SK Hynix Beat Every Record. The Stock Fell Anyway.

The memory chipmaker posted its best quarter on record, but a narrow miss against sky-high estimates, layered onto fresh competitive fears from China, still sent its stock reeling.

Key Takeaways

• Revenue reached 79.32 trillion won and operating profit 60.54 trillion won, up 257 percent and 557 percent year over year, both records that still fell short of consensus.
• SK Hynix shares fell nearly 9 percent on Nasdaq and the KOSPI dropped 10.8 percent as a Chinese IPO, an Nvidia financing report, and a lithography advance struck the same week.
• HBM4 shipments began, cash climbed to 88 trillion won, and 2026 capital spending now targets the high end of a 40 trillion won range as SK Hynix expands under strict discipline.

The Anatomy of a Disconnect

Some earnings reports need footnotes to explain why the market moved against them. SK Hynix’s second quarter needs almost none, which is precisely what makes it worth studying. Revenue reached 79.3187 trillion won, up 257 percent from a year earlier; operating profit reached 60.5426 trillion won, up 557 percent, figures so large they briefly obscured that they were also, in Wall Street’s language, a disappointment. Net profit reached 93.9226 trillion won, a margin of 118 percent, lifted by 63.3 trillion won in gains from asset sales and revaluations. First-half revenue crossed 100 trillion won for the first time in company history, with basic earnings per share of 132,126 won.

By the close of trading the following day, SK Hynix’s (NASDAQ: SKHY) shares had fallen close to 9 percent, from 143.02 dollars to their lowest level since the company’s July 10 listing. In Seoul, the common stock swung as much as 20 percent lower intraday before settling down 14.7 percent, and the KOSPI index fell 10.8 percent, its worst session since March. Samsung Electronics lost more than 13 percent in sympathy. A quarter built almost entirely of superlatives produced a reaction that looked, for a few hours, indistinguishable from a profit warning.

Where the Number Broke

The explanation sits in a single word carrying disproportionate weight this cycle: consensus. Local brokerages had modeled operating profit near 64.7 trillion won; LSEG’s SmartEstimate, weighted toward analysts with the strongest recent accuracy, pointed to something close to 64 trillion won. Against either benchmark, actual profit fell short by roughly 6 percent, an unremarkable gap in most industries and an unforgiving one in a sector where valuations assume flawless execution. A profit increase exceeding fivefold, in an ordinary market, reads as vindication. Here, it read as a warning that AI memory earnings might carry a lower ceiling than investors had priced.

Executives offered a more prosaic account than the reaction implied was necessary. Shipments of certain high-value products, HBM4 among them, slipped into the second half, and the resulting shift in mix pulled down the blended DRAM selling price. Management expects both volume and pricing to strengthen through the third and fourth quarters as HBM4 output ramps and the migration to 1c-nanometer DRAM continues. The gap between guidance and delivery may prove a matter of timing rather than demand, though markets rarely wait patiently to find out which.

A Week Built to Overwhelm

The earnings miss did not arrive in a vacuum. It landed inside a single week that delivered three unrelated shocks to the same trade. ChangXin Memory Technologies, China’s largest DRAM producer, debuted on Shanghai’s STAR Market and surged 466 percent on its opening day, raising roughly 8.6 billion dollars in the largest semiconductor listing in mainland Chinese history and briefly touching a market value near 487 billion dollars. The debut reawakened a fear that has shadowed the industry for years: that Chinese supply, once mature enough, could erode the pricing power on which SK Hynix’s margins now depend.

A second jolt came from reports that Nvidia was negotiating financing worth as much as 250 billion dollars to help OpenAI secure a data center campus in Ohio. Critics described the arrangement as circular, since capital extended by Nvidia would eventually return to it as chip revenue, and the concern traveled into credit markets. A third development, a report that a Shanghai manufacturer had begun mass-producing immersion lithography tools domestically, chipped at the assumption that Chinese chipmaking remained dependent on Dutch equipment maker ASML, though the technology trails the leading edge by several generations. None of the three touched SK Hynix’s own operations, yet together they framed how its earnings were read.

The Line CXMT Has Not Crossed

CXMT’s advance is genuine, though its geography matters more than the headline suggests. Its gains sit in commodity DRAM, the standard memory used in phones, laptops, and general-purpose servers, where it holds roughly 7.7 percent of the global market and where its process still trails Samsung and SK Hynix by two to three generations. High Bandwidth Memory, the product behind most of SK Hynix’s current profit, rests on a different foundation entirely. Rather than a single die, it stacks multiple DRAM layers vertically, connects them through microscopic vertical wires, and mounts the assembly beside the processor on a silicon interposer, producing bus widths many times wider than standard DDR5 allows.

Reaching that architecture at scale requires packaging capability CXMT has not yet shown, and its HBM output remains a fraction of SK Hynix’s total DRAM wafer capacity. The lithography breakthrough changes where China sources its tools, not whether Chinese firms can execute the packaging HBM demands. SK Hynix began mass shipments of HBM4 in the second quarter, with a fuller ramp planned for the second half, and industry estimates place its share of Nvidia’s HBM4 allocation for the Vera Rubin platform at roughly 60 to 70 percent, with Samsung and Micron dividing what remains. Samples of the next-generation HBM4E have already reached customers, built on a process the company describes as mature and stable.

Fortifying the Balance Sheet for a Longer Cycle

The balance sheet tells a calmer story than the share price. Cash and short-term investments rose to 88 trillion won, up 33.6 trillion won in a single quarter, while interest-bearing debt fell to 18.6 trillion won, leaving net cash at 69.4 trillion won and a debt-to-equity ratio of 7 percent. That strength arrives as investment needs are widening. Full-year 2026 capital spending is now guided to the high end of a roughly 40 trillion won range, reflecting an accelerated schedule at the M15X facility and preparation for the Yongin Phase 1 cleanroom, due to open in early 2027. Further commitments, including a new packaging facility designated P&T7 and a new NAND base called M17, will be phased in against confirmed customer demand rather than built ahead of it.

SK Hynix has also finalized long-term supply agreements with roughly ten customers, contracts typically running five years that pair volume commitments with deposit mechanisms designed to dampen short-term price swings for both sides. Management said it is reviewing further shareholder return measures and intends to disclose specifics before year-end.

The Trade Ahead

The timing of the selloff deserves its own scrutiny. It arrived less than three weeks after SK Hynix’s own Nasdaq debut, an offering that raised 26.5 billion dollars, the largest listing ever completed in the United States by a non-American company, surpassing Alibaba’s 2014 offering. Stock priced at a premium during extraordinary demand leaves less room for disappointment than stock priced conservatively, and SK Hynix shares had already fallen roughly 47 percent from their June peak in the six weeks before this report. Rate expectations added further pressure: under Chairman Kevin Warsh, the Federal Reserve signaled that more policymakers now favor higher rates this year than previously expected, a shift that makes richly valued growth stocks costlier to hold.

The company’s own guidance, set against that pressure, points toward acceleration rather than retreat. Third-quarter DRAM bit shipments are expected to rise roughly 10 percent from the second quarter, full-year DRAM demand growth is projected in the mid-20 percent range, and NAND demand growth in the high teens. Management has characterized recent efficiency gains in AI models as an expansion of the addressable market rather than a threat to it, arguing that cheaper inference draws in more users than it saves in memory. Whether that holds, and whether this quarter’s miss proves a timing artifact rather than the first sign of a slower cycle, will decide whether the current share price marks a floor or simply a pause before the next leg down.