China's Largest Memory Maker Earned a 49% Net Margin Last Quarter. It Is Not Deflating the Memory Cycle — It Is Being Paid By It.

The question under every memory-cycle argument this year is whether China ends it. It is the largest single input into what Micron ($MU) is worth, it is the reason our own bear case exists, and until two months ago it could not be answered with a number, because the company at the centre of it published none.

That changed on July 27, 2026, when CXMT — ChangXin Memory Technologies, China's largest DRAM maker — listed on Shanghai's STAR Market. An IPO requires a prospectus. The prospectus has financials in it.

They do not say what either side of this argument expected.

What the prospectus discloses

The table at the top of this page is the whole disclosure. Three lines carry the argument.

2025: CNY 61.80 billion of revenue, up 155.6%, and CNY 1.88 billion of net profit. That is a 3.04% net margin, in the year the memory market went vertical, from a company that lost CNY 7.14 billion the year before on CNY 24.18 billion of revenue.

The March 2026 quarter: CNY 50.80 billion of revenue, up 719.1%, and CNY 24.76 billion of net profit. A 48.74% net margin. One quarter did 82% of the revenue of the entire prior year.

Guidance for the first half of 2026: CNY 50–57 billion of net profit.

Converted at CNY 6.74 per US dollar — the rate on August 19, and one rate applied throughout, because press coverage of the same yuan figures has quoted anything between $8.6bn and $9.1bn for 2025 revenue depending on which rate the desk used — that is $9.17 billion of 2025 revenue and $7.54 billion in the March quarter.

The comparison everyone reaches for, and why it misleads

Micron's most recent filed quarter, ended 2026-05-28, was $41,456 million of revenue. So:

• CXMT's entire 2025 is 22.1% of that single Micron quarter.
• CXMT's March 2026 quarter is 18.2% of it.

The natural conclusion is that CXMT is a rounding error and China is not a risk yet. That conclusion is available, it is arithmetically correct, and it is the wrong read — for the same reason a growth investor never compares levels without comparing rates.

CXMT is growing at roughly twice Micron's rate inside the same price spike. Two notes on that comparison before anyone quotes it: the two quarters are not the same three months, and the smaller base does most of the work in any growth ratio. But it is the direction that matters, and the direction is that the gap is closing rather than widening.

The finding: China is inside the bubble, not attacking it

Here is the part that reframes the whole question.

A price war looks like volume without margin. CXMT's 2025 looks exactly like that — 155.6% revenue growth converted into a 3.04% net margin, which is what selling bits near cost produces.

Its March quarter does not look like that at all. A 48.74% net margin is not the behaviour of a company flooding a market to take share. It is the behaviour of a company taking the same prices everyone else is taking.

Set the profit split side by side — that is the second chart on this page. In the March quarter CXMT kept $3.67 billion of profit on $7.54 billion of revenue. In the quarter to May, Micron earned roughly $28.1 billion of net income on $41.5 billion, a 67.7% net margin. (That Micron figure is ours: diluted EPS of $24.67 times 1.138 billion diluted shares, not a line in the filing.)

Two companies, two countries, one price. China is currently a beneficiary of the memory cycle, not a threat to it. The Chinese entrant that was supposed to break DRAM pricing has spent the last two quarters collecting the same windfall as the incumbents.

That is a genuinely bullish datapoint for the near term, and it should be said plainly by a site whose model carries a bear case built on the opposite.

There is a second reading of it that is not bullish at all, and it is the one worth holding: a 49% net margin is what funds the capacity that ends the cycle. CXMT was loss-making as recently as 2024. It is now generating the cash to build fabs without needing anyone's permission or capital — and it raised more on top of that.

Where the risk actually sits: wafers, and the absence of HBM

The prospectus is as interesting for what it does not contain.

