The AI Memory Trade Just Had Its Black Tuesday. Speculators Were Already Net Short the Nasdaq at a One Year Extreme.

By Kresmion Research, June 24, 2026

The AI memory trade had its worst day in months, and heading in, large speculators were already net short the Nasdaq at the bottom of their one year range. On Tuesday June 23, South Korea's KOSPI fell 9.99 percent and tripped a circuit breaker, Samsung and SK Hynix each dropped more than 12 percent, and in the US the chip complex went with it: Micron fell about 13 percent, Qualcomm about 8.6 percent, Nvidia about 4.1 percent. Micron was up roughly 250 percent year to date going into that session. This is the part the price chart does not show: who was leaning out before it broke, and who got caught leaning in.

This is not a recommendation and it is not a forecast. It is a positioning post-mortem, built from the data that does not make the headlines: futures positioning, insider filings, and options skew, set next to the tape.

Key takeaways

What broke

The memory chip trade had gone close to vertical. SanDisk was up roughly eightfold on the year and jumped 11 percent the week prior after Apple flagged what it called unavoidable memory price hikes (247 Wall St). The driver was real: Gartner has DRAM prices up about 125 percent and NAND up about 234 percent in 2026, with the largest makers sold out of much of this year's output on fixed price multi year contracts (Stocks Down Under). The KOSPI, where the memory giants live, had roughly doubled on the year going into the selloff.

Then June 23 happened. The KOSPI fell 9.99 percent to 8,203.84 and triggered a circuit breaker, with Samsung off 12.3 percent and SK Hynix off 12.5 percent. Foreign investors sold about 5.79 trillion won, roughly 3.8 billion dollars, of Korean shares in a single day (KED Global). The US chip complex followed. One detail matters for honesty: Micron's 13 percent drop was not an earnings miss. Micron reported the next day, June 24, after the close. The Tuesday move was contagion and a rates shock, not a number (Motley Fool).

What Kresmion's positioning data showed going in

Here is the part you cannot see on a candlestick. Kresmion tracks the CFTC Commitments of Traders report across thirteen futures markets and normalizes each to its own trailing year. As of the report dated June 16, one week before the crack, large speculators in Nasdaq 100 futures were net short 8,908 contracts. Against open interest that is about 2.6 percent net short, and it sat in the 5th percentile of the past 59 weekly reports, a z score of minus 1.88. This was not a one week blip. The same group had been net short in 7 of the prior 12 weeks, with three separate extreme short flags on April 28, May 12, and June 2. Late last November this position was net long more than 57,000 contracts. By late April it had crossed to outright short. The fast money spent two months quietly stepping off the Nasdaq while the index made new highs.

That is the cleanest version of the story: the futures crowd was already positioned for trouble in the index that cracked, and it had been for weeks. CFTC positioning is published Fridays for the prior Tuesday, so this read is dated June 16 and is eight days stale by the time of the selloff. It does not capture the crash itself. What it captures is the setup.

The part that complicates the smart money story

If the article stopped there it would be too clean, and the data does not let it. Those same speculators were also net short volatility. As of June 16 they held a net short of 66,739 VIX futures contracts, about 16 percent of open interest, a position that pays out as long as nothing breaks. So the picture is not a tidy one of foresight. It is a crowd that was net short the index but also short the insurance against a shock, positioned for a slow grind lower rather than a violent one. When the shock arrived, the short volatility leg was on the wrong side of it.

The broad equity read was milder. S&P 500 speculators were net short about 7.5 percent of open interest, but at a z score of only minus 0.21, nowhere near an extreme. The conviction, such as it was, sat specifically in the Nasdaq book and in being short calm. That distinction is the difference between a hedge and a call, and the honest answer is that it was probably some of both.

The insiders were trimming too

The futures tape was not the only place capital was leaning out. On Kresmion's insider board on June 24, of 85 active clusters, 57 were net selling and 28 were net buying, with 253 million dollars of insider sells against just 29 million of buys, a ratio of nearly nine to one by value. Two of the four strongest sell clusters were chip names. Applied Materials showed five insiders selling 35.9 million dollars including the CEO, ranked third. Nvidia showed five insiders selling 40.3 million dollars on a single day, June 17, six trading days before the selloff, ranked fourth, and that group included both CEO Jensen Huang and CFO Colette Kress.

The honest caveat is large and it belongs here. Five executives selling on the exact same day is the signature of a pre scheduled 10b5-1 plan, not a panic. Kresmion's data carries no plan flag, but coordinated same day C suite sales at a company like Nvidia are almost always programmatic. And 40 million dollars is a rounding error against a multi trillion dollar market cap. The dollar figure is not the signal. The pattern is: insiders across the AI hardware complex were net sellers into the run, not buyers.

The cleaner real time tell came from options. On June 23, Nvidia's 180 strike put carried a higher implied volatility than its 220 strike call, a skew toward the puts of about 2.6 volatility points, and the 200 strike put was the single most crowded line on the board at 107,000 contracts of open interest. Unlike a scheduled insider sale, that is live money paying up for put protection the day the tape rolled over.

