at different times, developers can reduce the water footprint of their AI models.

As the paper has yet to be peer-reviewed, it’s possible that some of its arguments and conclusions may need to be re-evaluated. Still, the researchers’ findings are impressive, and it’s hard not to agree with their final assessment: “AI models’ water footprint can no longer stay under the radar — water footprint must be addressed as a priority as part of the collective efforts to combat global water challenges.”