Coatue funded the models. Now Coatue is funding the power plants that run the models. Thats the clearest signal yet that the marginal dollar in AI has moved down the stack from software to infrastructure.

The logic is straightforward once you see it. Model improvements are hitting diminishing returns on a per-dollar basis. Power availability is hitting increasing returns because every additional gigawatt unlocks capacity for multiple hyperscalers simultaneously. If youre an allocator trying to decide where the next 10x sits in the AI value chain, the power layer has a structural advantage: models commoditise, but the grid interconnection queue is a five-year bottleneck with no shortcut.

The fact that its Coatue making this move rather than a traditional energy fund is the part worth paying attention to. They have better visibility into AI compute demand than almost any other allocator because they sit on the cap tables of the companies generating that demand. When the investor with the best forward demand data pivots to infrastructure, thats a pricing signal about where the constraint has moved.