AI Data Centres Are Drinking Your Water
A 2026 UN University report puts hard numbers on AI's environmental impact: by 2030 its water use will match the basic needs of 1.3 billion people. Here is what every ChatGPT prompt really costs.
When you ask ChatGPT to write an essay, nothing physical seems to happen. A few seconds pass, text appears, you close the tab. But the AI environmental impact in 2026 is enormous and growing fast, and the latest numbers are hard to ignore. A UN University report released in June puts a figure on AI data centre water usage that should stop anyone mid-prompt: within a few years, the water cooling these machines could match the basic annual needs of more than a billion people.
What the UN University report actually found
On 3 June 2026, researchers at UNU-INWEH, the UN University's water institute, published an assessment of where AI's resource demand is heading. UN News covered the findings the next day. The headline projection is stark: by 2030, the water consumed to cool AI data centres is on track to equal the basic annual domestic water needs of every person in Sub-Saharan Africa, all 1.3 billion of them.
That water is not a side effect you can wave away. Data centres run hot, and the most common way to keep thousands of processors from melting is to evaporate clean water through cooling systems. The same water people drink, farm with and wash in is being poured into server halls so that chatbots can answer faster.
The electricity numbers are just as heavy
Water is only half the story. The report projects that AI data centres will consume around 945 terawatt-hours of electricity, roughly triple the combined annual electricity use of Pakistan, Bangladesh and Nigeria put together. That is three countries, about 650 million people, outdrawn by server farms running models.
This is already reshaping real places. In Ireland, data centres now account for around 21% of all metered electricity, more than every home in the country's urban areas combined. The strain got serious enough that Dublin paused approvals for new data centres until 2028. A national grid is effectively being rationed so that the machines can keep growing.
A single image costs a thousand times more than text
It helps to see where the demand comes from. One widely-used AI service processes roughly 2.5 billion prompts every single day. Each one is small on its own, but multiply a small cost by billions and you get the figures above.
And not all prompts are equal. Researchers have found that generating a single AI image can need around a thousand times more energy than a simple text classification task. So the casual habits add up quickly:
- A throwaway question to a chatbot: small, but billions of them daily.
- A generated image for a slide or a meme: up to 1,000x the energy of basic text work.
- A full essay or report drafted by AI: many prompts, many revisions, all drawing power and water.
- An entire cohort of students doing this every assignment season: a measurable load on the grid.
Your assignment is part of the demand
It is tempting to think this is a problem for tech companies, not students. But the demand curve is built from ordinary use. Every ChatGPT-written assignment, every AI-generated diagram, every humanised draft run through three different tools, adds to the prompts that justify the next data centre. The next one that needs more water and more power from a grid that is already stretched.
You did not cause the global build-out by asking for an essay. But the honest accounting is that mass AI use in education is one of the reasons demand is climbing this fast. Singapore students, like everyone else, are quietly part of the count.
“We don't run a single server. Your assignment is written by a person at a desk, not cooled by a river. Zero AI means zero data-centre load.”
The human alternative has a footprint of roughly zero
Here is the part that makes us a little proud. When a real graduate writes your essay by hand, the environmental cost is a person, a laptop and a cup of coffee. No model is queried. No GPU farm spins up. No water evaporates to cool a server rack on your behalf. The footprint is the same as it has been for as long as people have written essays.
That is not why CodedByHumans exists, we built it so your work is genuinely human and impossible for Turnitin to flag, but it is a real bonus. Choosing a person over a machine happens to be the lighter choice for the planet too.
If you want work that is Turnitin-safe, written by a real Singapore graduate, and adds nothing to the AI water bill, message us on WhatsApp for a free quote. Real grads, zero AI, nothing to detect, because nothing is AI.
Frequently asked questions
How much water does AI data centre cooling actually use?
According to a June 2026 UN University (UNU-INWEH) report, AI's water consumption is projected to equal the basic annual domestic water needs of all 1.3 billion people in Sub-Saharan Africa by 2030. Most of that water is used to cool the processors inside data centres, often by evaporating clean water through cooling systems.
Why does generating an AI image use so much more energy than text?
Image generation runs far larger and more complex computations than a simple text task. Research cited in the 2026 reporting found that producing a single AI image can require around a thousand times more energy than a basic text classification, which is why heavy image and video use drives data centre demand so sharply.
Is one student using ChatGPT really an environmental problem?
A single prompt is tiny. The issue is scale: one widely-used AI service alone handles about 2.5 billion prompts a day. Mass everyday use, including assignments, is a major reason data centre water and electricity demand is climbing, and that collective demand is what justifies building more facilities.
Does using a human writing service avoid this footprint?
Largely, yes. When a real graduate writes by hand, no AI model is queried and no data centre spins up to serve your task, so there is effectively no extra water or GPU electricity cost. At CodedByHumans the work is done entirely by people, which means zero AI and zero added data-centre load.
Need this done — by a real human?
Essays, code, and software written by graduates, never by AI. Turnitin-safe, guaranteed.