SK hynix just posted the best quarter in its history, and the stock fell 14.7% in Seoul.

Operating profit came in at ₩60.54 trillion, up more than 550% year over year. Revenue hit ₩79.32 trillion, a 257% jump. DRAM average selling prices rose roughly 30% quarter over quarter and NAND rose about 50%. Every one of those figures is a record.

The market wanted more. Consensus had penciled in ₩64 trillion of operating profit, so the print landed as a miss despite being extraordinary in absolute terms. Micron fell 8.9% in sympathy, closing at $820.53 and finishing 32% below its June 25 peak.

An expectations reset, not a cycle turn. Nothing in SK hynix’s numbers says memory pricing rolled over. Guidance points the other way: DRAM shipments up about 10% next quarter, full-year DRAM demand growth near 25%, NAND demand up 18%. Pricing is still climbing. What broke was the assumption that these companies could keep clearing whatever bar analysts set after a 250% rally. Cyclicals almost always look strongest right before sentiment cracks. That’s uncomfortable, but “peak optimism” and “peak earnings” are different events, and they rarely arrive in the same quarter.

The threat worth watching. The more durable concern arrived the same week. CXMT, China’s largest DRAM producer, completed an $8.5 billion IPO and surged 465% on debut. That capital funds capacity, and capacity is what actually ends a memory upcycle. Two things limit the near-term damage: more than 98% of CXMT’s revenue is commodity DRAM, with effectively no presence in high-bandwidth memory, where the AI margin lives, and its DDR5 cost per bit still runs over 30% above competitors. Fabs take 18 to 24 months to reach volume. The pressure is a 2028 story, not a September one.

What to actually watch. Forget the headline reactions. Monthly DRAM contract pricing is the real signal, and it leads the stocks by a wide margin. If August and September contract prices hold or rise, the miss was noise. The second checkpoint is gross margin guidance at Micron’s next report on September 29. Margin compression, not a revenue wobble, is how memory cycles announce their end.

Supply chain commentary still describes a constrained market through 2027, with consensus modeling earnings peaking in 2028. This remains a cycle, and cycles end. The evidence that this one is ending is not here yet.

This is commentary, not investment advice.