Is AI the New Oil? We Need to Talk About What It’s Costing Us

We talk a lot about how smart AI is getting.
But here’s something we don’t talk about enough: it’s quietly wrecking the planet.

Not in some sci-fi dystopia. In a real, physical, “it’s-happening-now” kind of way. Every time we ask a model to summarize a doc or spit out a joke, there’s a real-world cost i.e. electricity, water, hardware, minerals.

And it adds up. Fast.

🔋 Power Hungry: How Much Juice Does AI Really Need?

Let’s start with the basics. Training GPT-3 used up more than 1,200 megawatt-hours of electricity. That could’ve powered over a hundred homes in the U.S. for a full year.

Now take GPT-4. That one? Easily 50x more, depending on the estimate. We’re talking small-city-level energy.

And guess what? That’s just the training. The real energy drain comes after that, when you and millions of others start using it. Every prompt spins up servers. Add that up over billions of queries, and you’re looking at gigawatt-hours per day.

It’s invisible to us. But behind the scenes? It’s a power-hungry monster running 24/7.

💧 Water You Doing, AI?

Here’s the part that really got to me.

Google burned through 5.6 billion gallons of water in 2022. Microsoft used over 11 million gallons in a single month, just for cooling during GPT-4’s training.

This is drinking water we’re talking about. Clean, processed water. The kind people in some parts of the world literally walk miles to get. And it’s being evaporated to cool chips.

And here’s the kicker: a lot of these data centers are in water-stressed areas. So while locals are being told to conserve, massive AI farms keep guzzling water like it’s nothing.

🖥 The Trash Heap No One Sees

We act like AI lives in some magical “cloud.” But it’s all built on hardware, GPUs, memory, server racks.

And this stuff doesn’t last. Hardware gets outdated quickly. Every new generation of AI needs faster, beefier chips. The old ones? Tossed.

In 2022, we produced 62 million tons of electronic waste. Barely one in four devices got recycled.

The rest? Dumped. Burned. Left to leach chemicals into soil and water in some distant landfill.

The cloud is real, but it’s built on a mountain of physical waste.
⛏ Digging Up the Future, One Mine at a Time

Want to build AI chips? You’ll need copper, silicon, cobalt, gallium, and rare earth minerals. Where do those come from?

Mostly: destructive mining operations. In many cases, unregulated. And often in poor regions where environmental oversight is low and human rights abuses are high.

Entire ecosystems get wrecked. Rivers turn toxic. Workers risk their lives — sometimes for dollars a day.

And we call this progress?

🧼 Greenwashing 101: The Feel-Good Illusion

Big Tech loves to talk sustainability. “We’ll be carbon neutral by 2030!” “Water positive by 2030!”

Sounds nice, right?

But here’s the reality: most of these companies won’t even tell us how much CO₂ their AI models produce. GPT-4? Gemini? Claude? Their carbon footprints are black boxes.

Instead, they buy carbon offsets, some of which don’t actually offset anything. It’s like saying, “We planted a tree, so we’re good,” while a server farm across the globe eats through a coal-powered grid.

It’s not accountability. It’s marketing.

🔁 The Efficiency Paradox

“Newer models are more efficient,” they say.
Sure. But what happens when that efficiency makes AI cheaper to run?

You guessed it, we use it more.

More apps. More AI assistants. More everything.
Just like with crypto. Efficiency didn’t shrink the footprint. It made the system grow faster.

AI’s following the same pattern, only at a bigger scale.

🚨 So… What Can We Actually Do?

We can’t sit back and hope billion-dollar companies will police themselves. That ship sailed.

But we can push for:

Clear reporting: emissions, water use, energy usage. We deserve to know.
Regulations: especially around data center energy and water limits.
Better incentives: reward small, efficient models — not just the next trillion-parameter beast.
Mindset shifts: Not every app needs a GPT-4. Maybe a small model would do just fine.

🌍 Who Pays for All This?

Every time you ask ChatGPT to plan your road trip, or generate some cute tweets, here’s what might be happening:

A liter of clean water, gone.
A few more grams of CO₂ in the atmosphere.
One more chip closer to becoming e-waste.

It’s not just “tech” anymore.
It’s planetary infrastructure, and it’s coming at a price.
Final Thought

AI doesn’t have to be a villain in the climate story.
But right now, it’s acting like one.

We can change that, but only if we stop pretending the cost doesn’t exist.

Let’s make that part of the conversation, before it’s too late.