AI’s hidden water cost, and why it matters

What is wrong with using AI to turn yourself into a 1980s movie star? Nothing, apparently. Type a prompt, upload a photo, and within seconds, you have a glamorous, grainy portrait of yourself that looks as though it belongs on the cover of an old film magazine. It is nostalgic, effortless, and, most importantly, fun.

It is a privilege to enjoy something without ever having to ask what it takes. However, AI does not generate these retro images out of thin air. Behind the seemingly effortless “generate” button are massive data centres packed with servers that require vast amounts of electricity and water to operate. Because this infrastructure remains largely out of sight, the environmental cost of using AI is easy to overlook and even easier to ignore.

Tap here to add The Daily Star as a trusted source

We use AI to write emails, finish assignments, brainstorm projects, summarise books, make presentations, edit photographs, generate videos, and create memes. We ask it to rewrite one sentence in ten different ways. We generate an image, dislike it, generate another, change the lighting, clothes, or background, and generate again. Each command feels insignificant. But collectively, they are not.

Generative AI requires enormous amounts of computing power. The specialised chips running these systems consume electricity and produce heat. The cooling systems used to manage the heat consume significant amounts of water. Water is also used to generate electricity that powers data centres. The more computing-intensive the task, the greater the resource demand.

So, how much water does one prompt actually use? There is no single universal number. It depends on the model, the data centre, the cooling technology, the weather, the electricity source, and even the length and type of the request. One widely cited study estimated that GPT-3 could consume roughly 500 ml of water for every 10 to 50 medium-length responses. However, Google reported in 2025 that a median text prompt on its Gemini apps used about 0.26 ml of water, roughly five drops. Google also acknowledged that water consumption varies depending on the infrastructure and methodology used to measure it.

The disagreement is part of the problem. AI companies still don’t provide enough standardised, transparent information for the public to understand the water footprint. One recent estimate puts the global water footprint of AI at between 4.2 and 6.6 hundred crore cubic metres annually by 2027.

The world is already struggling with water. In 2024, 220 crore people still lacked safely managed drinking water, according to the United Nations. The UN also reports that 340 crore people lacked safely managed sanitation. A 2026 report from the United Nations University says the world has moved beyond a conventional water crisis into what it calls “global water bankruptcy”.

This is hardly an abstract concern for Bangladesh. Water scarcity, groundwater depletion, and salinity already affect communities across the country. Against that reality, the question is not whether we should stop using AI. That would be neither realistic nor particularly useful. AI can improve education, research, healthcare, and productivity. The problem is an industry that makes computation feel weightless while its physical infrastructure remains hidden, and a culture that encourages us to generate endlessly because the marginal cost to the user appears to be zero.

We need better disclosure from AI companies, more water-efficient data centres, greater use of recycled and non-potable water, and serious limits on the siting of massive data centres in water-stressed regions. Companies should be required to disclose how much water their systems consume, while governments should treat water availability as seriously as electricity when approving new data centres. Users, meanwhile, can ask a simple question before pressing generate: do I actually need to use AI for this?

The question, then, is not whether we will use AI, but whether we are willing to acknowledge what that convenience consumes. Progress cannot mean building technologies that make consumption seem invisible. The environmental costs behind AI must be visible, measurable, and shared by the companies profiting from its expansion.

Hurmetun Nesa Labiba and Fatema Tuz Zuhra are research associates at Bangladesh Institute of Governance and Management (BIGM).

Views expressed in this article are the author's own. 

Follow The Daily Star Opinion on Facebook for the latest opinions, commentaries, and analyses by experts and professionals. To contribute your article or letter to The Daily Star Opinion, see our guidelines for submission.