On 27 July 2026 ChangXin Memory Technologies listed in Shanghai and closed 465.82 per cent above its offer price, briefly the most valuable listed company in China. Korean memory shares fell hard through the same month, and the listing became the explanation. We went through the prospectus, the two Korean quarterly filings and the sequence of trading days, and the explanation does not survive the dates.
Last verified: 4 August 2026. Built from CXMT’s Shanghai listing prospectus as reported by Reuters, SK hynix and Samsung Electronics second-quarter 2026 results releases, Counterpoint Research and TrendForce market data, the United States House Select Committee on the Chinese Communist Party, and the Department of Defense 1260H list. Prices and share counts move; the technical and regulatory record is the part we expect to hold.
Editor’s opinion
The gap is wide, and a listing did not close it
Our view is that CXMT remains a long way behind Samsung Electronics and SK hynix on the things that decide memory profits, and that nothing in the July record changes the distance. On process, CXMT’s volume node is where the Korean two were in 2019. On high bandwidth memory, the product that pays for this cycle, CXMT is targeting first HBM3 output while both Korean firms are shipping HBM4. That is not a gap measured in quarters.
We also think the market attributed the wrong cause. Korean memory shares had their worst day of the month on 13 July, two weeks before CXMT listed, and the trigger was profit-taking after SK hynix’s Nasdaq debut. A Morgan Stanley call to cut memory exposure landed on 6 July. The 28 July fall followed record results that missed forecasts. CXMT is in that story, but it is one input among several, and it is not the largest.
What we will not do is pretend the closing distance is imaginary. CXMT went from roughly 3 per cent of the DRAM market to roughly 8 per cent in four quarters, and its first-quarter revenue rose more than sevenfold. Those are real numbers from an audited document. Our reading is that most of that came from commodity DRAM pricing roughly doubling in a single quarter, on a product line that is almost entirely commodity — which is a statement about the cycle and about CXMT’s exposure to it, not about closing a technology gap.
Andrew Lee. This section argues a view rather than establishing a fact. It is not a forecast and not a recommendation, and the material it draws on is listed in the sources below.
The listing that set the number
CXMT sold shares at 8.66 yuan and closed its first session at 49.00 yuan, a rise of 465.82 per cent. It traded as high as 55.03 yuan intraday. The close valued the company at roughly 3.28 trillion yuan, which put it above Industrial and Commercial Bank of China and made it, for that afternoon, the largest listed company in the country by market value.
The raise was about 57.9 billion yuan, near 8.6 billion dollars, from roughly 6.7 billion shares representing about 10 per cent of the free float after issuance. That is one of the largest Asian offerings of the year. The prospectus says the money goes to production lines and technology upgrades.
One detail is worth holding onto, because it tells you how the listing was received at home rather than abroad. Before the debut, the China Securities Regulatory Commission held a series of meetings with listed companies, brokerages, fund managers and academics, against concern that CXMT would pull money out of other Chinese technology stocks. The domestic worry was about crowding out, not about Korea.
A first-day close of 465.82 per cent above the offer price is a statement about how the offer was priced and how Chinese retail demand behaves in a hot listing. It is not a measurement of the company’s technical position, and we do not treat it as one anywhere in this piece.
What the prospectus actually shows
The listing produced something that did not exist before: an audited filing. Until 27 July, most published figures on CXMT were estimates from research houses. Now there is a document, and the document is striking.
First-quarter 2026 revenue was about 50.8 billion yuan, up roughly 719 per cent on the same quarter a year earlier. Profit attributable to shareholders was about 24.8 billion yuan, against a loss of about 1.56 billion yuan in the year-earlier quarter. The prospectus puts CXMT’s share of the global DRAM market at 7.67 per cent for 2025.
Those are large numbers and we are not going to soften them. A company that lost money a year ago now earns tens of billions of yuan a quarter. The question this piece is about is what produced it.
The market moved more than the company did
TrendForce puts first-quarter 2026 DRAM industry revenue at about 97 billion dollars, up about 81 per cent on the previous quarter. Commodity DRAM contract prices rose between roughly 93 and 98 per cent quarter on quarter. The whole industry roughly doubled its revenue in three months without anyone shipping a new generation of product.
That matters for reading CXMT specifically, because CXMT’s output is almost entirely conventional DDR and LPDDR. Its exposure to the part of the market that repriced is close to total. A sevenfold revenue increase against a doubling in price, on a rising bit volume off a small base, is arithmetic we can follow. It does not require the company to have caught anyone.
