Two days ago the semiconductor stocks fell 11%, and the financial press reached for the bubble headlines. Then Thursday showed up and quietly mugged the narrative.
The PHLX semiconductor index — the main scoreboard for chip stocks — rose 3.06%. That puts it back up 83% on the year.
The spark was not the usual star. It was Micron (MU), which jumped 4.5% after unveiling a $250 billion U.S. manufacturing plan. That number used to be a decade of spending for the entire industry.
The rest of the memory bench followed it out the door: Sandisk (SNDK), Western Digital (WDC), ON Semiconductor (ON), Marvell Technology (MRVL), and Advanced Micro Devices (AMD).
What sat out is the real tell. Nvidia (NVDA), Alphabet (GOOGL), and Microsoft (MSFT) barely moved.
In a normal week, those three names are the whole story. This week they were the audience.
Shovels get paid first
The logic is old and blunt. Nobody knows which AI model or assistant wins the decade. Everybody knows the buildout eats staggering amounts of memory, and memory pricing is the tightest bottleneck in the chain.
So investors are asking, with real money, when the hyperscalers — the giant cloud companies building AI data centers — turn an estimated $710 billion of AI spending into actual profit. While they wait for the answer, they are rotating toward the suppliers being handed that check. A rotation — money moving out of one group of stocks and into another — is exactly what Thursday was.
SK Hynix now lists in the U.S. at roughly six times forward earnings — six times next year’s expected profit. That is either the cheapest seat in the theater or a warning label. Memory is the most boom-and-bust corner of tech, and six-times is what the market charges for that history.
Not everything got a pass, either. Salesforce (CRM) slid 2.2% on an analyst downgrade the same day.
The sizing question
Here is the part a market column will not tell you, and it is the whole planning point. The real decision is not “do I believe in AI.” It is how much of your retirement rides on it.
That is a concentration question — how much of your money sits in one theme or one stock. As a general planning principle, an AI sleeve should be built so an 11% chip week is a headline, not a lifestyle change.
Maybe you own the theme only through the giants. Maybe you own nothing and feel the itch to fix that in one afternoon, at week-one-of-a-rotation prices. Either way, the second look belongs on the sizing, not the ticker.
Sizing is not a prediction. It is an admission that nobody rings a bell at the top, so the plan cannot be built to need one.
You do not wait for the storm to learn the roof leaks. A review is fifteen minutes. Bring your latest statement, and we will see whether your AI exposure is sized for the week the chip index drops another 11% — and you shrug.