The Hidden Cost of Curiosity
What does asking AI a question really cost our planet—and should we feel guilty about using it?
I’ve been wondering for quite a while about the resources required to power artificial intelligence. Every time I ask ChatGPT a question, generate an image, or get help with a project, somewhere there’s a computer doing the work.
And computers need electricity.
But how much? And what happens when millions of people are using AI every day?
Recently, I watched a television show featuring a data center being built near a small community. The facility was drawing so much electricity that the aging electrical grid struggled to keep up. Problems cascaded, eventually contributing to electrical fires.
Now, television dramas aren’t exactly known for their restraint when it comes to disasters. But it made me wonder:
Wait a minute. Could that actually happen?
And if I’m using AI to research an interesting subject, solve a problem, or generate a cartoon picture of myself, am I contributing to the demand for these enormous facilities?
The short answer is yes.
The longer answer sent me down a rather interesting rabbit hole.
What Happens Behind That Little Blinking Cursor?
When I type a question into ChatGPT, the answer doesn’t magically appear out of thin air. (Although sometimes it certainly feels that way.)
Powerful processors in data centers perform calculations to generate the response. Those processors consume electricity and produce heat, which means they need cooling systems that may use additional electricity, water, or both.
Then there are the resources required to manufacture the chips, construct the buildings, and maintain the infrastructure.
Even a simple question has a physical footprint.
It’s strange to think about, isn’t it? We tend to imagine digital activities as somehow separate from the physical world. We’re not driving anywhere or plugging in an enormous machine.
We’re just typing.
But behind that blinking cursor is an entire physical infrastructure.
An ordinary text question generally requires a relatively small amount of energy, although the amount varies with the model, the task, and the equipment.
The bigger issue emerges when we multiply those small amounts by millions—or billions—of interactions.
How Much Electricity Are We Talking About?
According to the International Energy Agency (IEA), data centers accounted for approximately 1.5% of global electricity consumption in 2024. The agency projects that consumption could more than double by 2030, with AI driving much of the growth.
Of course, data centers don’t exist solely for AI. They also support websites, cloud storage, streaming services, business applications, and much of our everyday digital world.
But AI has introduced some particularly demanding computing needs.
Large AI-focused facilities can require more than 100 megawatts of power—enough to supply many thousands of homes, depending on household electricity consumption.
Now imagine asking a community’s decades-old electrical grid to accommodate that kind of demand.
Suddenly, my television show doesn’t sound quite so far-fetched.
But Doesn’t Everything We Do Online Use Resources?
Here’s another question that occurred to me.
Suppose I spend thirty minutes learning about astronomy by asking ChatGPT questions.
Would that consume more electricity than watching a thirty-minute educational video about astronomy?
Surprisingly, there isn’t a simple answer.
Streaming a video involves transmitting existing content and displaying it on my device. AI must generate responses using computationally intensive processors.
Depending on the model, the number and complexity of my questions, and the devices involved, either activity could potentially use more electricity.
And what about everything else we do online?
Streaming movies, scrolling social media, playing online games, storing thousands of photos and screenshots in the cloud, and participating in video calls all require physical infrastructure.
AI isn’t unique in having an environmental footprint.
That doesn’t mean all digital activities consume equal resources. Far from it! But we should be careful about assuming one is environmentally irresponsible while another is harmless without reliable measurements.
Perhaps another worthwhile question is:
What am I getting from the resources being used?
Thirty minutes spent learning something new has value, whether I’m asking AI questions or watching an educational video.
Could a Data Center Really Cause Fires?
Back to that television drama.
Could a data center overload an electrical grid so badly that equipment fails and starts fires?
The underlying concern is real, although the television version deserves a generous helping of dramatic license.
Electrical infrastructure can fail. Transformers can overheat, power lines can develop faults, and malfunctioning equipment can create ignition hazards.
In dry, windy conditions, an electrical ignition can become a serious wildfire.
However, electrical grids have protective systems designed to prevent dangerous overloads. A data center doesn’t ordinarily cause fires simply because it uses a lot of electricity.
Before connecting an enormous facility, utilities generally evaluate available capacity and determine whether upgrades are needed.
Those upgrades might include new substations, transformers, transmission lines, or additional power generation.
Unfortunately, improvements can be expensive and take years to complete, while demand for data centers continues growing.
So no, asking AI to write a poem isn’t likely to make your neighborhood burst into flames.
But the pressure large facilities place on electrical infrastructure is a legitimate concern.
Apparently, the television writers weren’t entirely making things up!
Electricity Isn’t the Only Resource
Then there’s water.
Computer processors generate heat, and that heat must go somewhere.
Some data centers use cooling systems that consume water through evaporation. Others rely more heavily on air cooling, closed-loop systems, or different technologies.
Water consumption varies considerably, and location matters.
A facility using substantial water resources in a water-rich region presents a different challenge from one operating in an area experiencing drought.
Electricity generation may also require water, depending on the energy source.
Add the resources needed to manufacture and eventually replace specialized hardware, and the environmental footprint becomes more complicated than simply checking the electric meter.
And What About Those AI Image Trends?
Now we come to something that has really caught my attention.
You’ve probably seen them.
Someone discovers a prompt that transforms an ordinary photograph into a glamorous holiday portrait, a fantasy character, or an elaborate cinematic scene.
They share the prompt.
Everyone tries it.
