Skip to content
VST—no feedCEG—no feedNRG—no feedTLN—no feedNEE—no feedSO—no feedDUK—no feedAEP—no feedD—no feedEXC—no feedPEG—no feedXEL—no feedGEV—no feedETN—no feedPWR—no feedVRT—no feedCMI—no feedBE—no feedNVDA—no feedAMD—no feedAVGO—no feedMSFT—no feedAMZN—no feedGOOGL—no feedVST—no feedCEG—no feedNRG—no feedTLN—no feedNEE—no feedSO—no feedDUK—no feedAEP—no feedD—no feedEXC—no feedPEG—no feedXEL—no feedGEV—no feedETN—no feedPWR—no feedVRT—no feedCMI—no feedBE—no feedNVDA—no feedAMD—no feedAVGO—no feedMSFT—no feedAMZN—no feedGOOGL—no feed
Join the waitlist
The evidence

So the bottleneck moved.

Put the three steps together and you get a claim that is specific enough to be wrong, which is the only kind worth making.

  • Everything digital is compute. The output of the industry is arithmetic, delivered on demand.
  • Compute is power. The buildings are specified in watts and the contracts are for supply.
  • Power is the constraint. The supply side moves on a decade clock and the demand side moves on a quarterly one.

The interesting consequence is about attention rather than about energy. Attention is still pointed at the chip layer, because that is where the drama has been, and where the recognisable names are. Meanwhile the layer that actually sets the rate is full of companies most people cannot name and would find boring if they could.

How this could be wrong

Stating the failure conditions is part of making the claim honestly. The argument weakens if efficiency gains at the model or chip layer outrun demand growth badly enough that total draw flattens. It weakens if interconnection reform genuinely compresses the queue. It weakens if on-site generation becomes normal enough that the grid stops being the gate.

Those are real possibilities and none of them has happened yet. If one does, this site should say so rather than quietly rewriting the thesis.