Jesus christ dude. A few years ago we were marching hopelessly towards extinction, and then a handful of psychopaths decided we should be sprinting. One day a more intelligent lifeform will read our story as a classic tragedy.

This is large, yes, but to put it in perspective:

Currently about ⅔ of datacenter energy is spent on non-AI, like websites, databases, streaming, SAAS, email etc etc etc; ⅓ is AI. So yes, that’s a lot.

But it is also true that global datacenter power generation is currently around an estimated 130 GW worldwide, so the >7.5 GW here is around a 5% increase in just one project. (Well, if the ⅓ estimate is true, 45 GW of the 130 GW is AI, so that will go 45 GW → 52.5 GW or so. So like a… 12%-24% increase, something like that)

And it is also true that it will be primarily/mostly used for AI-related things.

And even worse, seems like this would be prime real estate for solar. Welllllll… you can’t generate 7.5ish GW of constant power, you need extra and battery storage. So unfortunately you need 225,000 acres of solar panels. Larger than Chicago or Austin, covered solid with panels. You’d need upwards of 150 GWh of battery storage. THe US currently has about 30 GWh of battery storage all across the US. So it would drive up the cost from natural gas’s $10B to a project of $120B or so.

So that sucks.

The only mitigating factor I can really find is that every new site increases overhead, so at least if it’s all in one place, it will be a bit more efficient. Also they should not need that full power generation all the time, so while it’s still massive, they are talking the worst case scenario.

I’m not trying to defend this thing. I just hate it when the argument gets reduced beyond reasonable truth, and I looked this stuff up because I was curious. I’m not supportive of this.