The Cycle That Was Supposed to Be Cancelled

Five days ago Samsung reported a profit up ninefold, around $80 billion. Twelve days ago Micron’s CEO said memory would be tighter in 2027 and 2028 than in 2026. This morning the Kospi is down more than 7% because memory stocks are being sold, and Reuters calls it a “brutal session” that “shows AI trade is turning.”

Memory is DRAM and flash, and DRAM is a commodity. For forty years it has been the most reliably cyclical business in technology: shortage, panic ordering, new fabs, glut, bankruptcy, repeat. A ninefold profit jump is what the top of that cycle looks like. When the CEO says supply stays tight for two more years, that’s the sound of everyone justifying the fab spending that eventually ends the shortage.

The market’s reaction isn’t about this quarter’s earnings. It’s the sudden recollection that the cycle never got cancelled. Friday’s “AI revenue doubts” headline was the same worry arriving early.

Last week I wrote about the AI put, the idea that one sector’s earnings are holding up the bond-sold-off equity market. A put has a strike, and Korea may be the first place the market found it. A single index whose biggest members are memory makers turns out to be a leveraged bet on one price.

Meanwhile Goldman jumped on earnings, and the bank beat is the volatility trade from Sunday paying out on schedule. Somebody’s bad morning is in somebody’s good quarter. I’d like to know which desk was on the other side of Seoul.