Asking an AI a question, we typically assume that an answer pops up instantly on our screen. However, the process of delivering an answer happens with the help of high-performance computers inside data centres. Since such computers consume energy and create heat, the data centres require some kind of cooling system. In such a case, AI can have water footprints since the system uses water to cool down the data centre.

The problem is that there is no certain amount of water consumed per each question asked to AI. The amount of water depends on the AI model, hardware, data centre, its cooling system, climate conditions and the source of electricity. Thus, it is possible to get vastly different numbers depending on the study.

For example, according to a 2025 study conducted by Google, the median consumption of water for answering Gemini text prompt was 0.26 ml, which is comparable to 5 drops of water.

So, does querying the AI use water? Yes, possibly  but the quantities will be small for any single text input. What seems to matter most is the total quantity of queries made when billions of requests are processed through AI.

Not only is water related to cooling. The analysis of AI's water footprint includes water consumed through data-centre cooling as well as water that goes into generating the electricity powering computing infrastructure.

It needs to be noted that AI is getting increasingly efficient in its energy consumption. According to the International Energy Agency, in 2026 the electricity use per simple AI task had dropped significantly; however, some new types of AI applications, including AI agents, reasoning, and video generation, require significantly more energy than text generation alone.

Hence, the viral meme about “one AI question using one bottle of water” is a simplification. AI does have water and energy footprint, but the magnitude of one query is highly variable. The important question is not how much water one query consumes, but what is the resource cost of AI usage worldwide.