Everyone wants to talk about the chatbot.
That’s where the excitement lives. Type a prompt. Get a lesson plan in seconds. It feels like magic. Weightless, almost. Like a Google search that writes back.
But that feeling? It’s an illusion.
And before districts start redesigning their curricula around generative AI, they need to answer a duller, heavier question:
Can we actually afford this?
The Tower in the Room
Remember the computer labs of the late 90s?
Kids typed on a keyboard, stared at a monitor, and assumed the brain lived in the beige box on the desk. It didn’t. The real work happened inside a massive, humming tower. Or in the server room down the hall.
Generative AI works the same way, only the tower is now a fortress.
Located hundreds of miles away. Sometimes just five miles down the road. A sprawling data center, not a desktop PC.
Think of your prompt as a TV remote.
The hardware in the cloud? That’s the wiring inside the set. The response? The image on the screen.
You don’t see the wiring.
But you should care about it. Every prompt. Every automated feedback comment. Every lesson plan. They cost electricity. They consume water. They demand specialized processors. They require scarce computing capacity that doesn’t grow on trees.
Generative AI is not just code. It is a physical thing that demands power, land, and water.
The Energy Tab
Most educators focus on using AI. Literacy. Governance. Stanford reviews say adoption is outpacing the evidence for student outcomes. UNESCO urges human oversight. All vital conversations, yes. But they miss the foundation.
A different set of researchers looks at the footprint.
PhD Xiaofan Liang maps how AI reshapes land use. PhD Shaolei Ren tracks the thirst.
Here is the number that matters.
In 2023, U.S. data center electricity consumption hit roughly 176 terawatt-hour s. That’s 4.4% of all electricity in the country.
Put that in human terms.
That’s enough power to keep 17 million homes running for a whole year.
Look at the grid.
The appetite for AI power is growing, and it is voracious.
Why It Never Gets Cheaper
Schools used to buy software differently.
You licensed a platform. A LMS. An assessment tool. The price was set. Predictable. Foreseeable. As you added users, the marginal cost often dropped. Economies of scale.
AI? Not so much.
Industry insiders call it “inference costs.”
Every time you ask the model something, you burn resources. It generates new data. It costs money. Every. Single. Time.
Scale does not lower the per-unit cost. It raises the total bill.
So here is the dilemma.
Most districts are dabbling. Pilot programs. Limited licenses. Features bundled into other apps. We don’t know the cost of universal access.
What does it cost if every student in the district uses AI daily?
We have no answer. And if the price spikes, the district is stuck.
Privacy is a Price Tag
Then there’s data.
Parents are worried. Rightly. They don’t want student details flowing into commercial models.
So the solution?
Host it yourselves. Use private deployments. Keep the keys.
Sounds secure, right?
It’s expensive.
If you want control, you need servers. You need cybersecurity. You need networking hardware. You need technicians to fix it.
Data privacy is not just a policy setting.
It is an infrastructure build. The more privacy you want, the deeper your wallet gets.
The Moving Target
Meanwhile, the market is shifting under their feet.
OpenAI. Anthropic. The pricing models change before you finish reading the Terms of Service. Infrastructure investments remain astronomical. The economics are unclear.
And when do they hit schools with this?
When ESSER federal funds ran out. When states are arguing over every dime of ed-tech spending. When teachers are burnt out. When mental health crises are spiking.
Do districts understand the commitment they’re signing up for?
Probably not.
It’s not a purchase. It’s a dependency.
The Community Split
One last thing.
Data centers are expanding wildly across the U.S.
Local communities are fighting back. Or at least, they’re negotiating.
Public meetings are full of angry neighbors talking about water use. About strain on the local grid. About heat exhaust. About land use.
Educators might think this is far from the classroom.
It’s not.
Every AI lesson plan depends on those buildings being built, permitted, and powered.
No Easy Finish
We are debating how to use AI while we haven’t figured out if we can support it.
The infrastructure is taking shape. The economics are fuzzy. The governance is messy.
Maybe that’s the point.
Schools should probably figure out what they can keep before they decide how much to lose.
