THE ANATOMY OF AN AI & MEMORY SELL-OFF

METHODOLOGY · READ THIS FIRST

Three source grades appear throughout this article.

FV means a primary filing or company disclosure.

KR means Korean-language reporting of Korea Exchange, KOFIA or regulator data, read in the original and translated.

T2 means a bank note, brokerage research, survey or market-data vendor, read secondhand.

Company numbers here are filing-grade. Most positioning and flow evidence cannot be, because prime brokerage data and fund manager surveys go to clients only, and Korea Exchange statistics sit behind a portal I could not access. Where a claim rests on secondary reporting, the tag says so.

SECTION 1 · THE EVENT

WHAT HAPPENED

On 22 June 2026, two funds made their 52-week highs on the same day., which tracks the semiconductor complex, closed at 655.01., which tracks memory, closed at 80.72. Both finished that session higher, by 2.4% and 5.2%. T2

The next morning they fell 7.9% and 14.3%. By 17 July, SOXX was down 20.3% and DRAM down 34.7%. T2

Across the same weeks the S&P 500 barely moved. The damage stayed inside semiconductors, memory and the crowded edges of the AI trade and other high-beta stocks.

Although both funds hold the same theme, one fell nearly twice as far as the other. Anything that explains the decline has to explain that gap, and most of what has been written does not try.

The picture becomes even more puzzling if you look at the underlying assets within the ETFs, because the earnings are still delivering and then some:

Samsung reported quarterly operating profit of ₩89.4 trillion. A year earlier it earned ₩4.68 trillion and the stock still closed down 6.9%. FV T2

Nine days later TSMC filed a fifth consecutive record quarter with a 67.7% gross margin. Its shares rose 1.23%, while the semiconductor sector fell 4.46% the same session. FV T2

Something other than earnings was setting these prices. Two different things, in fact, but before examining what those were, it’s worth understanding the mechanics that de…