$5.5T in AI capex by 2030 isn’t a “tech sector” number. It’s starting to look like railroads / the grid.

JPMorgan recently put global AI-related capex at about $5.5 trillion through 2030. Goldman’s number for just Meta, Microsoft, Amazon, and Alphabet is already $5.3T over a similar window. Either way, that’s not software-cycle money.

At that scale it stops looking like another SaaS wave and starts looking like the old buildouts: railroads, the electrical grid, autos, then the internet. A new technology shows up first. Then the physical world has to get bigger to carry it.

Power plants, data centers, chips, copper, steel, cooling, transformers, transmission. All of that has to be built, not prompted.

What I keep coming back to: AI looks like a digital revolution on the screen. Underneath, it may be the biggest “re-materialization” trade of this century. The model is software. The constraint is concrete, metal, and megawatts.

Curious how people here are playing it. Still just NVDA / the hyperscalers, or are you looking further down into utilities, copper, grid equipment, and the boring stuff that actually has to get poured and wired?