Nine days ago the question on this site was whether the KOSPI’s October collapse was a buying opportunity or the start of a longer rethink of the AI trade. SK Hynix answered part of it on Wednesday. Third-quarter revenue came in at 104.6 trillion won, the first quarter above 100 trillion in the company’s history and up 328% from a year ago. Operating profit was 83.3 trillion won, up 632%. Both beat consensus by 8% to 9%.
The margin is the number to sit with. 79.7% operating margin, after 72% in the first quarter and 76.3% in the second. That is a commodity memory maker earning margins that most software companies never see. The stock closed up 10.9% at 1,487,000 won, Samsung Electronics added 9%, and the KOSPI jumped 6.6% to 6,012, back above 6,000 for the first time in ten sessions. Foreigners were net buyers for a second day after three days of selling.
Why the margin may not be the peak
The usual memory script says margins like this attract capacity, capacity arrives, prices break. SK Hynix spent the call arguing the script is broken this time. Management says demand will exceed supply through 2030. Two reasons: leading-edge DRAM and HBM processes are hard to ramp, and new fabs take years to build. The first Yongin fab only reaches full operation next year.
The stronger evidence is the order book. The company now has long-term agreements with 14 large customers that cover more than half of its output for 2027 and 2028. The contracts carry a price mechanism that moves with the market inside an agreed range, and some customers are asking to extend terms beyond five years. When buyers ask for longer contracts at moving prices, they are telling you they expect to be short.
Prices back that up. DDR4 8Gb was about $26 at quarter-end against $6.30 a year earlier. 128Gb MLC NAND is roughly eight times its year-ago level. HBM4 is expected to pass HBM3E in both volume and revenue in the fourth quarter, so the mix gets richer from here, not poorer.
The payout is the tell
On top of the 40 trillion won buyback finished last week, the board added 60 trillion won: 30 trillion in cash dividends and 30 trillion in buybacks and cancellations. That brings 2026 shareholder returns to about 100 trillion won. The third-quarter dividend is 20,000 won a share, half regular and half a special payout funded by gains on the Kioxia stake. A new 20 trillion won buyback was approved too, though only half of it will be cancelled. The rest goes to employee compensation.
Companies that fear a cycle top hoard cash. SK Hynix is handing out roughly as much as it plans to spend on new capacity, and it still sits on 98.7 trillion won of net cash. That is either supreme confidence or the cleanest signal yet that the board sees no cliff in 2027.
What could still go wrong
The cost lands on someone. CFO Kim Woo-hyun expects PC and phone shipments to fall because memory is too expensive and too scarce. We already saw smartphone shipments drop 20% on memory costs. Several PC makers are reportedly pulling assembly back to China to offset the bill. If device volumes keep shrinking, the demand that supports these contracts narrows to a handful of hyperscalers, and that is a concentration risk the margin hides.
China is the other variable. CXMT is reportedly starting HBM3E production with yields of 20% to 30%. That is not competitive today. It is a start.
For holders of the leveraged chip ETFs we looked at yesterday, Wednesday is a reminder that the swings run both ways and fast. For long-term holders, the more useful number is the contract coverage. More than half of two future years sold before the wafers exist. That is not how a peak usually looks.
Earlier on this theme: Memory Stocks Are Pricing a Peak While 2027 HBM Prices Are Set to More Than Double and Where Is the KOSPI Heading.