Hi Traders,

Last Tuesday, on SpaceX's first earnings call as a public company, someone asked Elon Musk what's slowing the AI buildout down. He didn't say chips. He said memory.

Memory output grows about 20 percent a year, Musk told analysts. Demand grows 200 percent, maybe faster. His conclusion: "Economics 101 would suggest that the price increases. It does not decrease."

Normally I'd file that under one guy talking his book. Musk buys enormous amounts of memory for Tesla and SpaceX, so he'd love for everyone to know he's supply-constrained. Except he's the seventh executive to say a version of this in 2026, and the other six don't work for him.

Seven People Who Sign the Checks

Tim Cook told the Wall Street Journal in June that this is a hundred-year flood, and that he'd never seen anything like it in over 40 years. He's spent those decades inside electronics supply chains at IBM, Compaq and Apple. Apple is raising device prices because of it.

Jensen Huang told memory makers to build all the fabs they want, because Nvidia will absorb whatever comes out of them. Michael Dell walked the arithmetic at a Bank of America event in April: roughly 25 times more memory per AI accelerator by 2028, times 25 times more accelerators, which lands on 625 times the demand.

Skeptics say the realistic number is closer to 180 times. That's still a number that breaks a supply chain.

Micron's Sanjay Mehrotra sees tightness running into 2027. SK Group chairman Chey Tae-won went further at Nvidia's GTC in March, saying the wafer shortage runs to 2030 with supply short by more than 20 percent. And on July 30, Amazon lifted its 2026 capital spending to $220 billion from $200 billion, named memory as the reason, and said it still wouldn't have enough capacity this year or next.

Then the Group Got Taken Apart

While those seven were describing a shortage with no end, memory stocks came unglued. The Philadelphia Semiconductor Index fell roughly 19 percent through late July, tracking its worst month since 2008. Micron, Samsung, SK Hynix and the Roundhill Memory ETF each dropped more than 20 percent off their highs during the slide, and SanDisk lost 36 percent in five sessions.

The selling wasn't about demand. SK Hynix posted a record 76 percent operating margin, then guided 2026 capex up 50 percent to $31 billion, which spooked people into thinking the AI spending cycle is overheating.

Add China's CXMT ramping supply, and what you get is a crowded trade unwinding. Not a thesis breaking.

Own the Toll Booth, Not the Traffic

Cheaper AI models were supposed to be bad for this trade. They're the opposite. When intelligence gets cheap, people use far more of it, and every unit of compute drags memory along with it.

So, three ways to own the bottleneck instead of the application. Micron (MU) for DRAM and HBM, the high-bandwidth memory that sits right next to the GPU. SanDisk (SNDK) for the NAND flash storage all that data has to live on.

And SK Hynix (SKHY) for the HBM inside Nvidia's top systems. Fair warning on that one: the ADR (a US-listed proxy for the Korean shares) only started trading here on July 10, the float is thin, and it swings 20 percent in a session without much provocation.

Compute is the story everyone tells. Memory is the toll every one of those stories has to pay.

Be there when I play it and tap this link to join my Turbo Income Trading service.

Here for a good time AND a long time,

Han