Every answer artificial intelligence gives you passes through a real building, full of machines that seriously eat power. Behind the talk of "the cloud" stands concrete, cables and a monthly bill bigger than a whole city's. Here's who builds these buildings, who powers them, and who ends up paying.
A warehouse, not a cloud
When you ask a chatbot something, the answer doesn't come "from the cloud". It comes from a specific warehouse, somewhere in Virginia, New Mexico or Ireland, full of racks of servers humming day and night. These machines are starving for power and cooling water. Without them, the smartest model in the world is a pile of silicon that computes nothing.
So when you read a story about a new AI (artificial intelligence) model, behind the headline there's always a construction site. The company has to find land, build a warehouse, connect it to the power grid, and secure water. Dull, heavy industrial work, which turns out to be more expensive and harder than the software itself.
Why suddenly everyone's talking about nuclear plants
Here comes the part that sounds unbelievable but is true. Microsoft signed a deal with the energy company Constellation to restart a shut-down reactor at the Three Mile Island nuclear plant in Pennsylvania. The site is known for the worst nuclear accident in US history, which happened in 1979 in the neighboring reactor. Now Microsoft is bringing the stopped reactor back to life and buying all its power, just for itself, for twenty years ahead.
Stop here for a second. They're not buying a rate, the kind you pay on your home meter. They're buying the entire output of an entire power plant, before it's even switched back on. It's as if a factory skipped the city grid and the ordinary meter altogether - and put up its own substation right next to the plant, because the shared grid can't guarantee enough voltage for its round-the-clock furnaces.
Microsoft isn't the only one. Big AI companies are running around the world buying direct access to gas plants, wind farms, even building their own small reactors next to their own warehouses. The reason is simple: the ordinary power grid wasn't built to suddenly absorb a crowd of new factories, each drawing as much as a small city.
Who builds it and with whose money
These warehouses aren't built by the model's inventor with pocket money. They're built with investor money, bonds, and sometimes direct government loans. Investors bet on it because they expect the AI company to stay a customer with a long-term power supply contract, the same way a factory signs a fifteen-year raw material supply deal to convince a bank to finance a new plant.
The problem is that when so much money heads toward power and concrete, the line between "AI company" and "energy company" blurs. The firm selling you a chatbot increasingly looks like a firm running power plants and negotiating with state governors. Software has become the smaller part of the business.
Who pays in the end
Power doesn't appear out of nowhere just because a paying customer is ready to pay. When a company reserves a huge chunk of a region's electrical capacity, it joins the same queue for the same grid as hospitals, factories, and your apartment block. In the state of Virginia, which has the densest concentration of these warehouses in the world, they already draw as much power as every fourth light bulb would draw from the local grid. Power companies are building new lines to keep up, and those costs get spread across every subscriber's bill, not just the warehouses'.
The bill doesn't go to one place, it gets split. Once by the investors, the ones financing the build. Once by you, because the cost of the long-term power contract gets folded into the subscription you pay for the chatbot. And once by the warehouse's neighbors, whose own bill goes up because the local grid now also serves a factory that never sleeps.
Who holds the power now
Here's the twist worth paying attention to. While everyone watches who's making the smartest model, the real power is shifting toward whoever can guarantee electricity. A state with spare grid capacity is suddenly a sought-after partner. An energy company with an old, forgotten reactor is suddenly worth a fortune. A regulator who signs a permit faster carries more weight than an engineer who writes code.
What will stand the test of time
At home, my AI systems run on ordinary servers, with a perfectly ordinary power bill. That's why the difference hits me right in the eye. These companies no longer talk about a smarter model. They talk about power plants, reactors and deals with governors. This is an infrastructure and political story, not a technological one, and the scale is a different galaxy.
The most ironic part to me is exactly the choice of Three Mile Island. The place all of America remembers as a symbol of nuclear failure is now sold as a symbol of progress, because someone has to power the chatbots. Nobody asks whether it's wise. They only ask whether the reactor will be running on time.
So next time you read a story about an AI breakthrough, don't just look at what the model can do. Ask where it draws its power from. That decides who ends up holding the power - not the chatbot.