The Unwind, Five Weeks Later: Is The Memory Selling Done, Or Do We Keep Rolling Over?

On June 23 I wrote that the tape was not selling AI — it was selling the positioning around AI. Factor unwind, not fundamental repricing. Five weeks later I can grade my own homework, and the grade is mixed in a way that is actually useful.

The de-grossing part was right and it was fast. The part I want to correct in public: I treated "memory" as one thing. It is not. MU and SNDK have spent five weeks proving they are two different trades wearing the same jersey, and that split is the single most important thing on my screens going into August.

Today, oil came back in hard, crypto is holding its shelf and coiling, and the rotation map I drew in June has quietly rebuilt itself with different leadership. Here is the read.

What I got right: the tells. The vol-of-vol blowout, the broken correlation between gold, bonds, and the dollar, the fact that the epicenter was Seoul and not Silicon Valley — all of that pointed at a book being taken down rather than a demand story breaking. Crowded-trade unwinds resolve in days-to-weeks, not quarters, and this one did. MU's fiscal Q3 print the following night confirmed the demand side, and the panic low in the memory complex has not been retested since.

What I got wrong, or at least too coarsely: I lumped the complex together. "The AI basket gets sold, then the good balance sheets get bought back first" was directionally fine, but it hid the dispersion. The winners inside memory have compounded. The second-tier names got a bounce and then went sideways on declining volume, which is not the same thing as a recovery.

The other thing worth saying out loud: the rotation destinations I named — gold, energy, utilities and power, defensives — did the work. Energy in particular did exactly what it was supposed to do as the sponge for growth money that had to go somewhere. That trade is still alive today, and today's crude tape is the proof.

Micron is the clean side of the trade. The June gap-down was liquidation, the earnings print answered the demand question, and price has spent five weeks doing what recovered leadership does: higher lows, shallow pullbacks that get bought inside two sessions, and relative strength against the semi index on down days. My targets from the July 6 and July 18 notes got hit and I said so at the time.

The reason I'm not chasing here is different from the reason I was cautious in June. In June the risk was forced supply. Now the risk is that the easy repricing is behind us and the next leg has to be paid for with actual capex confirmation from the hyperscalers. That is an earnings-cycle question, not a tape question, and I don't get paid to guess at it in advance.

So the honest MU answer to "is the selling over": yes, the positioning-driven selling is over. It ended in June. What replaced it is a normal uptrend with normal risk, and I trade that with size discipline rather than conviction speeches. I want pullbacks into the rising 20-day with volume drying up. I do not want breakout entries into a name that has already round-tripped a panic.

SanDisk is the other half of the story and it is the half that tells you the complex is not healed uniformly. I was long from $1,340 and I booked it on July 21 because the character of the tape changed — the second-leg thesis had played out and the buyers stopped showing up on dips. Nothing in the five weeks since has made me want the position back.

The tell is relative, not absolute. On green days SNDK participates. On red days it gives back more than it made. That is a distribution footprint, not accumulation. When a name stops outperforming its own group on the up days but keeps matching it on the down days, someone is using strength to get out, and I am not interested in being the counterparty to a patient seller.

The structural point: MU sits closer to the part of the memory stack the AI buildout is actually starved for. The further out you go on the periphery of that stack, the more the story becomes "this rallied because memory rallied" — and periphery names are the first to be abandoned when the money gets selective. Five weeks after a factor unwind is exactly when money gets selective.

So: not the same trade, not the same risk, not the same answer. MU's selling is done. The periphery is still working through supply. If SNDK reclaims its July high on expanding volume I will look again — that would invalidate the distribution read cleanly, and I'd rather be paid than right.

First, breadth inside semis. If the group can advance with more than a handful of names participating, the June unwind is fully digested and the complex trends. If every up day is three names carrying the index, we are in a narrowing tape and the next de-grossing event will be uglier than June because the crowding rebuilt into fewer tickers.

Second, the behavior of the losers. In a finished correction, the beaten-up names stop making new relative lows even if they don't rally. In an unfinished one, they keep bleeding quietly while the headline index makes highs. Right now the periphery is bleeding quietly. That is a caution flag, not a sell signal.

Third, volatility structure. The June event was a vol-of-vol story. If VVIX starts creeping again while spot vol stays pinned, dealers are getting nervous ahead of something and I size down before I know why. That single indicator saved me more money in June than any fundamental view I had.

Net: I do not think the memory selling resumes as a complex-wide event. I think it continues as a rolling, name-by-name rotation out of the periphery and into the two or three names with real HBM leverage. That is a stock-picker's tape, not an index tape, and you trade it with a scalpel.

Crude was the cleanest thing on my screens today. Energy caught a real bid, the integrateds led, and the services names — which had been the laggard leg all month — finally participated. That last part is what makes today different from the last three false starts.

The macro backdrop I laid out on July 22 has not changed: the geopolitical premium never fully came out, supply discipline is holding, and institutions remain structurally underweight the sector relative to two decades of history. Add the rotation flow out of crowded growth and you have a group that gets bought on dips by people who are not price-sensitive.

How I'm playing it: long the group, not the headline. Integrateds and the highest-quality producers for the core, services only on confirmation because that leg is beta and it will hand back the move if crude stalls. My invalidation is simple — if crude gives back today's range low and the services names roll first, the bid was tourists and I'm out of the beta leg immediately.

One more thing. Energy strength on a day when growth is not being liquidated is a better signal than energy strength on an unwind day. In June, energy was up because something else was down. Today it was up on its own bid. That is a real rotation, not a hedge.

Bitcoin has refused to break its shelf through the entire unwind and everything that followed. Every attempt to press it lower has been absorbed, and the ranges have been compressing for weeks. Compressing range plus a floor that won't break is the setup that produces a move — and the direction is decided by the break, not by my opinion about it.

I lean up, and I have said why since July 20: the stablecoin flywheel is compounding into every quarter, and the equity expression of that flywheel keeps getting stronger regardless of what spot does day to day. I am still long COIN and my $200-plus target is still the number I'm playing for. Nothing in the last week has changed that thesis; the tape has just been slow about paying it.

The discipline part: coiled markets punish people who front-run the break. I want the resolution with volume, and I want the first pullback after it to hold. If bitcoin loses the shelf instead, the crypto-equity complex will not be immune no matter how good the flywheel story is, and I will treat my COIN risk accordingly rather than argue with the tape.

Five weeks later, here is where the money actually sits. Energy is the cleanest long with a real bid and no crowding problem. Precious metals held the June rotation gains and are basing again, which is usually constructive. Power and grid names remain the AI derivative that keeps working even when the AI names don't. Memory has split into leadership and periphery. Crypto equities are coiled with an asymmetric setup and a defined risk line.

What is crowded now: a narrower group of AI winners than in June. The June unwind cleared out the marginal longs and the money came back into fewer tickers with more conviction. That is a better tape to trade and a worse tape to be complacent in, because the next de-grossing event finds less liquidity underneath.

The bottom line: the selling that started on June 23 is over as a positioning event. It is not over as a rotation. Money is still leaving the periphery and consolidating into leadership, and that process has weeks left in it. I'd rather own the two things with a real bid — energy and the coiled crypto expression — than pay up for the third bounce in a name that stopped outperforming a month ago.

This is my opinion and my read of the tape as of publication on July 29, 2026. It is not investment advice and not a recommendation to buy, sell, or short any security. I am flat SNDK and long COIN as of publication, and positions can change without notice. Levels and theses discussed here can invalidate in a single session. Do your own work and verify independently before risking capital.