Training One AI Model Can Use as Much Power as 100 Homes in a Year

The energy bill behind the magic is enormous, and mostly invisible. We break down why the next AI race is really a race for electricity and for the water used to cool the machines — and what that means for where these systems get built.

History rhymes here in a way that is hard to ignore. We have seen a version of this before, in a different industry and a different decade, and the way it played out then tells you a great deal about how it is likely to play out now. The mechanism, it turns out, is much older than the technology.

So what changes next? The second-order effects are where this gets genuinely interesting — the knock-on consequences that nobody is pricing in yet, precisely because they are still one step removed from the obvious story. Those quieter effects, not the headline itself, are the ones actually worth watching from here.

None of this is fully settled, and it would be dishonest to pretend otherwise. The available evidence points firmly in a direction without quite proving the destination, and anyone selling you certainty is, in the end, selling you something. What we can do is map the terrain clearly enough that you can judge it for yourself.

The headline is the easy part. What follows is the harder question: who actually benefits, who quietly pays, and which incentives were already in motion long before any of this reached the news. We start there, because that is usually where the real story hides — not in the announcement, but in the conditions that made it inevitable.