MU Broke $900. China's CXMT Is the Reason, and It's Not Going Away

---

Update (August 4, 2026): Everything This Piece Flagged Has Now Printed

CXMT priced, listed and detonated. The raise came in at $8.55 billion (57.9 billion yuan), the biggest mainland semiconductor IPO on record, and the stock debuted up 466% on Shanghai's STAR Market on July 27, briefly making CXMT the most valuable China-listed company. The full anatomy is in the CXMT explainer.

The market's $840 arrived. MU traded through it in the last week of July, closed July 31 at $823.03 after a 5.9% Friday drop, and printed near $801 in the August 3 premarket before closing that session green at $829.50. The 72%-priced path below resolved yes, inside the month.

The re-entry trigger fired on one condition and failed on the other. SK Hynix reported a record ₩60.54 trillion operating profit that still missed consensus by 5.6%: the whisper was right. Level met, fact failed, so the two-month-calls trade below never activated. The scoring is in the new section at the bottom.

Today, August 4, MU is rebounding: up roughly 6% near $880 at midday (an intraday snapshot, not a close) after SanDisk and SK Hynix unveiled the first High Bandwidth Flash industry standard and the sector bounced off an oversold month. SanDisk reports Wednesday; Micron's own next date is its September 22 earnings.

---

TL;DR

• Updated August 4: CXMT raised $8.55 billion and debuted up 466%. MU hit the market's $840 (July 31 close: $823.03), but the SK Hynix half of the re-entry trigger failed, so the trade never fired. MU trades near $880 midday today, rebounding about 6%. The full marking is at the bottom.
• On July 15, MU fell 8% at the afternoon lows, sliced through $910, tagged the $900 line we'd been flagging since July 13, and closed near $903, down about 7% on the session.
• The trigger: CXMT, China's state-backed DRAM maker, priced an $8.5 billion IPO, the biggest in China this year, to scale production. It's already the world's #4 DRAM producer with revenue up sevenfold in the first half, and Apple is testing its chips for China-market devices.
• This was the first 2026 selloff in MU with a genuine fundamental behind it. Every prior leg was positioning. This one attacks the supply-shortage thesis itself.
• The nuance that decides everything: CXMT threatens commodity DRAM. On HBM it is years behind rather than absent (corrected August 4, details below), and HBM is why you owned MU. Levels and plan below.

Meet the Problem

The resume CXMT brought to market. The IPO raised $8.55 billion and the stock debuted up 466%. This is what a real competitor looks like.

The entire memory supercycle rests on one sentence: demand is exploding and supply can't respond until 2028. CXMT's IPO announcement is China disputing the second half of that sentence with $8.5 billion and a state mandate. A company growing revenue sevenfold in six months, already fourth in the world, raising the country's biggest IPO specifically to add DRAM capacity, is not a rumor or an analyst note. It's scheduled, funded supply.

And the Apple detail is the one that stings. Apple testing CXMT chips for China devices means the quality gap is closing fast enough that the world's most demanding customer is willing to look. Commodity DRAM pricing power erodes from exactly this direction.

What CXMT Threatens vs. What It Can't Touch

Be precise here, because the 8% haircut priced both and only one is real today.

Threatened: commodity DRAM. The stuff in phones, PCs and standard servers. This is where CXMT's scale lands first, where Chinese state subsidies distort pricing fastest, and where Micron's margin recovery came from in 2025. Real risk, arriving over quarters.

Not threatened yet: HBM. High-bandwidth memory, the AI-server product that's sold out into 2027, requires packaging and yield capabilities CXMT doesn't have at scale. Nobody credible puts Chinese HBM4 at scale inside two years. Micron's AI story, the reason the stock 5x'd, runs through HBM, and today's news doesn't touch it. Even the bears' own coverage concedes the HBM story stays intact while commodity pricing gets contested.

(Correction, August 4: an earlier version of this paragraph said CXMT "cannot touch" HBM. That was too strong. CXMT has HBM3 samples out, reportedly with Huawei for evaluation, and is targeting mass production. What protects the margin pool is the gap, not a prohibition: fewer than 2% of its roughly 265,000 monthly wafer starts run HBM, at low yield, three to four years behind Samsung and SK Hynix's HBM4 roadmaps. "Years behind at 2% of its own capacity" is the accurate sentence, and unlike "cannot," it stays true.)

So the repricing question: how much of MU's multiple was commodity-DRAM pricing power versus HBM growth? Today's 8% says the market thinks a meaningful chunk was the former. That's fair. What's not fair is extending the CXMT threat to the sold-out HBM book, and that mispricing is where the trade eventually lives.

