The clear target for AI investment is chips, while the hidden link keeps switching.

In the first half of the year, storage took over the baton from computing power. WDC doubled from its lows, MU's HBM orders are scheduled until next year.
The market has already priced in the logic that "AI needs to store data."
But there’s one layer that hasn't been fully priced: electricity.
Training a GPT-5 level model costs tens of millions of dollars in electricity each run.
VRT works on data center cooling, CEG supplies nuclear power directly—these aren’t tech stocks, but with each step forward in AI, they take a bigger share.
Storage has been driven up, but electricity and heat dissipation might be the next wave.
My logic is simple: the most certain aspect of the AI arms race isn’t who will win, but that the physical layer will always be insufficient.
Insufficient computing power → buy chips.
Chips heat up → need cooling.
Cooling → need electricity.
Every link in this chain will be cycled through. #NVDA