Speaking on technology journalist Alex Heath’s Sources podcast, Altman acknowledged that he was citing this data from memory and that the figure might vary slightly.
However, he stated that the comparison accurately reflects how low the real-world water consumption of modern AI operations is compared to common agricultural practices.
Altman argued that widespread concerns regarding high water consumption in data centers are based on outdated cooling infrastructure and do not reflect the design of modern facilities. Stating that large-scale data centers today operate with closed-loop systems, Altman said these facilities require a similar amount of water for daily use as standard commercial office buildings.
The OpenAI CEO expressed that the common belief that a high amount of water is consumed per query does not align with the data and persists despite public corrections.
According to internal company estimates, the direct water footprint of an average ChatGPT prompt is approximately 0.32 milliliters.
However, environmental researchers and policy experts warn that it is difficult to make precise calculations because companies do not disclose standard data regarding water consumed indirectly through cooling methods, regional water resources, and electrical grids.
News Source: 12punto
