The hidden plumbing behind every prompt<\/p>\n
The servers that power Large Language Models (LLMS) and tools such as ChatGPT generate<\/a> enormous amounts of heat as processors run strenuous operations<\/a>. To keep them from overheating, data centers rely on water-based cooling systems. Said systems are vast industrial operations that consume significant volumes of freshwater. According<\/a> to a paper published in April 2023 by the University of California, each 100-word AI prompt is estimated to use roughly 519 milliliters, almost one small bottle of water.\u00a0<\/p>\n Far more computationally intensive than text, image generation pushes that figure significantly higher.\u00a0<\/p>\n Depending on the model and data center, a single AI-generated portrait can require<\/a> up to 100 times as much power as a basic text query. More energy means more heat, which requires more cooling, thus more water.\u00a0<\/p>\n Data centers rely on local water supplies, natural sources, and in some cases, seawater. The cooling methods include using<\/a> evaporative cooling or closed-loop systems. In the former, cold water absorbs heat from servers and exits the building as steam, requiring a constant fresh water supply. In the latter, hot water is chilled and recirculated instead, using little to no extra water but significantly more energy. Whether one method is better than the other has not been studied<\/a> due to undisclosed numbers.<\/p>\n At the global scale, the numbers become staggering. ChatGPT alone is reported to receive around 2.5 billion queries<\/a> in a single day, a figure that encompasses everything from homework help using text prompts to viral AI portraits, and each category carries a different environmental price tag. According<\/a> to a 2025 study by the Institute for Environmental Studies in the Netherlands, the total water footprint of AI systems could reach between 312.5 and 764.6 billion liters in 2025, equivalent to the global annual consumption of bottled water.<\/p>\n Invisible infrastructure, visible consequences<\/p>\n The surge in demand fueled<\/a> by viral trends like Studio Ghibli art style portraits, referring to Ghibli, a renowned<\/a> Japanese animation studio known for its soft, whimsical aesthetic, as seen in films like \u2018Spirited Away,\u2019 released in 2001, and \u2018My Neighbor Totoro,\u2019 released in 1988, and AI caricatures has significantly amplified AI\u2019s environmental impact. When a single prompt format goes viral, and millions of social media users attempt it within days, the aggregate load on data centers spikes<\/a> sharply, and so does water consumption.<\/p>\n According to a 2025 Ecolab study, a sustainability company that provides solutions and services related to water treatment, sanitation, and hygiene, found<\/a> that more than half of AI consumers were initially unaware of its massive water consumption for infrastructure and operation, laying bare how the world has embraced AI\u2019s conveniences while remaining largely blind to what powers them.<\/p>\n The recent caricature trend caused ChatGPT\u2019s website to go down for thousands of users, with reports peaking at around 13,000 outage reports<\/a> at a single moment. OpenAI CEO Sam Altman has previously urged<\/a> users to reduce their use during image-generation surges, in March of 2025, admitting that the high usage is straining technical infrastructure.<\/p>\n
![\"Shop](\"https:\/\/www.europesays.com\/ai\/wp-content\/uploads\/2026\/04\/shopify-banner.png\")