Water Consumption of AI: How Tech Giants are Draining the Planet 2024

Water consumption of AI: How tech giants are draining the planet

As the demand for artificial intelligence (AI) products and services increases, so does the need for cooling down the data centers that power them. This means that the world’s biggest technology companies, such as Microsoft, Google, and Meta, are using more water than ever before, raising environmental and social concerns.

How much water does AI use?

According to a paper published in Nature this week, the water consumption of AI could reach between 4.2 billion and 6.6 billion cubic meters by 2027, which is about half of what the UK uses in a year. This is because AI models, especially the ones that use generative AI, require huge amounts of computing power and data to process and generate text, numbers, and other information.

To prevent overheating, these models run on massive server farms that use chilled water to absorb heat from the air. Some of this water evaporates, while some can be recycled. However, this still puts a strain on the water resources, especially in areas where water is scarce or polluted.

Water is also used in other forms of energy production, such as oil and gas extraction, thermal power plants, and hydroelectric dams. Therefore, the water consumption of AI is not only a direct result of cooling down the data centers but also an indirect result of the electricity they consume.

What are the impacts of the water consumption of AI?

The water consumption of AI has several negative impacts on the environment and society. First, it contributes to the global water crisis, which affects more than two billion people who lack access to safe drinking water. Second, it worsens the effects of climate change, such as droughts, floods, and wildfires, which threaten the livelihoods and health of millions of people. Third, it creates conflicts and inequalities, as some communities and regions have to compete for or share the limited water resources with the tech giants.

For example, in West Des Moines, Iowa, a data center cluster that hosts OpenAI, one of the leading AI research organizations, consumed 6 percent of the district’s water in a month, according to a lawsuit filed by its residents. They claimed that the data center was depleting the aquifer and lowering the water pressure, affecting their quality of life and property values.

Another example is the ChatGPT chatbot, a popular AI service provided by OpenAI, which uses its older model GPT-3. Shaolei Ren, an associate professor at the University of California, Riverside, estimated that asking the chatbot for 10 to 50 responses would use as much water as a 500ml bottle, depending on the location and time of the request. The newer model, GPT-4, which has more parameters and power, would likely use even more water, but the exact amount is unknown.

What are the solutions to the water consumption of AI?

The tech giants that use AI have acknowledged the problem of the water consumption of AI and have set some goals and initiatives to reduce it. For instance, Microsoft, Google, and Meta have pledged to replenish more water than they use by 2030, by supporting projects that improve water efficiency, conservation, and restoration. They have also invested in renewable energy sources, such as solar and wind, to lower their carbon footprint and electricity demand.

However, some experts and activists argue that these measures are not enough and call for more transparency and accountability from the AI industry. They urge the AI firms to disclose more data and details about their water consumption and environmental impact, such as how much water is used by different AI services and models, and how they compare to other industries and sectors.

They also suggest that AI users and developers should be more aware and responsible for the water consumption of AI, and avoid using or creating AI tools that are unnecessary, wasteful, or harmful.

Kate Crawford, a research professor at USC Annenberg who specializes in the societal impacts of AI, said: “Without better transparency and more reporting on the issue, it’s impossible to track the real environmental impacts of AI models. And this matters at a time when many parts of the planet are experiencing deep and extended droughts, and fresh drinking water is already a scarce resource.”