A 557% Profit Surge That Counts as a Miss: What SK Hynix's Quarter Reveals About the AI Trade

Consider the sequence. On Monday, Korean chip stocks suffered their worst session of the year, with the KOSPI briefly down 10 percent, on fears that the AI boom's financing is circular and China's toolmaking breakthrough is real. On Tuesday morning, SK Hynix reported the largest quarterly operating profit in the history of Korean business — 60.5 trillion won, up 557 percent, at a 76 percent margin. It missed estimates. The stock rose anyway.

Each of those facts is individually explicable. Together, they describe a market that no longer knows what it is pricing.

The miss that wasn't about demand

Start with the miss itself. Consensus wanted roughly 64 trillion won of operating profit on 84 trillion won of revenue; SK Hynix delivered 60.5 trillion won on 79.3 trillion won. Nothing in the release suggests demand softness — DRAM prices rose about 30 percent quarter on quarter, NAND in the mid-50s, supply remains effectively sold out, and the company steered capex to the top of its 40–50 trillion won range.

What the miss reveals is that sell-side models have gone vertical. When estimates compound 500-percent growth quarters into the indefinite future, "disappointment" becomes a mechanical inevitability — a property of the expectations curve, not the business. This is the first structural fragility of the AI trade: the numbers can be historic and insufficient simultaneously.

Scarcity is doing strange things

The second signal comes from the product mix. Digitimes reports both Samsung and SK Hynix are reallocating flexible DRAM capacity toward server DDR5, because some 64GB RDIMMs now generate more revenue per wafer than HBM — the supposedly premium product the entire supercycle narrative is built on. Meritz Securities estimates memory suppliers will meet only 75–80 percent of demand in the second half of 2026, possibly deteriorating to 60 percent in 2027.

Read that carefully: the shortage is now so broad that the commodity product out-earns the flagship. That is spectacular for near-term margins, and quietly corrosive for the long-term story — because it means today's profits are a scarcity phenomenon as much as a technology one, and scarcity always ends. CXMT's 466 percent Shanghai debut this week is the market putting a trillion-yuan price tag on exactly how it ends.

Ten contracts and a leap of faith

SK Hynix's best argument for durability is structural: ten customers on long-term agreements, with AI data center demand approaching 70 percent of revenue. Contracted volume converts the memory cycle's historical whiplash into something resembling infrastructure revenue.

But the counterparties to those contracts are the same hyperscalers and AI labs whose financing structures spooked the market on Monday — the reported $750 billion web of Nvidia-linked commitments, the $250 billion OpenAI backstop under discussion. Long-term agreements are only as durable as the capex budgets behind them. SK Hynix has effectively swapped cycle risk for counterparty concentration risk, and the market spent Monday repricing precisely that.

What Tuesday actually settled

The rebound — KOSPI up 2.8 percent, the stock up 3.7 percent — settled the narrow question: demand is real, cash is being paid, the quarter was not the top. It did not settle the broader one. A market that punishes a 557 percent profit surge for missing by five percent, then forgives the miss within hours, is a market trading on conviction about 2028, not results from 2026.

For the Asia AI trade, that is the honest takeaway from this remarkable 48 hours. The fundamentals have never been better; the expectations have never been more demanding; and the gap between them is now the primary source of volatility. Record quarters no longer move the story forward. Only guidance does — and guidance, unlike operating profit, cannot be audited.