Global DRAM revenue share
First quarter 2025 against first quarter 2026
Upper bar in each pair is the first quarter of 2025, lower bar the first quarter of 2026. Revenue share, drawn to scale against each other. Note that the largest single move in this chart is SK hynix losing seven points to Samsung, which has nothing to do with CXMT. Source: Counterpoint Research quarterly DRAM tracker.
The chart contains a fact that gets lost in the CXMT discussion. Between these two quarters the biggest change in the table is not CXMT adding five points. It is SK hynix losing seven and Samsung gaining four. The leadership of the DRAM market changed hands in this period, and CXMT was not the reason.
Four places to look for the gap
Market share tells you what was sold. It does not tell you what can be built. For that, four things are worth separating.
Process node
CXMT’s volume process is its G4 generation, which corresponds to the 1z class that Samsung and SK hynix put into mass production in 2019. Its G5 process is in development. Reported production sits in the 15 to 17 nanometre class, where it makes DDR5 at 8,000 megabits per second and LPDDR5X at 10,667 megabits per second in volume.
Those are shipping products, not samples, and that is the part people underrate. But the node comparison is about three generations, and in memory a node generation is a cost structure. Being three years behind on process is being three years behind on cost per bit at the same price.
High bandwidth memory
This is where the distance is largest and where the money is. CXMT is understood to be targeting first HBM3 output in 2026 and HBM3E in 2027, placing it three to four years behind the leaders on the product line.
Against that: SK hynix said in its second-quarter 2026 release that HBM4 mass shipments had begun and that the part met customer speed requirements with what it described as the industry’s best power efficiency. Samsung reported expanding HBM4 supply in the same quarter and shipping the industry’s first HBM4E samples. So the comparison is first-generation HBM3 against volume HBM4 and early HBM4E. Two full generations, on the product that carries this cycle’s margins.
It follows that CXMT’s revenue surge is not evidence of progress here. Its bit sales are conventional DRAM. It did not participate in the HBM market that produced most of the Korean profit.
What the Korean quarter looked like
SK hynix reported second-quarter 2026 revenue of 79.32 trillion won and operating profit of 60.54 trillion won, an operating margin of 76 per cent, with net profit of 93.92 trillion won. First-half revenue passed 100 trillion won for the first time. The company said it had long-term agreements with around ten core customers.
Samsung Electronics reported consolidated revenue of 171.5 trillion won and operating profit of 89.5 trillion won. Its Device Solutions division — the semiconductor business — accounted for 127.5 trillion won of revenue and 89.2 trillion won of operating profit, a margin of about 70 per cent. Its Device eXperience division, which sells phones and consumer electronics, recorded an operating loss of 0.8 trillion won, squeezed by the same component prices.
Read those two Samsung numbers together before quoting either. Group operating profit of 89.5 trillion won and semiconductor operating profit of 89.2 trillion won are almost identical, and the reason is that the consumer division lost money. Citing the group figure as a semiconductor figure happens to be nearly right this quarter, and would be wrong in most others.
Access to the tools
Here we have to correct a claim we keep seeing made too strongly, including in the material we started from. CXMT is not on the United States Entity List. The House Select Committee on the Chinese Communist Party, describing firms that meet its advanced-chip production thresholds, notes that such companies are typically also on the Entity List “with one major exception being CXMT”. That exception is the company in this article.
What did happen is narrower. In June 2026 the Department of Defense updated its 1260H list of Chinese military companies to 188 entries and included CXMT alongside YMTC, citing indirect state ownership and links to Chinese industrial and defence-technology bodies. The department had left both firms off a February version, which it published and then withdrew. Inclusion on 1260H does not by itself impose export controls or sanctions.
So the honest version of the tooling argument is this. CXMT does not have access to extreme ultraviolet lithography, which constrains how far it can scale, and that constraint is real and durable. But it has not been cut off the way YMTC was, and the reporting through 2026 points to an administration that is enforcing existing rules rather than writing new ones while trade talks continue. Anyone arguing the gap is protected by sanctions should look at which list CXMT is actually on.