Then everyone shares their results.
And suddenly social media is overflowing with AI-generated images.
The trend spreads like wildfire.
(Sorry. Given how this curiosity started, I couldn’t resist.)
These images can be hilarious, beautiful, imaginative, or just plain fun.
But creating an image generally requires substantially more computation than answering a simple text question.
Research comparing AI tasks has found that image generation can consume considerably more energy than some text-processing tasks. However, the difference varies by model, hardware, image resolution, and task.
There’s no reliable universal rule that every image uses exactly ten, fifty, or one hundred times as much energy as every text response.
Still, when millions of people participate in a trend—sometimes generating dozens of variations until they find the perfect one—the combined demand adds up.
Which brings me to a slightly uncomfortable question.
Should We Feel Guilty About Generating Images Just for Fun?
I don’t think guilt is particularly useful here.
But I also don’t think pretending these activities have no environmental cost is the answer.
Consider all the other things we enjoy that consume resources: television, movies, video games, travel, crafting, and countless everyday hobbies.
We don’t automatically conclude that every recreational activity is irresponsible simply because it isn’t essential.
Enjoyment, creativity, entertainment, and curiosity have value, too.
That doesn’t mean resources are unlimited or that our choices don’t matter.
Maybe we don’t need fifty nearly identical AI portraits when five would do. Maybe we can be a little more intentional about how we use these tools.
But should someone enjoying a whimsical AI image feel personally responsible for a billion-dollar data center?
I don’t think so.
And that leads to a much bigger question.
Who Is Ultimately Responsible?
Technology companies make decisions about where to build data centers, what equipment to install, which cooling systems to use, and how to obtain electricity.
Utilities evaluate electrical capacity and infrastructure needs. Governments and regulators establish rules governing construction, environmental impacts, and resource use.
Local communities may also have a say.
Data centers can bring construction jobs, tax revenue, and economic investment. But communities have legitimate questions:
- Who pays for electrical-grid upgrades?
- Could existing customers face higher electricity rates?
- How much water will the facility consume?
- Will the project create lasting local jobs?
- Do the benefits justify the resources required?
A data center might be a welcome investment in one community and a poor fit for another.
That’s why thoughtful planning, transparency, and accountability matter.
Individual users contribute to overall demand, certainly. But the people making enormous infrastructure decisions have considerably more influence over how that demand is met.
Can AI Become More Efficient?
To me, this is one of the most interesting—and encouraging—questions this rabbit hole uncovered.
Computing technology doesn’t stand still.
Researchers are developing more efficient chips, smaller and more specialized AI models, improved software, and better data-center cooling systems.
Some facilities are exploring ways to reduce water consumption, reuse waste heat, or obtain electricity from lower-carbon sources.
If future AI systems can accomplish the same tasks using less energy, the environmental cost of individual interactions could decrease substantially.
There’s just one complication.
When technology becomes cheaper and easier to use, we tend to use more of it.
So improved efficiency doesn’t automatically guarantee lower overall electricity consumption.
Still, there’s plenty of room for innovation, and that’s something worth watching.
The Curious Problem of Asking AI About AI
There’s something wonderfully ironic about how I investigated this subject.
I wondered whether using AI was contributing to environmental problems.
So naturally, I asked AI.
Which required electricity.
Then I asked more questions.
More electricity.
And eventually, I used AI to help research and develop an article about the electricity required to use AI.
At some point, the irony becomes impossible to ignore.
But there’s a useful lesson here.
AI can be a wonderful tool for exploring unfamiliar subjects and discovering questions we might never have thought to ask.
However, AI companies aren’t necessarily impartial sources when discussing their own environmental impact.
Independent research, reliable statistics, and transparent reporting matter.
And ironically, AI can be a useful tool for finding those independent sources—provided we actually check them!
Curiosity should lead us toward better information, not simply whichever answer makes us feel most comfortable.
So, What’s the Hidden Cost of Curiosity?
Every AI question consumes resources. Every generated image has an environmental footprint. And collectively, our growing appetite for AI contributes to demand for enormous data centers requiring electricity, water, land, and infrastructure.
Those costs are real.
But so are the benefits.
AI can help people learn, create, solve problems, improve accessibility, conduct research, and accomplish things that might otherwise be difficult or time-consuming.
The challenge isn’t simply deciding whether AI is good or bad.
It’s learning to use a powerful technology thoughtfully while expecting the companies building it to do the same.
For me, that means being aware of the resources involved, making reasonable choices, and paying attention to the larger questions about infrastructure and accountability.
It doesn’t mean feeling guilty every time I ask a question.
And it probably won’t stop me from occasionally generating a ridiculous picture just because it makes me smile.
After all, curiosity is part of what makes life interesting.
The important thing is making sure the hidden costs don’t stay hidden.
Sources and Further Reading
- International Energy Agency (IEA) — Energy and AI (2025). Global data-center electricity consumption, projections, and grid challenges.
- Lawrence Berkeley National Laboratory — United States Data Center Energy Usage Report. U.S. electricity demand and data-center growth.
- U.S. Congressional Research Service — Research on data centers, electrical infrastructure, and utility planning.
- Nature — Light Bulbs Have Energy Ratings—So Why Can’t AI Chatbots? (2024). Research into the energy demands of AI tasks.
- Research on AI image-generation energy consumption — Studies comparing the computational and energy demands of image-generation models.