The counter case: rebalancing, not a top

The strongest argument against reading any of this as the start of an unwind is that the memory shortage is structural, not a story. The top three makers have sold out much of 2026 production under fixed price multi year deals, the shortage is forecast to run into 2027 and 2028 because fabs take years to build, and the Korea Exchange called the selloff a rebalancing rather than a verdict on the country (CNBC). The KOSPI bounced 3.26 percent the very next session, led by the same chips. A position that is short an index backed by contracted revenue can be right on the tape and wrong on the thesis.

There is also a leverage explanation that has nothing to do with foresight. Korean retail margin balances were at records and single stock leveraged chip ETFs had grown large enough that regulators issued warnings, so the move may have been a forced deleveraging of a crowded, heavily margined book rather than informed selling, after regulators warned over the single stock leveraged chip ETFs that had ballooned past nine billion dollars (Motley Fool). And the macro trigger was its own event: the June 17 FOMC delivered a dot plot that flipped to signal a rate hike, with nine of eighteen officials now penciling at least one and the median rate projection rising to 3.8 percent (CNBC). Net short positioning did not cause that. It just happened to be facing the right way when it hit.

What would change the read

Two observable events settle which story this was, and neither is a prediction. First, May PCE lands Thursday June 25. April headline ran 3.8 percent and core 3.3 percent, both above the Fed's freshly raised 3.6 percent forecast (Morningstar). A hot print hardens the rate hike fear that lit the fuse. A soft one removes it. Second, Friday's CFTC report, dated June 23, will show whether speculators covered that Nasdaq short into the drop, which would mark it as a completed hedge, or pressed it, which would say they think there is more to come. If the chip names keep bouncing the way they did Wednesday while positioning stays short, the futures crowd was early and wrong. If the bounce fails while insiders keep selling, the setup was telling you something. Kresmion's regime model, for its part, just printed its lowest reading in three months on six straight down sessions, and its signal board ran net cautious by 17 to 9. That is a model leaning risk off, not a prediction that it is correct.

Frequently asked questions

What does it mean that speculators were net short the Nasdaq?

In the CFTC Commitments of Traders report, the non commercial category covers large speculative traders such as hedge funds and managed futures. As of June 16 they held more short Nasdaq 100 futures than long, a net short equal to about 2.6 percent of open interest. That reading sat in the 5th percentile of the past 59 weekly reports, meaning they had rarely been more short in the past year. It is a measure of how the fast money was positioned, not a forecast.

Did the smart money call the AI selloff?

Partly, and the honest answer carries a catch. Speculators were net short the Nasdaq for two months before the June 23 drop, which is the right direction. But they were simultaneously net short volatility, which is a position that loses money in exactly the kind of shock that hit. So they were leaning short on price and complacent on risk at the same time. That is closer to a crowded hedge than a clean prediction.

Was Micron's 13 percent drop caused by earnings?

No. Micron reported earnings the following day, June 24, after the close. The 13 percent fall on June 23 was driven by the Korean market selloff, the broader chip de rating, and a rate hike scare, not by a specific Micron number. The stock was up roughly 250 percent year to date going into the move, which left little room for disappointment.

Is the AI memory trade rolling over or just resting?

The data cuts both ways and the article does not pretend to resolve it. The case for an unwind is forced deleveraging of a crowded, heavily margined position with a hawkish Fed removing the tailwind. The case for resilience is that the memory shortage is structural, with makers sold out of 2026 output under fixed price contracts, which is why the same chips that crashed Tuesday led the bounce Wednesday. The cleanest things to watch are Thursday's PCE print and whether the bounce holds.

Sources

• Kresmion proprietary data, as of June 24, 2026: CFTC Commitments of Traders positioning across 13 futures markets (cot_reports), insider trading clusters (insider_clusters), equity options skew and open interest (options_chain_snapshots), the cross asset regime model (macro_regime_history), and the daily signal board (asset_signals).
• U.S. Commodity Futures Trading Commission, Commitments of Traders. https://www.cftc.gov/MarketReports/CommitmentsofTraders/index.htm
• KED Global, Korean stocks crash on tech selloff, June 23, 2026. https://www.kedglobal.com/korean-stock-market/newsView/ked202606230010
• Motley Fool, why Micron stock suddenly crashed, June 23, 2026. https://www.fool.com/investing/2026/06/23/why-micron-stock-suddenly-crashed/
• 247 Wall St, SanDisk jumps after Apple flags memory price hikes, June 18, 2026. https://247wallst.com/investing/2026/06/18/sandisk-jumps-11-western-digital-rises-7-after-apple-flags-unavoidable-memory-price-hikes/
• Stocks Down Under, DRAM and NAND memory shortage and AI demand. https://stocksdownunder.com/dram-memflation-ai-memory-shortage/
• CNBC, South Korea KOSPI selloff and foreign investors, June 11, 2026. https://www.cnbc.com/amp/2026/06/11/south-korea-kospi-selloff-foreign-investors-krx.html
• CNBC, Federal Reserve interest rate decision, June 17, 2026. https://www.cnbc.com/2026/06/17/fed-interest-rate-decision-june-2026.html
• Morningstar, May PCE preview, inflation seen firming. https://www.morningstar.com/economy/may-pce-expected-show-rising-inflation

Related Kresmion reading: Research Notes archive and the Learn library.