Musk's xAI Supercomputer Isn't the Beast He Claims (Yet)

Okay so I saw this deal and had to share... except it's not really a deal. More like Elon doing his usual thing where the hype runs ahead of reality. Remember when he said xAI's Colossus 2 supercomputer was already hitting 1 gigawatt capacity? Yeah, about that.

Turns out some smart folks at Epoch AI got their hands on satellite imagery of the facility, and the math doesn't add up. At all. The cooling infrastructure they spotted can handle maybe 350 megawatts tops. That's barely a third of what you'd need for a true gigawatt-scale operation. And trust me, when you're running that many GPUs, cooling isn't optional – it's literally what determines how much compute power you can pack in there.

Here's what bugs me about this whole thing. We're talking about massive AI training clusters that pull more power than small cities. You can't just say "we're at a gigawatt!" without the infrastructure to back it up. The satellite photos don't lie – you can count the cooling towers yourself if you're into that sort of thing.

Look, I get it. Building these mega-clusters is hard and expensive. But why claim you're already there when you're clearly not? The 350MW setup they actually have is still impressive. That's enough juice to train some seriously powerful models. Just... not the world-beating setup Musk's been tweeting about.

Bottom line? If you're tracking the AI arms race in 2026, take those gigawatt claims with a hefty grain of salt. The hardware's probably coming eventually, but right now Colossus 2 is more like Colossus 0.35.