Micron fell 8% today to $904.28, its worst day since this correction began, on 54 million shares. The trigger was not an order cut, a downgrade, or a guidance walk-back. It was an IPO on the other side of the world: CXMT, ChangXin Memory Technologies, China's DRAM champion, priced the largest Chinese chip listing in history on July 14. The market read it as the arrival of the China memory flood and sold everything with DRAM exposure. Intel, AMD, and Marvell all fell 5 to 7% in sympathy.
I own a large Micron position, so I spent today pulling the CXMT story apart properly: the prospectus numbers, the capacity claims, the technology gap, and what actually changed for Micron's earnings. Short version: the threat is real, but it is aimed at a market Micron already left, on a timeline that starts around 2028. Meanwhile the numbers that decide this stock did not move at all this week. The chart, though, earned its warning label. Here is the full picture.
The news, precisely
CXMT priced its Shanghai STAR Market IPO at 8.66 yuan per share on July 14, raising $8.55 billion base and up to roughly $9.8 billion with the overallotment, about 10% of the company. That doubles its original target and passes SMIC's 2020 deal as the biggest Chinese semiconductor listing ever. Shares start trading around July 27.
The prospectus is the real headline. First quarter 2026 revenue of 50.8 billion yuan, up 719% year over year. Net profit of 24.76 billion yuan, up 1,688%. First half profit expected above 50 billion yuan, roughly $7.4 billion. This is a company that lost money in both 2023 and 2024. The global memory shortage made it instantly, massively profitable, and now it has a listed currency, state shareholders holding 36% of it, and a national mandate.
Add the surrounding noise, YMTC's NAND share climbing from 8% to 13% in a year, a report that Apple is testing CXMT chips for China-market devices, GigaDevice guiding first-half profit up 1,099%, and you get today's tape. The fear is rational in outline: China did exactly this to solar, EVs, and batteries.
Who CXMT actually is
Here is what the prospectus and the industry data actually describe. CXMT's revenue is 99% commodity DDR and LPDDR, two thirds of it LPDDR4-class mobile memory. Its server DRAM, about a quarter of its mix, sells almost entirely inside China. Overseas revenue: 2.8% of the total. It holds 7.67% of global DRAM revenue per Omdia as of Q4 2025, around 8% per Counterpoint in Q1, up from about 3% a year earlier. That growth is real.
Now hold that against Micron's mix. Micron end-of-lifed consumer DDR4 and LPDDR4 this cycle, deliberately vacating the exact segments where CXMT lives, and rebuilt itself around HBM, high-capacity server DIMMs, and data center SSDs. Data center revenue ran above $25 billion last quarter alone. HBM is sold out through calendar 2027. Sixteen non-cancelable take-or-pay contracts totaling roughly $100 billion through 2030, backed by $22 billion of customer deposits, put price floors under about 40% of revenue. The overlap between the two product lines today is maybe 10 to 20% of Micron's revenue, and it shrinks every quarter.
One more detail from the filing: the IPO proceeds are earmarked for DRAM capacity and R&D. The amount earmarked for HBM is zero.
Wafers are not bits
The scariest stat circulating today is a research claim that CXMT exits 2026 at 350,000 wafers per month against Micron's 385,000, nearly matching the size of the company it is supposedly coming for. The stat counts wafers. Profits come from bits, and the gap between the two is the entire story.
CXMT has no EUV access and no path to it under export controls. Its most advanced node is roughly a generation and a half behind the big three. Its DDR5 die runs about 40% larger than Samsung's equivalent, its yields sit below the 85 to 90% industry-mature standard, and its cost per bit runs more than 30% above Micron, Samsung, and SK Hynix. Estimates of the technology gap range from 2 to 3 years on the friendly end to 5 to 6 years on the skeptical end. Even the analysts who take CXMT most seriously, and some very much do, put it about 4 years behind SK Hynix.
And there is a behavioral tell that the flood is not here: CXMT priced its chips just 5 to 10% below the big three last quarter. A company running 70% operating margins into a shortage is a price taker harvesting the boom, not a dumper crashing the market. The dumping risk is real, but it is a next-downcycle problem, roughly 2028, when supply normalizes and state backing lets CXMT run fabs at a loss. That compresses the depth of the next trough. It does not touch the earnings of this cycle.