Its named project spending goes to wafer lines and process upgrades for the DRAM CXMT already makes. There is no HBM project in it. High-bandwidth memory is the segment where Micron's margin lives, and the company that is supposedly about to end Micron's cycle has not earmarked prospectus money for it. Press reporting since has pointed to HBM packaging plans in Shanghai for late 2026; the prospectus, which is the document with legal consequences attached, is silent. Those two things are not compatible and we are not going to pretend to resolve them.

Meanwhile the capacity picture is the one thing here that is not disclosed anywhere. What exists is research:

• Reuters puts CXMT's three fabs at about 300,000 wafers a month today.
• Citrini Research projects about 350,000 a month by the end of 2026, against roughly 375,000 for Micron.
• Announced projects would take CXMT past 600,000.

Every one of those is an estimate by someone outside both companies. Neither Micron nor CXMT discloses monthly DRAM wafer starts, and this is exactly where a piece like this is supposed to stop rather than compute. So a note on the number that has been circulating: divide CXMT's revenue by an estimated wafer count and you get somewhere between $7,200 and $10,500 per wafer for the March quarter, against $36,900 to $46,100 for Micron on the same treatment — a gap of roughly four to six times. On 2025's revenue the same arithmetic gave a gap nearer fifteen to twenty. Both ends of that ratio rest on capacity estimates we cannot check and on Micron revenue that includes NAND, so the level is not a number anyone should quote. The change in it comes entirely from disclosed revenue, and the change is the point: whatever the gap was, it closed by roughly two thirds in a quarter.

The other half of the bubble question

CXMT priced its IPO at CNY 8.66 and closed its first session at about CNY 49, up roughly 466%, at a market capitalisation near CNY 3.3 trillion — about $490 billion, which made it the most valuable company listed in China, ahead of ICBC. The raise itself is the one figure our sources disagree on: the prospectus sought CNY 29.5bn, and coverage of the completed deal reports CNY 57.92bn (about $8.6 billion).

Micron, at $971.66 as of the August 18 close — the price basis our model carries — and 1.138 billion diluted shares, is worth about $1.106 trillion.

So a company with 18% of Micron's quarterly revenue and a 7.67% share of global DRAM carries 44% of Micron's market value. Whatever is being priced into memory equities, it is not being priced only in dollars, and "is China a risk to the memory bubble" has an answer nobody asks for: China is now a large part of it.

What this does to our Micron model

Our Micron model carries four cases. Computed from the file as it stands, at a $971.66 price basis:

The Bear case is the China case. Its note says so: "the glut arrives early… CXMT keeps taking the mainstream DRAM the incumbents walked away from, and the uncontracted book reprices hard." And its entire evidence base is two X posts — one paraphrasing a 2028-glut forecast, one asserting that CXMT runs "around 300,000 wafers a month, close to MU's total DRAM capacity."

Two things follow, and they point in opposite directions.

First: the tweet was right. Reuters and Citrini both land in the same place on capacity, independently and after that post. It is unusual to be able to say that, and it should be said. It also does not make a tweet an acceptable source for the most consequential assumption in a published model.

Second: the prospectus contradicts the mechanism, not the magnitude. The bear case assumes CXMT is "taking the mainstream DRAM the incumbents walked away from" — the price-war reading. A 48.74% net margin says that is not what happened in the March quarter.

Now the sensitivity, which is the part that decides whether any of this matters. Holding the model's own base assumptions and changing one thing at a time:

The China assumption is worth 1.76 times what the exit multiple is worth. In a model where the exit multiple is normally the loudest input — our own rationale calls it the thing that "decides almost the whole answer" — the operating consequences of a Chinese glut matter more. That is why sourcing it from a tweet was the weakest point on this site, and why a prospectus is worth an article.

Does the model change today? No. The prospectus does not tell us when CXMT's wafers become bits at prices that break Micron's uncontracted book; it tells us that has not happened yet and that CXMT is now vastly better funded to make it happen. Both readings survive it. Re-sourcing the bear case from X posts to the prospectus is a define-thicker-model change and a separate decision from this piece — and it is now a possible one, which it was not last week.

