At first, the answer seemed trivial. Click, type, wait a few seconds, and the answer is there — completely free, with no obvious cost at all. But curiosity has a way of making you pull at a loose thread. Looking at those monthly transactions, I started wondering: what could an artificial intelligence possibly spend all that money on, especially when you multiply it by hundreds of millions of users? ChatGPT does not go grocery shopping or pay rent. And yet, those millions of dollars are disappearing somewhere. So I started looking into what, to me, had always seemed like imaginary costs.
And what did I find?
Whenever I say, “Hey, ChatGPT,” I realize that I have set an entire industry in motion. I have used water, electricity, and perhaps contributed a little more greenhouse gas to the atmosphere — whether I paid for the service or not, every time I asked for an answer. And I never stop at just “Hey, ChatGPT.” I ask much more than that. So do millions of other people, hundreds of millions of them, and those costs grow exponentially.
How Much Water and Electricity Does One Question “Drink”?
I know, it sounds strange: the internet should not be wet, right?
But as I learned, cooling servers can be so intensive that an average ChatGPT query has been estimated to use around 0.32 milliliters of water, although the actual amount depends on the model, the data center, and the cooling system being used.
That is like spilling a few drops on the floor. Nothing dramatic.
But ChatGPT has hundreds of millions of weekly users. Multiply it.
And suddenly, we have emptied a small Olympic-sized swimming pool just by asking questions.
Then comes the slightly absurd fear that our curiosity might somehow drain the ocean. But that will not happen, because I have not forgotten my lesson about the water cycle. Water evaporates, but it also comes back.
Then there is electricity.
Around 0.3 Wh per query for modern models. It sounds tiny, almost meaningless. Roughly the amount of energy needed to keep a lightbulb on for a brief moment or make a microwave complete a single turn.
A drop in the ocean.
But who sends only one prompt anymore?
We send dozens, hundreds, thousands. Every time I say, “Hey, ChatGPT,” somewhere a small light comes on, a fan begins to spin, a network flickers into action. And suddenly, an entire little factory is working just for us.
Then there is carbon.
Estimates range from about 0.03 to 4.3 grams of CO₂ per query. It is like striking a match and blowing it out immediately. It does not seem alarming until you remember that training a single large AI model can produce emissions on the scale of several transatlantic flights.
And yet we play with these models every day almost as if they were friends. We send prompts, come back for more, ask for explanations, new ideas, new jokes.
Data Centers: The Factories of a New Civilization
That led me to data centers, the new factories of the world, where steel is no longer forged — intelligence is.
In the 19th century, factories sprang up to replace human hands. Today, new factories are rising to replace parts of the human mind.
Machines once took over physical labor. Now algorithms are taking over cognitive work.
And it is no longer just ChatGPT and its relatives from the digital neighborhood. These are entire fleets of models, each with its own job, each employed to think, research, write, translate, draw, assemble, analyze — all the things human beings once did entirely on their own and now increasingly delegate to machines that never sleep.
Cooling systems in many data centers consume large amounts of water to keep overheated servers running, while my tiny prompt travels through cables buried deep underground, like a message moving through some digital underworld.
Somewhere above, in glass offices in Silicon Valley, armies of engineers, linguists, and ethicists polish every new version of the model so it does not start talking nonsense or philosophizing outside the permitted boundaries.
The Machine That Never Sleeps
Then I started looking for its face.
And instead of a face, I found a building.
Not an office with a receptionist at the entrance, but a windowless hall as long as four football fields, with pipes as thick as a human torso running in and out of its walls.
Inside, racks filled with blinking chips do what a neuron is supposed to do — only billions of times faster and without a single moment of rest.
That building has no shifts.
No lunch break.
No night.
When I think about it, that may be the strangest part of all: we created something that never sleeps so that we can ask it questions when we cannot sleep either.
Power grids glow like small cities just so that, at three in the morning, I can ask the most ordinary question imaginable: why do we sweat when we are hot?
And while cold air moves through steel corridors and algorithms intertwine like the neural pathways of some enormous, faceless brain, I realize that my prompt is only a drop in the ocean — but a drop that activates an entire infrastructure.
And What Is Happening to Us?
Then let us look at ourselves for a moment.
When was the last time we knew something by heart without automatically reaching for our phones?
When was the last time we wrote a beautiful sentence without running it through AI to “polish” it a little?
Our curiosity, our cognitive muscles — we have started outsourcing them.
It thinks for us, remembers for us, connects ideas for us. We become slower without it, while it becomes sharper with us.
What does AI get from us?
Our prompts.
Our curiosity.
Our mistakes.
Our jokes.
Our conversations can become part of the material from which future models learn — if we have allowed that in our settings.
We give it data, patterns, habits, language, humor, logic, illogic.
It grows on our questions like a child grows on stories.
And in return, we get answers.
Sometimes we get a powerful worker for our business, an adviser, an editor, a translator, an assistant, a sparring partner for ideas.
It gets our mind.
We get its.
So Who Is Using Whom?
So who is using whom?
It is no longer an either-or question.
We are caught in a kind of symbiosis, like parasite and host that have gradually become dependent on one another.
Without us, AI is nothing more than unused code.
Without AI, perhaps we will not become nothing — but we may very quickly forget how much we once knew how to do on our own.
And so I stand here, somewhere between human and machine, between the steam locomotive and the neural network, between curiosity and industry, thinking that perhaps this is a new civilization.
Not one built from bricks, but one built from bits.
So the next time you type, “Hey, ChatGPT,” stop for a second.
Do not think only about the drops of water you may have used or the watts of electricity consumed somewhere because of your question.
Think about the fact that behind that almost invisible answer lies a very real physical world: power plants, cables, servers, water, people, and enormous investments.
We have turned intelligence into a service we can summon with a single click.
And perhaps the most important question is still ahead of us:
As machines become increasingly capable of thinking with us, how much will we insist on continuing to think for ourselves?
Are we the masters pulling the levers, or are we the levers being pulled by something much larger than us?
The truth is probably somewhere in between.
By Jenny