Memory stocks stopped trading like different companies in June

SanDisk makes NAND flash. Micron makes DRAM and high-bandwidth memory. Different products, different fabs, different customers, and for most of the time we can measure them, they traded like it — as two companies in the same industry rather than as one asset.

Between 22 June and 31 July 2026 they stopped.

The three windows are not a clean partition — the return from 18 to 22 June falls between the first two rows — so read them as three separate measurements rather than as parts summing to a whole.

Everything below asks what that finding is worth.

What else was happening in those 28 sessions

The Korean market was falling hard, and the falls were led by memory.

By 8 July the KOSPI was more than 20% below the record it set in June. On 13 July SK Hynix fell over 15% in Seoul, its largest single-day decline on record. On 28 July the index dropped more than 11% in a session led by chip stocks, with Samsung down over 11% and SK Hynix down 12%, amid concerns over China’s chipmaking progress and the AI spending boom. On 30 July the government said it planned to put roughly $14bn into its sovereign wealth fund for strategic investments after the rout. On 31 July SK Hynix and Samsung rose more than 20%.

In the same weeks, one leveraged holder of this complex was being taken apart. Situational Awareness LP held SanDisk as its second-largest long position — $724.4m, 18.8% of its long book, at the period end of 31 March 2026. By 30 June, in a filing made public on 14 August, SanDisk was its largest line at $5.67bn and Micron its second at $5.57bn. It told partners it lost about 67% in July, and on 30 July it sold the bulk of its stock portfolio to Citadel. Before 30 July TEXXR had logged four stories naming the fund, two of them on startup rounds it led; it logged seven in the three days from 30 July.

What connects them, and what does not

The honest answer is: the timing and the securities, and not much else that this record can show.

What is established. Two memory names that normally trade as different companies traded as one thing for six weeks. A large levered holder of one of them was liquidating during those weeks. The Korean memory names fell and then rebounded on dates that bracket that liquidation closely.

What is not established, and matters. There is no dated evidence here that Korean retail investors were net sellers on any particular day, still less forced sellers. There is no measure of Korean market leverage, margin-call volume, or forced-sale volume in this corpus. So the tempting sentence — that one wave of forced selling drove both — is not available. Two things happening in the same weeks, in the same securities, is a reason to look; it is not a mechanism.

And the most-quoted Korean figure is not ours. Two numbers circulate: that around 3.6% of the South Korean population received margin calls, and that more than 1.2 million leveraged retail accounts were called. Both appear in our corpus only inside social posts embedded as discussion beneath the Bloomberg and Financial Times pieces. Neither is stated in any article’s own reporting and neither is corroborated by any publication we hold. They are claims made, not facts established.

The candidate explanation, offered as one

There is a well-described mechanism that would produce exactly this pattern, and it is worth naming precisely because it is a hypothesis rather than a finding.

When a price fall meets borrowed money, it can trigger margin calls or collateral shortfalls, which force selling, which drives prices lower, which triggers more calls. Leverage turns a price correction into a balance-sheet spiral, and the severity of the downswing tracks the leverage standing behind the position rather than the size of the move that started it. Ray Dalio’s Big Debt Crises draws the line between a contraction a central bank can end by easing and one that comes from “a financial deleveraging that leads assets… to be sold, causing asset prices to decline, causing equity levels to decline, causing more forced” selling. Roger Lowenstein’s account of 1998 in When Genius Failed gives the condition in one clause: the firm “was losing money faster than it could raise it.”

Correlations converging toward one during a liquidation is the signature of that process, because a forced seller does not sell what it dislikes — it sells what it owns, and what it owns is whatever it can get a bid for. That is the reason the 0.921 is interesting.

But a signature is not a proof. The same convergence would appear if both names simply responded to the same news, such as the China chipmaking concerns the 28 July coverage cites. This record cannot separate those two explanations, and this piece does not claim to.

The same pattern inside one book

One measurement does bear on it, from a place where we can see the positions.

Across the fund’s own March book, its long holdings and the shares its put positions referenced moved together at +0.816 from the filing period end to 30 July, +0.950 in July alone, and +0.972 over the final fortnight. Everything in the book converged as the month went on.

Two cautions on that. It is a correlation between the underlying share prices, so it says the two baskets became one factor; it does not measure whether the puts hedged anything, which would depend on net delta the filing does not report. And the fund is one holder, so this shows convergence inside a portfolio rather than across a market.

The fuller reading of that book is in the ninety days, and the filings are laid out on the fund’s page.

What would settle it

The 13F for 30 June, filed 14 August, showed both memory lines still held, and larger. It cannot show what was sold in July, or when.

The correlation is its own test, available to anyone. From 3 August to 26 October it ran at +0.835 over 59 sessions: below July’s +0.921, above the old +0.707. A stress artifact would fade further; a lasting change would hold.

Sources

Korean market coverage from the TEXXR record: KOSPI down 11%+ led by chip stocks, Samsung −11%+, SK Hynix −12% (Bloomberg, 2026-07-28); SK Hynix’s largest-ever single-day fall (CNBC, 2026-07-13); the KOSPI down 20%+ from its June record (Financial Times, 2026-07-08); retail investors caught in the selloff (Financial Times, 2026-07-30); SK Hynix and Samsung up 20%+ (CNBC, 2026-07-31); South Korea’s roughly $14bn sovereign-fund injection after the rout (Bloomberg, 2026-07-30).

The 3.6%-of-population and 1.2-million-account figures appear only in social posts embedded as discussion beneath articles 1173805 and 1174000 respectively. The fund counts are TEXXR article searches for “Situational Awareness”, logged before 30 July and from 30 July to 1 August, less two hits on other subjects.

Correlations are Pearson coefficients on daily close-to-close returns from our point-in-time archive of as-traded closes, unadjusted for splits, over the windows and session counts stated. The three windows overlap incompletely and are three measurements, not a partition. Fund-book correlations are value-weighted on the reported position values from the 13F-HR for the period ended 2026-03-31 (accession 0002045724-26-000008, filed 2026-05-18), priced 2026-03-31 to 2026-07-30, the priced names covering 93% of the long book and 76% of the put book by value.

Position figures are from the position lines of the 13F-HRs for 31 March and 30 June 2026, CIK 0002045724. A 13F reports long positions only, is filed up to 45 days after the quarter it describes, reports the value of the underlying shares rather than a premium or strike for option lines, and never reports leverage or ordinary short sales.

The 67% is the manager’s unaudited figure, from his 30 July letter to partners as reported by the WSJ; the Citadel sale is from the WSJ’s sources. Mechanism passages are quoted from Ray Dalio, Big Debt Crises, and Roger Lowenstein, When Genius Failed. Not investment advice.