One thing the bear case gets structurally right, and the prospectus supports: it is a margin case, not a revenue case. CXMT competes in conventional DDR4 and DDR5 — DDR5 was 31.9% of its 2025 revenue, up from 13.3% — and not, on the prospectus's own evidence, in HBM. Whether Micron's take-or-pay floors cover the conventional segments where that competition lands, or only the high-bandwidth lines, is the question that decides how much of Micron's revenue is exposed, and neither the 10-Q nor the prospectus answers it.

What to watch

1. CXMT's H1 2026 result against the CNY 50–57bn net profit guide. A beat means Chinese DRAM is still being paid cycle prices; a miss on flat volumes means the price war has started at home first.
2. Any HBM project appearing in a CXMT filing or capital raise. That converts the story from a pricing threat to a displacement one, and it is the single disclosure that would most change Micron's model.
3. Whether CXMT's revenue per wafer keeps rising. Rising revenue on flat wafers is CXMT moving up the stack. Flat revenue on rising wafers is bits flooding the conventional market — far more bearish, and the reading almost nobody would call bearish at the time.
4. Micron's next 10-Q language on average selling prices. The last one gave a low-260% ASP increase against a low-20% bit increase; that ratio is the exposure.
5. Whether anyone gets a wafer-capacity figure from a company rather than a research desk. Everything anyone believes about Chinese DRAM capacity currently traces to a small number of estimates.

CXMT's financials — 2024 revenue of CNY 24.18bn and a CNY 7.14bn net loss, 2025 revenue of CNY 61.80bn (+155.6%) and net profit of CNY 1.88bn, Q1 2026 revenue of CNY 50.80bn (+719.1%) and net profit of CNY 24.76bn, H1 2026 net profit guidance of CNY 50-57bn, DDR5 at 31.9% of 2025 revenue against 13.3% in 2024, a 41.02% 2025 gross margin, a 7.67% Q4 2025 global DRAM share against 91% for the top three, and the absence of any HBM project in the prospectus — are from CXMT's Shanghai STAR Market IPO prospectus as reported by trade and financial press; CXMT files in China and we have read the prospectus through that coverage rather than the document itself, which is a weaker provenance than this site normally accepts and is the reason every CXMT figure here is labelled. The July 27, 2026 debut, the CNY 8.66 offer price, the ~CNY 49 close, the ~466% gain and the ~CNY 3.3 trillion market capitalisation are from press coverage of the listing; the raise is reported as CNY 29.5bn in the prospectus and CNY 57.92bn in coverage of the completed deal, and we have not reconciled the two. All dollar equivalents are ours at CNY 6.74 per USD, the rate on August 19, 2026. Micron's revenue of $41,456M for the quarter ended 2026-05-28, the prior-year $9,301M, diluted EPS of $24.67 and the 84.56% gross margin are from our stored series and our Q2 coverage; the $28.1bn net income and 67.7% net margin are ours, derived as EPS times 1.138 billion diluted shares. The 343% DRAM revenue increase on a low-260% ASP increase and a low-20% bit increase is quoted from Micron's 10-Q for that quarter. Capacity figures — CXMT at ~300,000 wafers a month today (Reuters), ~350,000 by end-2026 against Micron's ~375,000 (Citrini Research), and 600,000+ from announced projects — are third-party estimates, not company disclosures, and neither company publishes monthly DRAM wafer starts; the revenue-per-wafer bands built on them are ours and should not be quoted as levels. Micron's $971.66 price is the basis carried in our model as of the August 18 close, not a live quote; the ~$1.106 trillion market capitalisation uses 1.138 billion diluted shares per house convention. All four Micron fair values and the sensitivity table are computed from data/models/mu.json as it stands today; the model's own bear-case assumptions and its two X-post sources are quoted from that file.