The number that moved the stock was the shortfall against consensus. Revenue of 79.32 trillion won against roughly 83.4 to 84 trillion expected, operating profit of 60.54 trillion against roughly 64 trillion — a 5.5 percent miss on the top line and a 6.6 percent miss on operating profit. The number that describes the business is a different one. DRAM average selling prices rose about 30 percent sequentially. NAND average selling prices rose in the mid-50 percent range sequentially. Enterprise SSD revenue more than doubled quarter over quarter. Those are not the marks of a demand problem, and treating a 5.5 percent variance against a forecast as evidence of one requires ignoring what the forecast was measuring.
Consensus for this quarter was built by analysts modeling a business whose operating profit grew 557 percent year over year and whose revenue grew 257 percent. The error bars on any such model are wider than the miss. When a company reports 60.5 trillion won in operating profit against an estimate of 64 trillion, the useful question is not whether the estimate was beaten but whether the estimate was constructed on a basis that survives contact with the quarter. Management’s own explanation was mix and timing: shipments of certain high-value-add products slipped into the second half, and the resulting change in product portfolio moved the blended ASP. That is a sequencing variance, and sequencing variances reverse or they do not. This one is testable within ninety days.
The strategic content of the quarter sits underneath the headline. SK Hynix began mass shipments of HBM4 in the second quarter and has completed customer sampling on HBM4E, with capacity expanding through the M15X and Yongin facilities and 2026 capital expenditure now guided to the upper end of the previously indicated 40 trillion won range. It has also finalized long-term supply agreements with approximately ten customers and is extending multi-year contract discussions further out. Those contracts are the single most underappreciated item in the release. A memory manufacturer that converts spot exposure into multi-year committed volume is doing the one thing the industry has never historically been able to do, which is decouple its revenue line from the cycle it cannot control.
That is also where the moat argument has to be examined honestly rather than accepted. SK Hynix’s competitive position rests on HBM, and specifically on being the primary supplier into the accelerator platforms that define AI compute demand. That position is real, it is difficult to displace on any timescale shorter than a product generation, and it commands pricing that produced a 76 percent operating margin this quarter. But concentration is not the same thing as durability. The company skewed hard enough toward high-bandwidth memory that its conventional DRAM franchise thinned, and the response visible in this quarter’s commentary is a deliberate pivot back — LPDDR6 deliveries targeted within the second half, and LPDDR6 brought to the SOCAMM standard for data center use. A supplier that has to announce it is re-entering the commodity end of its own market is telling you where it thinks the next competitive pressure originates.
It originates in Hefei. CXMT’s Shanghai listing landed days before this print and reframed the supply question in a way earnings cannot answer. CXMT is the fourth-largest DRAM producer at roughly 8 percent share, against Samsung near 36 percent, SK Hynix near 29 percent, and Micron near 24 percent. Eight percent does not clear a shortage, and CXMT remains constrained by export controls on advanced fabrication tooling, sits on the Pentagon’s list of firms with alleged military ties, and faces legislative interest in restricting American purchases. None of that matters to the multiple in the way the reported interest from tier-one Western customers in qualifying Chinese DRAM does. The bear case for the memory cohort was never that AI demand evaporates. It is that the incremental supplier changes the shape of the pricing curve at the exact moment incumbents are capitalizing peak margins into permanent expectations.
The stock reaction has to be read against a July that was already brutal. SK Hynix fell roughly 47 percent from its June highs into this print, SanDisk lost more than half its value, and Micron and Western Digital each shed roughly a third — this after year-to-date gains that ran into the triple digits and, for SanDisk, past 500 percent. The ADR, listed on Nasdaq only weeks ago, closed at its lowest level since debut on the day of the report, with the Seoul line down 6.5 percent. A cohort in a 47 percent drawdown does not fall another 6 percent because of a 5.5 percent revenue variance. It falls because the marginal holder is positioned in leveraged single-stock and thematic memory products whose daily reset mechanics force selling into weakness, and because a narrow thematic basket in which three names account for the large majority of assets transmits any single-name shock across the entire cohort within the same session.
The macro overlay compounds it. West Texas Intermediate sits near 83 dollars after a month in which crude swung roughly 20 percent higher on alternating Iran escalation and de-escalation headlines, and the FOMC concludes a two-day meeting today under a chair who has abandoned forward guidance and declined to publish individual projections. Futures markets are pricing meaningful odds of at least one quarter-point increase by September and non-trivial odds of more by year-end. Memory equities were re-rated during this cycle on the argument that AI demand is structural rather than cyclical, which is a duration argument. Duration assets do not care whether the earnings were good when the discount rate is being repriced upward on energy-driven inflation. That is most of what happened this week, and it has nothing to do with what SK Hynix shipped.
One further item deserves scrutiny that it will not receive in the coverage. Net profit of 93.92 trillion won exceeded revenue, producing a net margin above 100 percent. Whatever generated that is non-operating, and it should be excluded from any forward estimate built off this quarter. Analysts who anchor next year’s model to a headline net figure inflated below the operating line will produce another consensus that the company misses while performing well.
The decision-relevant number is the third-quarter shipment guide: DRAM bit shipments up approximately 10 percent, NAND up approximately 3 percent. If the high-value-add volumes that slipped out of the second quarter land inside that, the miss was a calendar artifact and the pricing environment is intact. If they do not, the mix explanation was a description of demand rather than of timing.