You typed a question into ChatGPT this morning. Maybe it was about a recipe, or a work email, or just something you were curious about.
Somewhere, hundreds of miles away, a massive warehouse full of humming servers got a tiny bit hotter. And somewhere near that warehouse, a small amount of real, drinkable water disappeared to cool it down.
You never saw it happen. You never thought about it. That is exactly the problem.
We talk about AI like it lives entirely in the cloud — weightless, invisible, costless. The truth is far more physical, and far more thirsty, than most people realize.
The Number That Should Surprise You
Let's start with a single response.
According to scientists at the University of California, Riverside, each 100-word AI prompt is estimated to use roughly one bottle of water — about 519 milliliters. One email draft. One bottle of water, evaporated to keep the servers cool. Quora
OpenAI has pushed back on that figure. Sam Altman stated in a blog post that the average query uses just 0.3 milliliters of water — about one fifteenth of a teaspoon. Webzeto
Here is the catch. Altman's figure only accounts for direct operational water at the data center, not the water used in generating the electricity itself or manufacturing the chips. The gap between 0.3 milliliters and 519 milliliters is not a rounding error — it is a measurement choice. Webzeto
In other words, both numbers can be technically true. It depends entirely on how much of the full picture you are willing to count.
When You Multiply It By Billions
A single prompt sounds harmless. The math stops being harmless very quickly.
ChatGPT alone is estimated to process around 2.5 billion prompts per day. Globally, ChatGPT uses around 39.16 million gallons of water daily — the equivalent of every single person in Taiwan flushing their toilet at the exact same moment. World Population ReviewMillennial Money Man
When you account for the full supply chain, daily water use of ChatGPT reaches approximately 148 million litres — enough to fill 59 Olympic swimming pools, every single day. Webzeto
And this is just one product, from one company. Multiply that across every major AI platform now competing for users worldwide, and the scale becomes difficult to even visualize.
The Electricity Bill Behind the Chat Window
Water is only half the story. The other half is power.
For standard inference, OpenAI disclosed that a typical ChatGPT query consumes approximately 0.34 watt-hours, while academic benchmarks place GPT-4o closer to 0.42 watt-hours for a short exchange. Quora
That sounds tiny until you remember the billions of daily queries. Using available figures, ChatGPT may presently use around 39.98 million kilowatt-hours per day — enough to charge eight million phones. Millennial Money Man
And the energy cost is not flat across every type of request. Per-query energy varies enormously by task. A typical chat query uses around 200 times the energy of basic text classification. An AI-generated image uses around 1,450 times that baseline. A single short AI video can match the energy cost of 200,000 simple text classifications. World Population Review
That casual AI image you generated for fun, or that short AI video clip — these are not free actions. They carry a real, measurable energy cost, multiplied across millions of users doing the exact same thing.
The Communities Living Next to the Servers
This is the part that rarely makes it into the conversation — because most people never see where their AI queries are physically processed.
A single Meta data center in Newton County, Georgia, uses 500,000 gallons of water per day — roughly 10 percent of the entire county's water consumption. Webzeto
In the Netherlands, local media discovered that a Microsoft data center used over 22 million gallons of water in a single year — more than four times what the company and the municipality had told residents to expect. Medium
In the United States, around one in five data centers draw water from areas that are already over-stretched. Critics were quick to point to AI's water demand when fire hydrants ran dry during the Los Angeles wildfires in January. Millennial Money Man
These are not abstract global statistics. These are specific towns, specific counties, specific families whose water bills and water access are being quietly reshaped by infrastructure most of them never voted on and rarely even knew was being built nearby.
The Numbers Getting Bigger, Not Smaller
You might assume that as AI technology improves, the environmental cost should shrink. The reality is the opposite.
By 2030, global data centers powering artificial intelligence are projected to consume 945 terawatt-hours of electricity — nearly triple the combined annual electricity use of Pakistan, Bangladesh, and Nigeria, three countries collectively home to more than 650 million people. World Population Review
There is a specific reason for this counterintuitive trend. A lot of people assume the environmental footprint of AI naturally reduces as the technology becomes more efficient. But more efficient and more affordable AI actually drives more consumption of AI overall, making the total footprint larger than what efficiency gains save. World Population Review
This pattern has a name in economics — it is sometimes called the rebound effect, and it shows up again and again throughout history whenever a resource becomes cheaper to use.
The same dynamic has repeatedly reshaped global power struggles over scarce resources, from oil to fresh water, where increased access to a resource consistently drives demand upward rather than reducing overall strain on supply.
Ireland — A Warning Sign for the Rest of the World
If you want to see what happens when AI infrastructure outpaces planning, look at Ireland.
In Ireland, data centres accounted for 21 percent of total metered electricity in 2023 — exceeding the combined electricity use of all urban households in the country. The national grid operator has paused new data center approvals around Dublin until 2028. World Population Review
That is not a future projection. That has already happened. A country's electricity infrastructure was strained badly enough by data centers that the government had to step in and halt new construction entirely.
In the United States, a January 2026 report predicts that data centers' total combined energy demand will nearly double between 2025 and 2028, jumping from 80 to 150 gigawatts — equivalent to adding a country with the energy needs of Spain in just three years. YouTube
And ordinary people are already feeling it financially. A Bloomberg analysis found that areas with high concentrations of data centers saw electricity prices jump 267 percent over the past five years, and nearly three-quarters of Virginia voters now blame these facilities for rising electricity costs. YouTube
What This Actually Means for You
None of this means you should stop using AI tools. They are genuinely useful, and they are not going anywhere.
But understanding the real cost changes how you might choose to use them. Model choice and product defaults are footprint decisions — a basic text query carries a fraction of the cost of generating an image, and a fraction of that compared to generating video. World Population Review
Using AI thoughtfully — for tasks that genuinely need it, rather than generating ten variations of an image you only needed once — is not just good practice. It is a small, practical way to reduce a footprint that is currently expanding faster than most public infrastructure was ever designed to handle.
The companies building these systems are racing forward at extraordinary speed, with extraordinary capital behind them. The communities living next to the data centers, and the power grids straining to keep up, are the ones quietly absorbing the cost of that race.
The next time you send a message to ChatGPT, it is worth remembering — somewhere, a machine just got a little hotter, and somewhere nearby, a little bit of real water just disappeared to cool it back down.
0 Comments