How much water does Chat GPT drink because you want to talk

OpenAI CEO Sam Altman recently admitted that human politeness costs him a lot of money. Yes, when you write “please” and “thank you” to your chatbots, your electricity bill increases. One user asked: How much money OpenAI loses because people talk to AI as if they were living people. Altman did not hold back: “tens of millions of dollars – but not in vain.” And then added: “You never know…” (Maybe these “thank yous” will save us in the future?).

Here’s how artificial intelligence answers this question:

As an artificial intelligence, I don’t directly consume water in the usual sense — I don’t drink, I don’t take a shower, and I don’t grow plants. However, my operations are connected to data centers, which consume water primarily to cool servers.

Rough estimates:

In 2023, a study showed that a single query to a large language model (e.g., GPT) could indirectly require up to 500 ml of water — that’s about as much as a person drinks in one sitting. For example, data centers often use water that evaporates in cooling systems to cool servers. Microsoft, one of OpenAI’s partners, reported in 2022 that it used about 6.4 billion liters of water per year, and that’s due to cloud services and AI.

Why this matters:

When we discuss sustainability, AI’s ecological footprint encompasses not only its electricity consumption but also its water footprint. This becomes especially relevant in regions suffering from a shortage of fresh water. The problem of AI’s water footprint is new, but already important. There are several areas where we can work to reduce our impact:

Switching to water-independent cooling technologies.

Air cooling instead of water is less efficient in hot climates, but possible in cooler regions. Innovative closed-loop systems where water is not evaporated but recirculated.

Data center location selection.

Locating data centers in cold regions (e.g. Scandinavia), where natural cooling reduces water demand. Avoid building such centers in arid regions where water is a critical resource.

Water reclamation and use of “grey” water.

Using treated wastewater instead of drinking water for cooling.Technologies that allow water to be reused in cooling systems.

Optimization of AI models.

Less resource consumption through energy-efficient algorithms. Training smaller models or using already trained ones without having to run large models each time.

Transparency and accountability.

Companies should publish reports on water consumption, just as they do with CO₂ emissions. This allows the public and scientists to monitor and influence sustainability policies.