Why the sell-off is not the evidence people think it is
The premise we set out to test was that Korean shares fell because the market concluded CXMT had caught up. Putting the trading days in order does not support it.
| Date | Event | CXMT involved |
|---|---|---|
| 6 July 2026 | Morgan Stanley tells clients to cut exposure to Samsung, SK hynix and Micron, arguing earnings momentum has passed its peak | No |
| Week of 10 July | SK hynix debuts on Nasdaq; shares rally sharply | No |
| 13 July 2026 | SK hynix falls more than 15 per cent in Seoul, its largest one-day decline in nearly two decades; the Seoul market drops 9 per cent and trading is halted | No |
| 27 July 2026 | CXMT lists in Shanghai and closes 465.82 per cent above its offer price | Yes |
| 28 July 2026 | SK hynix reports record quarterly results that miss forecasts; the shares fall sharply again | Partly |
The worst day came before the listing
The worst session of the month came on 13 July, when the market fell 9 per cent and SK hynix posted its biggest one-day drop in nearly twenty years. CXMT had not listed. What had happened was a Nasdaq debut, a rally, and an unwind — at one point the American listing carried a premium of about 25.6 per cent over the Seoul line, and a twice-levered Hong Kong fund tracking the name lost a quarter of its value.
Two more things sit in the same window. Morgan Stanley — the house that moved Korean semiconductor shares with its earlier “Memory, Winter is Coming” work — told clients on 6 July that the narrow semiconductor-led advance was ending. And the 28 July decline followed results that were the best in the company’s history and still below what analysts had modelled. A stock can fall on a record quarter. That is a statement about expectations, not about Chinese competition.
We would add that the same bank later described the memory sell-off as an entry point, and that other houses argued demand remained intact. We are not citing that as vindication of anything. We are noting that the sell-side view moved inside a single month, which is a reason to be careful about treating any of it as a finding.
What genuinely narrowed
Three things, and we would rather state them plainly than have a reader find them elsewhere and conclude we hid them.
CXMT’s share of DRAM revenue went from roughly 3 per cent to roughly 8 per cent in four quarters. On the prospectus figure it held 7.67 per cent for 2025. Whatever the driver, it is now a measurable part of the market rather than a rounding error.
It is shipping current-interface commodity parts in volume — DDR5 and LPDDR5X, not legacy DDR4. In the segment where products are close to interchangeable and switching costs are low, that is enough to compete on price. Morningstar, initiating on the stock, made exactly that point in the other direction: it argued DRAM is broadly commoditised with little differentiation, and assigned no economic moat, to CXMT and by implication to the segment.
And it now has capital. The raise was about 8.6 billion dollars, earmarked in the prospectus for production lines and technology upgrades. A company three generations behind with funding is a different proposition from one three generations behind without it.
What would change our mind
We would rather name the evidence in advance than reassess quietly later. Four things would move us.
Qualified HBM shipments to a named hyperscaler or accelerator vendor. Not a sample, not an announcement — a qualification. That is the step that would put CXMT in the market that actually pays.
Volume production on G5 with disclosed yield. The node gap is the cost gap, and closing it is the thing that would let CXMT compete when prices normalise rather than only while they are doubling.
Margins holding through a down leg. First-quarter 2026 was a quarter in which commodity contract prices nearly doubled. Every commodity DRAM maker looked good. The test is the quarter after that stops.
A change in tool access in either direction. Formal Entity List designation would tighten the constraint; a relaxation of lithography restrictions would loosen it. Both are live possibilities and neither is our forecast.
A practical note for readers outside Korea and China
One thing worth knowing if this article prompted you to look for the shares: a STAR Market listing is not generally accessible to foreign retail investors. Access runs through qualified institutional channels or funds holding the name, and inclusion in the Stock Connect programme — the route that would open direct purchase — had not happened at the time of writing. Korean retail investors, for instance, could reach CXMT only indirectly.
That asymmetry is worth sitting with. The company at the centre of a Korean market panic was one that most of the people reacting to it could not buy. If you want the mechanics of which listings are actually reachable and which merely look reachable, we set that out in our piece on Samsung’s American quotation, and the account-level route into Korean shares is in how foreign investors buy Korean stocks.
