Opinions

The Planet is Burning. But at Least You Know What You’d Have Looked Like in the ’80s

BYKhaoula Ghanem

September 21, 2026

If you’re wondering why it is so hot, somebody’s mother is on Facebook generating AI photos of herself in the 1980s when there is an actual photo album containing actual photographs of her in the 1980s sitting somewhere in her house. Please. I am begging everybody to locate the album.

We have just crawled out of a summer that was jointly the hottest ever recorded globally. August 2026 was the hottest August on record. Western Europe had its hottest summer ever recorded. June alone brought brutal heatwaves, wildfires, and shattered temperature records across parts of Europe. The planet is quite literally cooking, and we have decided this is an excellent moment to ask a warehouse full of GPUs, “What if I had a perm?”

And apparently one wasn’t enough.

Generate again.

Make the bangs bigger.

Make it more ’80s.

Why does it not look like me?

Generate again.

Girl, YOU WERE THERE.

First, we Ghibli-fied ourselves. Then came the action figures, starter packs, fake movie posters, fake magazine covers, etcetera, etcetera, etcetera. We ask ChatGPT what to eat, what to wear, where to go, what to text back, whether our situationship likes us, and what the dream we had at 4 a.m. means. We are treating machines like graphic designers, personal assistants, therapists, best friends, and Magic 8 Balls, and the machines need data centers full of computers that require electricity and water to keep running. At this rate, we are going to be cooked alive because somebody needed to know what their cat would look like as a human.

That sentence is deliberately ridiculous. The environmental consequences are considerably less so. Generative AI has a physical footprint that is remarkably easy to forget because its interface has been designed to make the machinery disappear. You type something into a clean little box and something comes back. There is no smoke coming out of your laptop, and no tap starts running beside your keyboard. And even though no electricity meter appears in the corner tallying up your prompt, the infrastructure exists anyways.

Behind the AI ’80s yearbook photo sits a global network of data centers. Inside those buildings are servers running around the clock, producing enormous amounts of heat. Heat is bad for computers, which means data centers need cooling. Cooling can require water. Electricity generation can require water. Manufacturing the semiconductor chips inside those computers requires extremely pure water. Those chips require minerals, chemicals, fabrication plants, and international supply chains before they ever arrive inside a data center. As for the cloud? Tragically, it’s not actually a cloud. It is buildings. And those buildings have plumbing. Before you have generated anything, your little AI 80s photo already has an industrial supply chain.

To be clear, your AI perm did not cause climate change. Fossil fuels remain the primary driver of global heating, and pretending otherwise would let the industries responsible for the overwhelming share of historical emissions off spectacularly easily. The problem is that we are building another extraordinarily resource-hungry industry on top of an environmental crisis that was already well underway while finding increasingly stupid things for it to do.

There are applications of artificial intelligence that could plausibly justify their environmental cost, like scientific research, drug discovery, climate modeling, and energy optimization. Give the servers something useful to sweat over. Researchers at the University of California, Riverside estimated in 2023 that roughly 20 to 50 GPT-3 queries could consume around 500 milliliters of freshwater, although the amount varies substantially according to where and when the model is operating. Half a liter sounds like nothing, but that’s exactly how this gets you.

Nobody is sitting down, generating one AI image, admiring it thoughtfully, and logging off. You generate four. One makes you look 63. Regenerate. The next one gives you seven fingers. Regenerate. Your hair looks weird. Regenerate. Now do the ’90s. Now make your boyfriend one. Now put both of you together. Now make your hypothetical children.

Congratulations on the family portrait. The children do not exist, but their environmental footprint does.

The environmental conversation around AI also tends to begin far too late. We obsess over what happens when somebody presses generate. By that point, an enormous amount has already happened. Before an AI model can tell you what you would have looked like at Studio 54, somebody has to manufacture the chips capable of telling you.

Semiconductor fabrication requires ultrapure water, or UPW, because microscopic contaminants can ruin the extraordinarily delicate manufacturing process. Producing that water itself requires water, and lots of it. Then there are the raw materials, factories, transportation, data-center construction, and electricity required before a single prompt has been entered.

Then comes training, deployment, and billons upon billions of prompts. Then eventually comes obsolete hardware and electronic waste because, naturally, the industry producing tomorrow at breakneck speed also produces yesterday at breakneck speed.

We are also talking about a technology whose demand is growing at extraordinary speed. Data centers are being constructed around the world, AI is being shoved into products that worked perfectly well without it, and generative features increasingly appear whether anybody asked for them or not. You search something. AI. You open Pinterest. AI. You write an email. AI.

The internet once asked us to consume content. Generative AI asks us to manufacture infinite amounts of it ourselves.

Researchers from Hugging Face and Carnegie Mellon University found that image generation was considerably more energy-intensive than many text-based AI tasks. Across the image-generation models they tested, producing 1,000 images consumed about 2.9 kWh on average, while the least efficient model they tested used substantially more. Meanwhile at MIT, researchers found that generating one image requires as much energy as charging your phone.

Perhaps every fleeting aesthetic trend does not need to become a computational event involving data centers halfway across the world. That does not mean nobody should use AI for fun. Environmental conversations become insufferable very quickly when responsibility gets dumped entirely onto ordinary people while corporations continue expanding infrastructure behind them.

The larger problem belongs to the companies building AI into everything while disclosing remarkably little about the resources individual products consume. If AI is going to become basic infrastructure, then its environmental footprint needs to become basic information, because the problem begins when an extraordinarily resource-intensive technological infrastructure is marketed to us as limitless, invisible, and essentially free.

Nothing industrial is free. Somebody pays for the electricity, somewhere supplies the water, somebody mines the minerals, some community lives beside the data center.

And eventually, all of us live with the climate.

Main image: Instagram/@sabrybeauty

The Planet is Burning. But at Least You Know What You’d Have Looked Like in the ’80s

Related Stories

More From Opinions

Newsletter

The only newsletter you won't unsubscribe from

Follow Us