HBM is the moat, and China is not close
Micron's earnings live in HBM, so the only China question that matters near-term is the HBM question. CXMT's HBM3 mass-production timeline has already slipped past 2026 per Digitimes. Modeled yields on its 8-high stacks sit near 25%, against a mature standard nearly three times that. Even the aggressive buildout scenarios put China at about 12% of global HBM wafers by 2028, nearly all of it consumed domestically by Huawei and Cambricon rather than exported into Micron's customer base.
Meanwhile the segment Micron actually sells into keeps tightening. TrendForce has 3Q26 server DRAM contract prices up another 13 to 18% quarter over quarter and sees the server shortage extending into 2027. And the single most important datapoint of the week landed July 12, from Korean industry press: 2027 HBM4 contract pricing is tracking from roughly $2 per gigabit in the second half of this year to $4 to $5 or more next year. More than double. When I mapped this trade on July 7, I wrote that 2027 HBM pricing needed to come in "multiples higher" for the estimates to hold. The first hard read came in exactly that hot.
The scorecard: what moved and what didn't
Since July 7, the week's actual data: fiscal 2027 consensus EPS was $149.64 then and is $149.64 today, not a penny of revision through two separate 8% down days. Zero downgrades and zero target cuts; KeyBanc went the other way, raising its target to $1,750 on July 14 after an Asia supply-chain trip, citing a tight memory environment through 2027. Hyperscaler capex guidance for 2026 still totals roughly $700 to 725 billion. NVIDIA's Rubin platform, carrying 288GB of HBM4 per GPU, is confirmed ramping in the second half. No insider sales were filed this week for the first time in a while. Burry has published nothing new on the position.
What did get worse is the tape. SK Hynix's blockbuster Nasdaq debut on July 10 (up 12.8% day one after a $26.5 billion raise, the largest US ADS offering ever) was followed by Korea's worst session in years and a 27% whipsaw, and MU now imports that volatility daily. The chart printed a lower high at $1,035 against the June high of $1,255, and today's $873.63 low undercut the July 7 low of $891.66. Rallies keep coming on fading volume and breakdowns on expanding volume, with put open interest stacked into the July 24 expiry. Today's close sits about 5% above the line I published on July 7: a high-volume weekly close below $860 breaks the 2026 uptrend structure and puts the 100-day average near $706 in play.
Trim or hold
My framework orders it this way: financials first, valuation second, trend as the trigger. The financials did not blink this week, and the first 2027 HBM pricing read strengthened them. The valuation got cheaper: about 6x fiscal 2027 consensus at $904, about 12x fiscal 2026. The market sold a 2028 commodity-DRAM risk against 2026 and 2027 earnings that are contractually locked, and the estimate tape confirmed it by not moving.
So the fundamentals do not support trimming, and I am not trimming on this news. What I am doing is respecting the trend leg. The exit rules I published July 7 are unchanged and none has triggered: a weekly close below $860 on expanding volume, fiscal 2027 revisions turning negative, 2027 HBM pricing landing flat instead of higher, or a hyperscaler capex pause. Two scheduled China catalysts sit inside the next two weeks, Samsung's full Q2 on July 23 and CXMT's Shanghai debut around July 27, and I expect the CXMT listing to produce SKHY-style sympathy volatility. If a gap through $860 toward $706 would force anyone into a panicked exit, that is a position-sizing problem to solve now, calmly, not a thesis problem. Sizing so the stop is survivable is risk management. Selling the headline is not analysis.
The calendar from here: Samsung July 23, MU options expiry July 24, CXMT listing around July 27, SK Hynix's first US-listed earnings in late July, Micron's own report around September 22, 2027 HBM contracts finalizing in Q4, and the December 9 CHIPS unlock, after which management has committed to returning 100% of excess free cash flow.
Disclosure: I am long MU. This is my own research and opinion, not financial advice. Do your own work, especially with a 2-beta trillion-dollar memory stock two weeks from four scheduled catalysts.