Artificial intelligence has the property of appearing invulnerable. Answers appear instantly on our screen, without smoke, without noise, without any noticeable consumption of resources. And yet, every time we write a “please” or a “thank you” on ChatGPT, fresh water evaporates somewhere on our planet. Not figuratively, but literally for real, because behind the feeling of intangible, almost magical artificial intelligence, there is a heavy material reality: huge data centers, power-hungry processors and “thirsty” cooling systems.

Sam Altman, CEO of OpenAI, recently admitted that human kindness towards chatbots – the extra words, the little social habits that we carry from our daily lives into the digital world – cost his company tens of millions of dollars in computing power and energy. What is not often mentioned is that this cost is not only in money, but also in natural resources, with water at the center. So, as the use of artificial intelligence grows around the world, a nagging question is starting to emerge: how “hungry” is our digital comfort really – and how willing are we to pay the price, even when we can’t see it?

To understand the relationship between a polite question and real-world environmental impacts, it’s useful to consider how large language models like ChatGPT work.

Every time a user enters a question – either “Please convert hectares to square meters”, or “Thanks, that helps!” or a simple “hello”, that text must be processed by powerful servers located in large data centers. These hubs are packed with GPUs and other specialized computing devices that consume a lot of electricity and generate large amounts of heat. This heat must be managed – and therein lies the influence of water. Many data centers use fresh water to cool servers directly or indirectly: the water evaporates in cooling towers to remove heat, or is used in power plants that power computers.

A Business Standard report on Altman’s remarks noted that, just from the computing power needed to process queries in a conversation where the user politely addresses the machine, saying “hello, how are you?” yes, thank you, and please,” OpenAI’s costs have reached “tens of millions of dollars.”

In an early study published in 2023, the conclusion was clear: a typical ChatGPT interaction with 10 to 50 responses could require about half a liter of fresh water, the equivalent of a water bottle. However, by mid-2025, some company executives expressed a different view. Altman himself is citing much lower water use per individual query, suggesting to the public that AI’s “thirst” for water is negligible, although environmental analysts argue that these claims downplay the much larger systemic water footprint when aggregated to global use.

But the individual water cost per query matters a lot. Independent researchers analyzing global data center water use even predict that by 2027, AI-attributable water demand could reach 4.2 trillion liters per year – a figure equivalent to the total annual freshwater withdrawal of entire countries like Denmark or about half the size of the United Kingdom. Viral posts and environmental commentaries have even repeated the claim that by 2027, AI’s total water use will equal the annual water use of all of New Zealand – a staggering figure for those trying to understand the scale of the potential impact.

If this equivalence is correct and can withstand scientific scrutiny, the key point is clear: AI’s water footprint will grow rapidly, in line with growth explosive use of AI, the construction of data centers and global digital activity.

When every “please” and “thank you” carries an additional computational load, and when billions of such interactions can contribute to the use of billions of liters of water, it challenges us to think not only about digital ways, but also about the real impact of the digital behaviors we take for granted. As artificial intelligence continues to become more and more integrated into our daily lives – from writing emails to writing poetry – understanding and managing its environmental footprint will be as important as being polite to a chatbot.

Source: prizrenpost