SK hynix reported the biggest quarter in its history on Wednesday. Revenue of 79.32 trillion won, up 257% from a year ago and 51% from the previous quarter. Operating profit of 60.54 trillion won — roughly $41.7 billion earned in three months — up 557% year over year and 61% sequentially. An operating margin of 76%, an all-time high for the company. It was the fifth consecutive record quarter.
The stock fell as much as 15% in Seoul before closing down 9.6%.
The reason is arithmetic, not narrative. Analysts had modeled roughly 84 trillion won of revenue and about 64 trillion won of operating profit. SK hynix came in under both — by about 6.6% and 5.5%. A company that grew operating profit six-and-a-half-fold got marked down for the shape of the beat, because at these expectations there is no such thing as a good number anymore. There is only a number relative to the model.
Our take: A 76% operating margin is not a business, it is a shortage with a ticker symbol. Nobody manufactures physical goods at three-quarters margin in a normal market — that number exists because AI server builders need high-bandwidth memory faster than three companies can make it. The market is not disputing the demand. It is pricing the question nobody at the earnings call can answer: how long does the price stay here? When the consensus already assumes vertical, missing vertical by 6% reads as the first derivative rolling over.
The number that isn’t what it looks like
Net profit landed at 93.92 trillion won — larger than revenue. That is not a memory-chip result. It is investment gains, including proceeds tied to SK hynix’s Kioxia stake, landing in the same three months as the operating record. The business itself is the 60.5 trillion won line. Worth remembering the next time a headline compares net profit to anything.
The operating story was mix. High-bandwidth memory, DRAM for AI servers and enterprise SSDs carried the quarter, and price increases on high-value products did the rest. HBM4 entered mass production during the period, with the company saying yields and quality are already approaching the maturity of HBM3E, the generation it replaces. HBM4E samples went to key customers, with volume production targeted for 2027.
That is the part the 9.6% drop obscures. A new memory generation reaching mature yields on its first production quarter is unusual, and it is the single best evidence that the supply side is executing. It is also, eventually, the thing that ends the shortage.
The whiplash is familiar. Memory names have gone from rout to rip inside 48 hours twice this month. Micron committed $250 billion to U.S. capacity on the same thesis. And the shortage has already reached consumer prices, with phone makers passing RAM costs to buyers.
What to watch
- The HBM4 ramp curve. Mature yields at launch pull supply forward. If the third quarter shows HBM4 volumes climbing faster than guided, the 2027 pricing assumptions in every memory model get looser.
- Whether the miss was volume or price. A 6% revenue shortfall against a 257% growth rate is noise if it is shipment timing and signal if it is realized pricing. The next print separates them.
- The gap between Seoul and New York. The Korea-listed shares closed down 9.6%; the U.S.-listed line edged higher by about 1.7% on Wednesday. Two sets of investors read the same release and disagreed.
- Big Tech capex. Microsoft and Meta report after Wednesday’s close. Their capital spending lines are SK hynix’s order book with a two-quarter delay.
SK hynix just proved the AI memory trade is real and got punished for not being more real. That is what a crowded position looks like from the inside.