After weeks of straight up, memory finally had a rough week. DRAM dropped up to 7+ percent in a single session and still falling afterhours. When something falls that hard you owe it to yourself to find out why, so here is the honest breakdown of the three things that hit at the same time, and why two of them are noise and one is worth watching.
One. SK Hynix announced it is doubling capacity
SK Hynix Chairman Chey Tae-won used Computex to announce a five year plan to double wafer capacity across HBM, DDR5, and NAND. The market read doubling supply as a threat to the shortage and sold off.
Here is the part everyone skipped. In the same breath, Chey said he expects the shortage to last through 2030. Read that again. The man doubling supply is saying it still will not be enough until the end of the decade. He is not expanding because the boom is ending. He is expanding because demand is so far ahead of supply that even doubling output keeps the market tight for years. And new fabs take three to five years to build, so none of this capacity even arrives inside the next two years. This is a supply headline that, read properly, is actually a confidence signal. Noise.
Two. SK Hynix is coming to the US market
SK Hynix told investors it received tremendously positive feedback on a planned US listing, potentially within this year. Micron and SanDisk dropped on the news.
The logic is subtle. Micron has been the only US listed pure play memory company, so it carried a premium as the easiest way for American money to bet on the memory crunch. If SK Hynix lists in the US, that premium leaks out of Micron because investors get a better, more direct option, the actual HBM leader with 57 percent market share. So Micron fell not because memory weakened, but because a stronger competitor is coming to its home market. And in the same disclosure, SK Hynix said it expects favorable HBM pricing to continue into next year and that Nvidia demand for its power efficient memory could tighten supply further from 2027. The dominant player just reaffirmed everything bulls believe. For the sector, this is bullish. Noise for the thesis, mild premium shuffle for Micron.
Three. Broadcom did not raise its AI guidance
This is the one that actually matters. Broadcom reported and failed to raise its AI chip revenue goalposts for next year, and that dragged the entire AI complex lower, memory included.
This is the first crack in the thing that actually drives the trade, AI demand expectations. For months the market has priced in relentless upward revisions to AI spending. One bellwether failing to raise, right after a major customer did an 80 billion dollar equity raise, made people nervous the spending might be plateauing. It is not my sell signal, Broadcom is a chip supplier, not a hyperscaler, and Google raised 80 billion three days earlier pointing the other way. But it is the first demand side question mark in this whole run, and it sits right next to the one signal I actually watch, which is hyperscaler capex guidance. Worth monitoring, not dismissing.
The pattern worth seeing
Two of the three scares were supply side, the new fabs and the oversupply narrative, and supply side fears are exactly what this thesis is built to withstand, because announced supply does not arrive inside my window and does not keep pace with demand anyway. The one that was demand side, Broadcom, is the only one that touches the part of the thesis that could actually break. The crowd sold on the two loud ones. The quiet one is the real watch item.
The stuff that confirmed the thesis the same week
While the selloff grabbed attention, the underlying signals kept pointing the other way. SK Hynix posted a 72 percent operating margin last quarter, higher than Nvidia. That is not a commodity business, that is pricing power. Contract prices for memory updated higher and are guided up through the second half. HBM remains sold out into 2027. And the demand engine kept roaring, Google raised nearly 85 billion in fresh equity specifically to fund AI infrastructure it says it cannot supply fast enough, with Berkshire Hathaway putting in 10 billion.
The one cultural signal that says it all
In Korea, SK Hynix work jumpers are now selling secondhand as the ultimate blind date outfit, and matchmaking agencies rank chip workers alongside doctors and lawyers. When a country starts reorganizing its dating market around your thesis, the supercycle has stopped being a spreadsheet and become a culture. Just remember which part of the cycle the Ferraris and the dating frenzy usually show up in.
Where that leaves me
Down week, thesis intact. The selloff was profit taking on an overheated run, triggered by two supply scares I am protected against and one demand question mark I am now watching. My tripwire was never the stock price or a scary headline. It is hyperscaler capex guidance turning down, and this week it pointed up. So I am holding, and I added a little into the drop.
I hold a long position in DRAM ETF (ticker: DRAM). Not investment advice. Do your own research.