Elon Musk just spent three hours saying one thing over and over again: AI is no longer limited by intelligence. It is limited by electricity.
Everything else in the conversation is downstream of that fact. The chips are coming. The models are improving. The capital is available. But the power? That’s the choke point.
And Elon is responding in the most Elon way possible: by attempting something absurdly hard. Data centers in space. Elon’s claim that space will be the cheapest place to run AI within ~36 months is deliberately provocative.
His logic is straightforward: Solar in space is ~5x more effective. No day/night cycle. No batteries. Fewer permitting constraints. No grid interconnection delays.
But let’s be clear: Launching data centers into orbit is not an easy problem. This is not a pragmatic solution for the next AI build cycle. It is a leapfrog strategy.
Elon explicitly states that the real bottleneck in terrestrial power is gas turbine blades and vanes — a casting constraint shared by only a handful of suppliers globally.
That’s not a physics problem. That’s a supply-chain and manufacturing scale problem.
Compared to orbital solar arrays, space-rated GPUs, hourly Starship launch cadence…
…scaling turbine blade casting is vastly more solvable.
To learn more, read the full take on Substack “Power Signal” at this link.
