The AI trade just got more practical
For a while, the AI story was basically: buy the software names, sprinkle in some chipmakers, call it a day. But first-half 2026 ETF flow data suggests investors are now moving down the stack — from the flashy app layer to the actual infrastructure that makes AI possible.
That means money is piling into the unsexy stuff: memory chips, electrification, power equipment, cooling systems, and data-center real estate. Basically, the picks-and-shovels phase of the gold rush, except the shovels now need a lot more electricity and air conditioning.
Why the crowd is changing lanes
The clearest evidence is the Roundhill Memory ETF, which reportedly sucked in $17 billion in Q2 alone and ballooned to nearly $25 billion in assets. That’s not a gentle drift; that’s investors stampeding toward high-bandwidth memory after realizing it’s one of AI computing’s biggest bottlenecks.
And it’s not just memory. Semiconductor ETFs pulled in $23 billion in net inflows during the quarter, while issuers launched 39 semiconductor ETFs in the first half of the year. Translation: Wall Street smelled demand and responded by making a buffet out of the AI hardware stack.
The new AI bottlenecks are boring — and expensive
The next phase of the trade is looking a lot less like “which chatbot wins?” and a lot more like “who builds the stuff that keeps data centers from melting?”
Investors are leaning into:
- utilities, because AI data centers need monstrous amounts of power
- nuclear ETFs, because reliable baseload electricity suddenly sounds sexy
- HVAC and cooling names, because servers run hot and nobody likes a fried chip
- data-center REITs, because cloud providers still need somewhere to park all that compute
That’s why funds like Global X Data Center & Digital Infrastructure ETF and iShares U.S. Digital Infrastructure and Real Estate ETF are getting attention too. The market is basically saying: if AI is the party, the infrastructure providers are the ones selling the venue, the speakers, and the ice machine.
Big picture
This is what a maturing trade looks like. The easy money narrative is moving from “AI will change the world” to “who gets paid to wire the world for AI?” For investors, that’s a useful map shift — because the hottest part of the AI rally may not be the loudest names anymore.