I love how the market spent two years pretending AI was just chips and then suddenly remembered that chips need electricity to do chip things.
Now every company connected to power, cooling, copper, data centers, generators, transformers, batteries, cables, concrete, HVAC, and maybe even office chairs is getting rebranded as “critical AI infrastructure.”
And the annoying part is that the thesis is kind of real. You actually do need power, memory, networking, cooling, land and grid upgrades to run all these AI data centers. You can’t just stack NVDA GPUs in a field and ask them nicely to train the future.
But at the same time, this is how bubbles evolve. First the real winner runs. Then the suppliers run. Then the suppliers’ suppliers run. Then some random industrial company says “data center demand” twice on an earnings call and gets treated like it invented fire.
So I’m trying to figure out where we are in the cycle. Are power/cooling/grid names still the smart second-layer AI trade, or are we already at the stage where Wall Street is buying anything with a wire attached to it?
What’s the best actual AI infrastructure play here, and what’s just getting dragged up by the narrative?