Nigiri with Extra Teraflops

Artificial intelligence doesn’t feed on ether or ethics - it devours electricity, cooling, and water in such quantities that even nuclear power plants start twitching nervously. Behind every dazzling generative paragraph stands not the spirit of innovation, but an orchestra of turbines, pumps, and transformers.

Data centers serving AI are no longer those boring, air-conditioned server rooms. They’re electrotechnical beasts that, in fits of training frenzy, can gulp down as much energy as a medium-sized city. Worse still, they do it capriciously… Ten megawatts today, a hundred tomorrow. Power grids, already wobbling on the tightrope of renewables, now have to cope with yet another moody diva in their lineup. And the high-voltage cables meant to handle this show? They’re simply running out.

The International Energy Agency predicts that by 2030, data centers’ energy use will double, reaching nearly a terawatt-hour per year. For context: that’s as much as all of Japan consumes today. In short, the world is gifting itself a brand-new Land of the Rising Sun - only to feed neural networks. Meaning, instead of sushi, we’re served a cold electricity bill.

But energy is only the opening act of this grotesque. GPUs in AI clusters generate more heat than a stadium of fans after a winning goal in the World Cup final. And that heat has to go somewhere. Enter water - the immortal heroine of every cooling system. The catch? It doesn’t fall from the sky in unlimited amounts, and data centers love to spring up in places where water is already scarce: Arizona, Texas, India. It’s not unusual for a server farm to drink more than an entire farming community. No wonder farmers aren’t applauding at AI Day conferences.

Of course, enthusiasts are quick to add: “AI will solve the problem itself!” That sounds about as convincing as a pyromaniac promising to put out a fire because he knows flames inside out. Sure, you can optimize liquid cooling, predict energy spikes, and shift heavy workloads to hours when wind and solar actually produce something. You can even build centers in colder regions and use waste heat to warm homes. All lovely - but why do I have the nagging feeling this will end up as a slick PowerPoint slide rather than a real-world implementation?

Because reality is rarely that kind, residents are beginning to notice that corporate server farms leave them only the bills - higher energy costs and less access to water. Governments in several countries are already limiting new data centers, fearing blackouts and social unrest. You could dress it up as “corporate social responsibility,” but the truth is more prosaic: nobody wants their fridge to shut off just because someone’s training a new language model.

The future seems to offer two scenarios. The first: AI runs headfirst into a wall. No power, no water, protests, and environmental taxes so steep that CFOs start sobbing into their quarterly reports. The second: someone finally designs this infrastructure with some brains, where optimization, heat recycling, local power sources, and sustainable regulations aren’t just forgotten village names in Ethiopia. In that world, AI could become a symbol of modern technology rather than wastefulness.

Which path will we take? If the story is written solely by corporate presentations, I’d bet on the first. But if we start treating energy and water with the same seriousness as model hyperparameters, there’s a chance for a happy ending. Until then, one question remains: do we want our digital future to be truly intelligent, or just “artificial”? Or maybe it’s not a question at all, but a bitter answer: artificial intelligence is neither intelligent nor artificial. It’s more like a teenager with an insatiable appetite - always hungry, always thirsty, and utterly convinced someone else should clean up after it. The problem is, the bill for electricity and water won’t land on the desks of philosophers or innovation marketers. It will land on ours.