When Positioning, Not Fundamentals, Sets the Price
A cross-cycle read on why oversold extremes in memory and AI names have repeatedly marked local bottoms, and why today's tape fits the pattern.
Some sessions are not about earnings. They are about flows.
The move originated overnight in Asia. Samsung Electronics fell close to 7% and dragged the Kospi down nearly 5%, and it did so despite a sharp jump in second-quarter profit.
A reminder that in a de-risking tape the print matters far less than the positioning behind it.
Compounding the tone, a Reuters report that DeepSeek is developing its own AI accelerator revived the familiar fear of reduced dependence on the incumbent silicon supply chain, Nvidia and Samsung included.
By the time the wave reached the U.S., the entire semiconductor complex was on the back foot.
As of mid-session, the sequence is textbook. The VanEck Semiconductor ETF is off -5.4%. Nvidia is relatively defended at -1.2%; the higher-beta complex is not, with Broadcom -2.5%, AMD -7.4%, KLA -8.3% and Marvell -9.1%.
The catalyst is real but marginal. The mechanism is de-grossing: a crowded, high-conviction trade being reduced under duress, not repriced on fundamentals.
Sentiment is the variable; the process is the constant
The distinction matters because it defines where the edge sits.
When a selloff is driven by fundamental reassessment, whether a demand air-pocket, a pricing collapse or a structural share loss, the correct response is to cut.
When it is driven by positioning, whether forced de-risking, headline-triggered stop-outs or cross-asset contagion out of a single Asian session, the correct response is the opposite: lean in where the crowd is capitulating.
Distinguishing the two in real time is the hard part, and it is where discipline beats narrative.
For eighteen months, the AI trade has repeated the same cycle: an exogenous shock out of Asia or a China-competition headline, a sharp multi-day drawdown across the semiconductor complex, and a recovery that begins precisely when sentiment indicators reach their extremes.
The tape does not remember the headlines. It remembers the levels.
There is one such level being tested right now, a widely watched momentum extreme that, on a set of core AI and memory names, has repeatedly coincided with local bottoms across this cycle.
Two names in particular, which I identify below, are printing that condition today with unusual clarity.