The Levels, Updated

• $900 broke intraday. The line we set on July 13 didn't hold clean, and there's no point pretending otherwise. A close meaningfully below it confirms the breakdown.
• $840 is the market's next number: prediction markets price a 72% chance MU touches it this month. Respect that. It's also roughly where the stock's HBM-only story finds valuation support.
• Reclaiming $940-950 on volume would mean the CXMT scare got absorbed the way every positioning scare this year was. Below $900, the burden of proof flipped to the bulls.

Marked August 4: the map worked. $900 never got reclaimed, $840 broke in the last week of July, and the tape ran to a $823.03 close on July 31 and about $801 in the August 3 premarket before turning. Today's rebound near $880 is happening exactly inside the $840-940 no-man's-land this section said not to chase. The current session-by-session picture lives in the August 3 breakdown.

The Options Angle

• The put-selling trade is suspended on MU. We've sold fear on every dip this month because the dips were flows. This dip has a fundamental attached, and selling puts into a live supply threat with a 72%-priced path to $840 is picking up dimes in front of a bulldozer that just announced its route.
• The trade with edge: own the HBM/commodity split directly. MU puts (or staying flat) against long SNDK or long SKHY-on-confirmation separates the CXMT-exposed commodity story from NAND and HBM stories CXMT doesn't touch. Today the market sold them all together; it won't forever.
• Re-entry trigger, stated now: $840 with the SK Hynix print in hand. If MU tags the market's number and SK Hynix's earnings (the falsification test from the thesis check) come in intact, that's the spot where the HBM story is on sale because of a commodity-DRAM scare. Two-month calls there, not here.

Trade log (added August 4; this piece predates the log rule)

#	Stance	Structure	Strikes and expiry	Cost or credit	Spot at writing	Implied move	Breakeven
1	Pass	Selling MU puts (suspended)	Any near-dated expiry	None logged	~$903, close Jul 15	Not sourced	Fails on a continued slide
2	Pair	MU puts against long SNDK	Nothing logged	Nothing logged	~$903, close Jul 15	Not sourced	Ungradeable
3	Long, conditional	Two-month MU calls at $840	Trigger: $840 tag plus an intact SK Hynix print	Never activated	~$903, close Jul 15	n/a	Condition failed, no fill

How This Aged, Marked August 4

Score the calls against three weeks of tape, losses and lucky escapes first.

The put-selling suspension was right, twice. MU went from a $903 close to $823.03 on July 31, a further 8.8% down, through a second sector-wide dump on July 28 and a 5.9% Friday drop. Any near-dated short put written into the July 15 candle got run over. This is the row that paid, and it paid by not being a position.

The $840 call resolved yes. Prediction markets priced a 72% chance MU touches $840 inside the month; it traded through the level in the last week of July and closed the month well below it. Respecting a level the market explicitly named remains the most repeatable habit on this site.

The re-entry trigger did its job by not firing. The design was two conditions: the $840 level plus an intact SK Hynix print. The level came, but SK Hynix reported a record quarter that still missed consensus by 5.6%, so the trade stayed untaken. What that cost and saved, honestly: staying out skipped a knife that kept falling to roughly $801 in the August 3 premarket, and it also skipped the rebound to about $880 at midday today. A two-month call struck at the $840 tag would be modestly ahead on today's snapshot. We grade a trigger on whether its logic held, not on the branch we couldn't have known: it held. The commodity-DRAM leg the miss confirmed is the same leg the thesis check had already amputated.

The pair trade is ungradeable, and that's on us. Row 2 shipped with no strikes, no expiries and no prices, before the trade-log rule existed. Directionally the split thesis behaved: the NAND name held up better than the commodity-exposed one, and SanDisk trades near $1,393 today, up about 8%, into Wednesday's earnings. But a recommendation without an entry has no return, so it gets no credit.

Sooo... Dead Money or Discount (Again)?

Neither, yet. This is the first MU selloff of 2026 that earned its red candle, and it deserves a real repricing of the commodity half of the business. It does not deserve the death of the HBM story, which remains sold out regardless of what CXMT builds. Let the stock find the market's $840, let SK Hynix report, and buy the AI-memory story back at a commodity-memory discount. Patience is finally the position.

Marked August 4: the stock found $840 and kept going, SK Hynix reported and missed, and the discount deepened before today's bounce. The patience position is still the position: MU has no dated catalyst until September 22 earnings, the CXMT supply story is now funded with $8.55 billion of real money, and the argument that decides everything is unchanged, which is how much of the multiple was commodity DRAM and how much was HBM. Today's High Bandwidth Flash headline is the market starting to price that split properly, which is all this piece ever asked it to do.