SK hynix produced numbers that would normally define a successful semiconductor cycle. Its official second-quarter release reported 79.3 trillion won of revenue, 60.5 trillion won of operating profit and a 76% operating margin. Revenue rose 257% from a year earlier; operating profit rose 557%. The shares fell anyway, and South Korea's technology-heavy market extended a historic decline.
The apparent contradiction is the story. The quarter confirmed that scarce high-bandwidth memory can generate exceptional economics. The market reaction showed that exceptional economics were already in the price. Once expectations assume scarcity will last, a record quarter is no longer enough; investors need evidence about how long pricing, yields and capacity discipline can resist normalization.
A record quarter can still miss the price embedded in the stock
Reuters reported that SK hynix's operating profit fell short of a 64 trillion won LSEG SmartEstimate and revenue missed an 84 trillion won estimate. Its earnings report on the market reaction also attributed part of the gap to slower HBM4 shipment recognition. A sixfold profit increase can therefore coexist with disappointment when the share price discounts an even steeper result.
The net-income line deserves particular caution. SK hynix reported 93.9 trillion won of net income, above operating profit. Reuters attributed 63.3 trillion won to gains on investment assets, linked by analysts to the Kioxia stake. That gain strengthens reported equity and cash capacity, but it does not reveal the recurring economics of producing memory. Operating profit and cash conversion are the cleaner measures of the manufacturing cycle.
This distinction also explains why valuation cannot be assessed from growth rates alone. A low comparison base magnifies year-on-year percentages, while the stock reflects the expected level and duration of future cash flows. The relevant surprise was not whether business was strong. It was whether strength exceeded a market already positioned for near-perfect execution.
Seventy-six percent margin is a scarcity reading
A 76% operating margin is not merely evidence of efficiency. It is a price signal from a market where demand exceeded available supply. SK hynix said high-value DRAM, HBM and enterprise SSD products drove the quarter, and that both DRAM and NAND prices increased. It began mass shipments of HBM4 and said 321-layer products had become the largest share of NAND production.
Scarcity has several layers. Leading HBM requires advanced packaging, acceptable yields, power efficiency and customer qualification; adding wafer capacity alone does not instantly create interchangeable supply. That supports the counterargument to the selloff: Chinese memory producers may add pressure in commodity DRAM before they can substitute for qualified HBM. Reuters reported that distinction, while also documenting investor concern over CXMT's capacity and technical progress.
But a scarcity margin creates its own incentive to invest. SK hynix is accelerating M15X, preparing the Yongin cluster and expanding packaging and NAND facilities. Samsung is also increasing capacity. The Associated Press reported record Samsung profit alongside market concern that enormous manufacturing investments may not earn adequate returns. Today's margin is therefore both evidence of a moat and an invitation for the industry to spend against it.
Contracts trade upside for visibility
SK hynix said it had finalized long-term agreements with about ten customers and was discussing more. These arrangements can protect utilization and help fund capacity, converting uncertain spot demand into a more bankable order book. They also change the upside. If contract terms limit price resets during a shortage, some future scarcity rent moves from the supplier to the customer.
That trade is rational in a cyclical business. The company ended the quarter with 88 trillion won in cash and cash equivalents, 18.6 trillion won of debt and a 69.4 trillion won net-cash position. Financial flexibility allows it to build capacity without repeating the balance-sheet stress associated with past memory downturns. Yet capital discipline must be judged when plants open, not when plans are announced.
The contract and capex questions meet at customer economics. Hyperscalers can request more HBM today, but sustainable demand ultimately depends on revenue or strategic value from AI services. Multi-year deposits and enforceable purchase commitments would make that demand more durable; loose volume intentions would transfer more risk back to SK hynix. Disclosure on contract protections, capital intensity and customer concentration would change the quality of the earnings thesis.
Korea's index turned one earnings miss into a market event
Company fundamentals alone do not explain the speed of the decline. A Reuters market report showed the KOSPI fell 10.84% on July 28, with SK hynix and Samsung representing more than half of the index. The following session brought another deep intraday fall as SK hynix results arrived. Concentration translated one industry's repricing into a national benchmark shock.
Leverage amplified it. Reuters described borrowed retail exposure and forced position closures as accelerants, while light trading volume suggested buyers had withdrawn. This means the index decline cannot be read as a clean forecast of memory demand. Part was a change in the fundamental narrative; part was market structure liquidating a crowded trade.
Evidence that would reverse the skeptical interpretation is specific. Faster HBM4 shipment conversion, durable customer commitments, stable yields and capex that does not outrun contracted demand would support a longer scarcity regime. Weaker pricing, rising inventories or capacity additions ahead of demand would show the margin was a cyclical peak. SK hynix has already proved that AI memory can produce extraordinary profit. The harder task is proving that the industry's response to that profit will not erase it.