When the chip bottleneck shifts to energy constraints

John on $NVDA: Even if TSMC does scale up magically, or they figure out how to build another fab, or Arizona increases capacity, you could still wind up in a weird scenario: Nvidia has the chips, they're ready to sell them, the hyperscalers want to buy them, and the AI labs want to run inference on them, but there's just not enough energy. The question is, when does the chip bottleneck shift to the energy bottleneck? That could be part of what is worrying people.