By Ricardo Salazar, SERC Chile researcher – Pontificia Universidad Católica.

So far in 2024, have you asked a question or requested information from ChatGPT, Bard, or Gemini? Artificial intelligence is transforming numerous fields significantly, becoming an indispensable ally for people. Among its qualities, it optimizes processes, boosts creativity, and facilitates decision-making. However, caution is required: every time you ask five or more questions to ChatGPT, at least half a liter of water is consumed.

You may be wondering why, and the reason is that for AI to respond to queries, perform translations, provide summaries, or other tasks, it requires data centers housing millions of powerful computers that process millions of calculations and requests in real time. These data centers consume more than 1% of the planet's total energy.

Microsoft’s latest 2024 sustainability report demonstrated that the company's water usage reached nearly eight billion liters during 2023, double the consumption reported in 2020; this is equivalent to the total annual water consumption (residential and industrial) in cities such as Valparaíso or Concepción, according to the 2022 and 2023 consumption reports from the Superintendency of Sanitary Services (SISS).

Google also reports high figures. In its latest sustainability report, it indicates that it consumed more than 2.2 billion liters of water. In both cases, consumption has increased over the years and is projected to continue rising. Compounding this increase in water usage is the fact that the water used as a coolant is fresh or potable water, much of which evaporates during the cooling process, preventing its reuse. Furthermore, many of these data centers are located in water-scarce regions. However, water consumption is not limited solely to cooling. Artificial intelligence also has a massive water footprint, as its manufacturing and implementation require thousands of liters of water. As mentioned, data centers house millions of computers, and the production of hardware—which includes physical components such as processors, RAM, hard drives, and motherboards—requires intensive mining of minerals such as boron, phosphorus, gallium, and germanium. The extraction of these minerals has a significant impact on the environment and contributes to water pollution. Subsequently, semiconductors and microchips require large volumes of water during manufacturing. Thus, the associated water footprint is added for every component involved, such as a cable, a sensor, or a label.

In the future, and as AI becomes more integrated into our daily lives, its water footprint will inevitably increase. Before this becomes unsustainable, I would like to propose for reflection and debate that we must begin considering alternatives that could be implemented today, such as the use of treated wastewater, the utilization of seawater, the construction of new data centers designed for lower water consumption, and the circular economy in manufacturing processes. For more information see here