Record Profits, Falling Stocks: Memory Equities Now Trade the Second Derivative
Every memory maker is printing records, and every memory stock is falling anyway.1 Samsung's Oct 8 preliminary Q3 guided operating profit to ₩107.4T ($80.2B), +782% YoY — the first tech company ever above ₩100T in a quarter — beating the LSEG SmartEstimate of ₩106.1T. Shares "dropped 0.7% Thursday morning" anyway (CNBC), and it still missed the higher FnGuide consensus of ₩108.1T. eToro's Josh Gilbert nailed the dynamic: "Samsung has delivered a record profit and still fallen short of expectations, which tells you just how demanding the AI trade has become… When expectations are so high for memory makers, the market doesn't hand out points for effort."
This is now a pattern, not a one-off:
- SK Hynix beat Q2 EPS consensus by +91.76% — and the stock fell 14.71% on the print. Samsung fell 5.91% on its own Q2 record (Investing.com preview table).
- Samsung shares are ~25% below their June record despite four consecutive record quarters (Reuters via Asahi). MU is -15.2% off its 52-week high with RSI 48, despite beating 4/4 quarters and 80.7% GAAP gross margin (/markets/MU/2026/10/12).
- Estimates are rolling over underneath the records: the LSEG SmartEstimate for Samsung was "slashed by 7.7% since the end of August"; 90-day EPS revisions are negative for both Koreans (Samsung -3.65%, SK Hynix -3.80%).
The sell-side has started pre-positioning for the turn: Bernstein expects HBM to be less profitable than conventional DRAM in 2027, prices normalizing in 2028 with industry gross margins easing from ~90% to the high-70s — and cut SK Hynix's target from ₩3.3M to ₩2.7M on HBM4 supply problems. J.P. Morgan has Samsung's HBM share rising to 34% in 2026 (from 20%) while SK Hynix falls to 46% (from 60%) — share rotation inside a shortage.
Even the bulls have changed the basis of the bull case. Gilbert again: "Memory buyers are also committing to multiyear supply agreements, giving Samsung greater visibility over demand in a business where investors have always worried about the next downturn." Translation: the market no longer pays for the shortage — it pays for cycle-dampening contracts (Micron's Contracted Model Meets Its First Guided Margin Step-Down — and a Strike Vote at Its Biggest Manufacturing Hub).
What it means: with price levels at records but momentum decelerating (DRAM/NAND Contract Pricing: The Deceleration Arc Is Now the Consensus — Q2 +60% → Q4 +10–15%, LTA Ceilings Binding, Spot Flat), the marginal buyer is trading the rate of change, not the level. This is what the equity tape does at prior memory peaks — discounting the turn two to three quarters before the P&L shows it. Whether the tape is early or wrong gets tested at the next checkpoints: SK Hynix's Oct 27 print and the 2027 HBM4 contract settlements.
-
An instance of Memory cycles turn on deceleration, never on falling prices. — The tape is discounting the turn from the rate of change — record price levels and consecutive beats no longer move memory equities. ↩︎