Hey Traders,

The selloff is hitting chip and semiconductor names harder. The physical infrastructure builders are holding up better. Let me walk you through it.

Picture a packed stadium when the crowd suddenly bolts for one exit. The game on the field did not change, everyone just decided to leave through the same door at the same time.

Advanced Micro Devices (AMD), Coherent (COHR), and Broadcom (AVGO), the semiconductor and photonics names (photonics is tech that moves data with light instead of electricity), were the most crowded longs coming into last week. A crowded long just means everyone piled into the same bet. They ran the hardest, and they had the most borrowed money stacked underneath them.

So when the rotation started Friday (money rotating out of one group of stocks and into another), they fell first and fell hardest. The data showed it clearly.

But look at what held up better: Power, cooling, construction, the physical layer of the buildout. These names already had their run earlier this year, back when I first flagged them.

GE Vernova (GEV) and Quanta Services (PWR) led that move. By the time this selloff hit, they were not the crowded trade anymore. The market had already worked off the froth, the speculative excess that piles up when a trade gets too popular.

That is the distinction I keep coming back to. The market had priced the digital side, the chips and software and semiconductor equipment, for perfection, so expensive that everything has to go right just to justify the tag. I said exactly that on BNN Bloomberg the day before Friday's selloff.

The physical AI trade is different. It is the picks and shovels of the buildout, the gear and labor every builder needs no matter which chipmaker wins, like selling shovels to every miner in a gold rush. It runs less on AI hype and more on signed construction contracts for the nonstop job of retiling the Earth for AI.

Data centers still need power. They still need cooling. Those contracts did not vanish because Nvidia (NVDA) fell three percent on a Friday.

Here is the part worth saying plainly: Both trades rest on physical scarcity.

Not enough chips, not enough memory, not enough power, not enough cooling.

The AI buildout is a five-year story at minimum. Both layers feed off the same constraint. You cannot build fast enough to meet what the market is demanding.

That scarcity does not care about a rotation. It does not care about a crowded unwind on a Friday. Companies have already made the capital commitments.

The hyperscalers, the giant cloud companies like Amazon, Microsoft, and Google, disclosed their spending. They signed the contracts. The only question is who delivers, and both the chipmakers and the infrastructure builders sit on that list.

You can see the scarcity in the order books.

GE Vernova says its gas turbines are effectively sold out through the end of the decade, sitting on a backlog north of $160 billion, with customers already reserving production slots for 2030 and 2031. That demand does not unwind on a Friday.

So yes, the selloff is treating them differently right now. But zoom out and they are the same theme. Five years of buildout, physical bottlenecks the whole way, and a market that keeps underestimating what it costs to run AI at scale.

Own both. Size them for where we sit in the cycle. I like the infrastructure names better in the short run, because semiconductors have hoovered in an enormous amount of AUM lately (assets under management, the dollars investors pour into a fund).

Look at the Roundhill Memory ETF (DRAM). It ran from $77 million in AUM to roughly $15 billion in a matter of weeks. Yes, you read that correctly: $77 million to $15 billion, with a B.

And for the IPO of an unexpected AI company named SpaceX, tap this link to join Andrew on Thursday as he pre-games (and makes a play) on Friday’s IPO.

Here for a good time… AND a long time… and a safer time!

Hans