Every time you ask ChatGPT to write an email or generate an image, somewhere in the world, a data centre is quietly draining the equivalent of several bathtubs of water.
That’s not hyperbole—that’s the hidden reality of our AI-powered future.
A recent leaked internal document has shed light on something the tech industry would rather keep private.
Amazon, it turns out, has been strategically minimising public disclosure about the massive amounts of water its data centres consume.
The memo reveals that Amazon used 105 billion gallons of water—enough to supply nearly one million American households for an entire year.
Why Does AI Drink So Much Water?
When most people think about AI’s environmental impact, they think about electricity consumption. But there’s another resource being consumed at an alarming rate: water.
Modern data centres are essentially giant computers that generate enormous amounts of heat.
To keep these machines from overheating, data centres rely heavily on evaporative cooling systems—think of them as massive air conditioners that use water instead of just electricity.
These cooling systems can evaporate about 80% of the water they draw, which is significantly higher than typical residential water use.
The Bigger Picture: An Industry-Wide Problem
Amazon’s water consumption might grab headlines, but it’s not alone in this.
Google, Microsoft, Meta, and every other company building AI infrastructure face the same fundamental challenge: how do you keep massive computational systems cool without draining local water resources?
The numbers are staggering when you scale them up.
The UN Environment Programme estimates that AI-related water demand could reach billions of cubic metres globally by mid-decade.
We’re talking about a resource that’s already scarce in many parts of the world becoming even more contested.
The Ethical Minefield
This raises some uncomfortable questions about the AI revolution we’re all supposedly benefiting from:
- Who gets to decide that generating marketing copy or creating digital art is worth depleting community water resources?
- What happens when there’s a drought, and local authorities have to choose between keeping the lights on at a data centre or ensuring residents have enough water?
The leaked Amazon memo suggests the company is well aware of these concerns. Internal documents reportedly show strategic thinking about how to minimise public awareness of water usage figures, particularly as they launched their “Water Positive” initiative in 2022, promising to return more water than they use by 2030.
The Real Cost of AI Queries
Every interaction with AI systems comes with a hidden environmental price tag that extends far beyond electricity consumption.
When you ask an AI to write a blog post, analyse data, or create an image, you’re not just using computational power—you’re consuming water from communities that may already be struggling with resource scarcity.
This isn’t about being anti-technology or anti-progress.
It’s about honest environmental accounting.
If we’re going to build an AI-powered future, we need to understand and address its real-world costs.
That includes being transparent about water usage, involving local communities in decision-making about data centre placement, and developing more efficient cooling technologies.
What Comes Next
This water usage issue is set to become the next major battleground for tech regulation and corporate responsibility.
Communities are starting to push back against data centre development, and regulators are beginning to ask harder questions about environmental impact assessments.
Some promising developments are emerging. Companies are experimenting with alternative cooling methods, including using recycled water, locating data centres in cooler climates, and developing more efficient chip designs.
But these solutions are still in early stages, and the pace of AI development is far outstripping the pace of environmental innovation.
The future of AI doesn’t have to be a zero-sum game between innovation and environmental responsibility. But getting there will require the same level of innovation and investment that’s currently going into making AI systems smarter and faster.
The question is whether the industry will step up to that challenge before the wells run dry.
