AI Could Trigger a New Water Crisis Around the World
Why does AI technology require such an enormous amount of water? And what does this mean for sustainability and the environment with this type of technology?
On top of the already-high electricity demands of AI systems, new research shows they also consume enormous amounts of water. According to the International Energy Agency, energy consumption for global data centers could more than double by 2026, meaning it could reach levels exceeding the energy use of major nations. While AI is being used to solve some of the world's biggest problems, like climate modeling and medical breakthroughs, the technology, ironically, is contributing to another kind of environmental crisis.
It was recently reported that OpenAI's ChatGPT, which uses the GPT-4 language model, consumes 519 ml of water - equivalent to a small bottle - to write a 100-word email. This came out of an investigation by the Washington Post in collaboration with the University of California, Riverside. Extrapolated over a year, that means a user sending one such email a week would use about 27 liters of water.
Why Is Water Needed?
Every time a user types a question into ChatGPT, it's quickly converted to code and processed through servers in large data centers around the world. These servers require enormous amounts of computation to deliver the best possible answer. That computation generates heat, and to prevent overheating, data centers use cooling systems often based on water cooling. The amount of water and electricity required varies depending on the data center's geographic location and climate.
Cities like West Des Moines, Iowa, are becoming increasingly attractive for data centers thanks to a temperate climate that reduces the need for water-intensive cooling. Despite this, global water use is rising rapidly. Between 2021 and 2022, Microsoft's global water use rose by 34% and Google's by 20%, with a large share of that increase attributable to AI development.
Consequences of Rising Water Use
Although AI hasn't yet led to a direct shortage of essential water resources, the rising use is concerning. Today's rapid population growth and shifting climate are worsening problems with depleted water resources and aging water infrastructure. Researchers like Shaolei Ren at UC Riverside argue that while the absolute amount of water AI models use matters, it's just as critical how the developers of these models manage the global water crisis.
Companies using AI, especially generative AI models, need to take responsibility and consider their impact on local water resources. Rising tensions between data centers and local communities over water use are already a reality, and if the trend continues, it could lead to social unrest.
AI's Role in Environmental Impact
AI's rising water consumption is just one part of a broader environmental impact. Analysts say advanced technologies like AI and machine learning could increase fossil fuel extraction by 15%, which could delay the transition to renewable energy. One example is a deal between Microsoft and ExxonMobil that, according to estimates, could generate emissions exceeding Microsoft's annual carbon offsets for 2020 by more than 600%.
Holly Alpine, former program manager for environmental sustainability at Microsoft's data centers, resigned this year over the company's environmentally unsustainable AI development. She warned that such deals could have enormous real-world consequences for the climate.
Summary
The rapid rise of AI technology undoubtedly has many benefits, but it also comes with a significant environmental cost. Reducing the negative effects on water resources and climate will require tech companies to take greater responsibility for their operations and how they affect our planet.
Source: https://fortune.com/2024/09/23/ai-water-usage-droughts-chatgpt-microsoft/