I've seen Musk note in an interview that at year-end the bottleneck will not be CPU/RAM etc, but electricity. And new powerplants are backlogged for years.

That's why he wants to go into space (10x solar potential because you don't have a day/night cycle, no clouds, no dust/rain, no temperature loss, no orientation issues, and no atmosphere reducing solar).

To me it seems ridiculous, for one because sending 150kg to space costs about $500k, and this is about the weight of a solar installation that costs $800 to install and generates about $1000 worth of electricity across 20 years at utility wholesale prices.

But suppose it was cheaper and viable, and earth-electricity was indeed capped, you could argue (if you believe the hype) that developing AI is an existential arms-race objective for US/China.

But from what I've understood that's just not the case at all. Something like 170+ coal plants are scheduled to be decommissioned, and the average coal and gas plant runs at 40-50% of capacity, because wind/solar is eating their lunch (cheaper marginal $ per kWh). i.e. there is so little demand that these plants keep using less capacity and shutting down superfluous plants.

You'd think if experts believed electricity was going to be a bottleneck, that venture capital / AI companies, or even traditional capital, would be buying up plants or signing guaranteed-usage contracts. But it doesn't seem to be the case.