When a company announces a new data center, they don't brag about square footage. They announce it in megawatts — "a 500-megawatt facility." That's a power rating, the same unit you'd use for a power plant, not a building. Once you notice it, it tells you almost everything about why the AI build-out looks the way it does.
Power is the real constraint, not space
A data center is a building full of computer chips, and chips do two things: consume electricity and turn it into heat. The limiting factor isn't how many servers you can physically fit — it's how much power you can feed them and how much heat you can haul away. You can always build a bigger box. You can't always get the electric grid to deliver hundreds of megawatts to that box. So the industry measures capacity the way it's actually limited: in power.
Why this is suddenly everyone's problem
AI made this worse. The chips that train and run modern models draw enormous power and run hot, packing more electricity demand into the same floor space than older data centers ever did. The result is that AI companies are now competing for electricity itself — striking deals for power plants, signing up for nuclear capacity, and choosing sites based on what the local grid can deliver. The bottleneck on AI growth is quietly becoming an energy problem, not a chip problem.
What it means for everyone else
You don't run a data center, but this shapes your world in two ways. First, the cost of the AI tools you use is tied to energy prices, which is part of why pricing can shift. Second, this demand lands on the same grid that powers your business and your town — which is why data centers, electricity rates, and AI are increasingly showing up in the same local news story. The build-out is a physical, energy-hungry phenomenon, not a purely digital one.
The honest caveat
Chips also keep getting more efficient, doing more work per watt every year, so it's not a straight line up forever. But for now, demand is outrunning those efficiency gains, and power remains the headline number.
The takeaway: Next time you see a data center measured in megawatts, read it as a tell — AI's real frontier right now is electricity. The companies that lock up power, not just chips, are the ones positioned to win.