Elon Musk Warns of “Stranded Compute” as Power Becomes the AI Bottleneck

Elon Musk says the AI infrastructure boom could create “a lot of stranded compute in partially complete data centers.”
The problem is timing.
GPUs can arrive faster than the power, substations, cooling and grid connections required to actually run them.
That sounds bearish for AI infrastructure at first.
But it could have the opposite effect on usable compute pricing.
If large amounts of hardware sit idle waiting for power, the effective supply of operational AI compute remains constrained even as total GPU shipments rise.
That would benefit operators that already have energized data-center capacity and could keep GPU rental rates elevated for longer.
The risk falls on companies holding expensive chips without the infrastructure needed to monetize them.
GPUs also depreciate quickly as new generations arrive, meaning idle hardware can lose economic value even before it begins generating revenue.
For $SPCX and other operators with secured power, the bottleneck could therefore become an advantage.
The bigger AI constraint may no longer be getting the chips.
It may be getting enough electricity online fast enough to use them.