The dated record
| What | Figure | Where it comes from |
|---|---|---|
| CXMT offer price and first close | 8.66 yuan to 49.00 yuan, +465.82% | Shanghai listing, 27 July 2026 |
| CXMT raise | about 57.9 billion yuan | Listing terms, about 10% of post-issue free float |
| CXMT first-quarter 2026 revenue | about 50.8 billion yuan, +719% | Prospectus, as reported by Reuters |
| CXMT first-quarter 2026 profit to shareholders | about 24.8 billion yuan | Prospectus; year-earlier loss of about 1.56 billion yuan |
| CXMT 2025 DRAM share | 7.67% | Prospectus |
| DRAM share, first quarter 2026 | Samsung 38%, SK hynix 29%, Micron 22%, CXMT 8% | Counterpoint Research |
| Commodity DRAM contract prices | +93% to +98% quarter on quarter | TrendForce, first quarter 2026 |
| SK hynix second quarter 2026 | Revenue 79.32tn won, operating profit 60.54tn won | SK hynix results release |
| Samsung second quarter 2026 | Group 171.5tn / 89.5tn won; DS division 127.5tn / 89.2tn won; DX division −0.8tn won | Samsung Electronics results release |
| CXMT volume node | G4, equivalent to 1z class of 2019 | Korean press reporting, July 2026 |
| CXMT HBM position | HBM3 targeted 2026, HBM3E 2027 | Published analysis of Chinese HBM development |
| Korean HBM position | HBM4 mass shipments; first HBM4E samples | SK hynix and Samsung results releases |
| CXMT and the Entity List | Not listed | House Select Committee on the CCP |
| CXMT and 1260H | Added June 2026, 188 entries | Department of Defense |
Sources
We take facts from these and write our own sentences. Where a figure originates in a company filing we say so, and where it originates in a research house estimate we name the house. We link to a document only where we hold the exact address for it; a link to a publisher’s front page would not get you to the source, so where we do not have the direct address we give you enough to find it.
Filings and official documents
- SK hynix, second-quarter 2026 business results, 29 July 2026. Revenue, operating profit, net profit, first-half cumulative revenue, HBM4 mass shipment and the long-term agreement count.
- Samsung Electronics, second-quarter 2026 results announcement, 29 July 2026 (Samsung Newsroom, global edition). Consolidated revenue and operating profit, and the divisional split for Device Solutions and Device eXperience.
- CXMT (ChangXin Memory Technologies), Shanghai Stock Exchange STAR Market listing prospectus, July 2026 — as reported by Reuters. First-quarter 2026 revenue and profit, the 2025 DRAM share figure of 7.67 per cent, and the stated use of proceeds. We did not read the Chinese filing directly and say so rather than imply otherwise.
- United States House Select Committee on the Chinese Communist Party, report on advanced semiconductor production in China. Source for the statement that firms meeting its production thresholds are typically on the Entity List, “with one major exception being CXMT”.
- United States Department of Defense, Section 1260H list of Chinese military companies, June 2026 update, 188 entries. Source for CXMT’s inclusion and for the fact that inclusion does not itself impose export controls.
Research houses and reporting
- Reuters, SK hynix shares in Seoul after the Nasdaq debut, 13 July 2026. The 15 per cent single-day decline, the 9 per cent market fall, the trading halt and the roughly 25.6 per cent premium on the American line.
- Counterpoint Research, quarterly DRAM revenue share tracker, first quarter 2025 and first quarter 2026. The four-company share table and the year-on-year market growth figure.
- TrendForce, first-quarter 2026 DRAM industry revenue and contract pricing. Industry revenue of about 97 billion dollars and the 93 to 98 per cent quarter-on-quarter move in commodity contract prices.
- Chosun Ilbo (English edition), report on Morgan Stanley’s 6 July 2026 recommendation to reduce exposure to Samsung Electronics, SK hynix and Micron. Cited as evidence of what the sell-side argued on that date, not as a view we endorse.
- Morningstar, initiation note on CXMT. Cited for its assessment that DRAM is a broadly commoditised product with low switching costs and no economic moat — an opinion, not a fact.
- Korean press reporting, late July 2026, for CXMT’s G4 process generation and its correspondence to the 1z class of 2019, and for volume DDR5 and LPDDR5X data rates. Cited as reporting of industry assessment rather than as company disclosure.
What we could not verify
Several figures that circulate in comparisons of these companies did not survive a check against a primary document, and we have left them out rather than attribute them vaguely.
CXMT’s operating margin. A widely quoted figure of about 70 per cent for the first quarter of 2026 comes from a subscription research house and does not appear in the material we could read directly. Because it invites a side-by-side with SK hynix at 76 per cent and Samsung’s semiconductor division at about 70 per cent — both of which we can compute from published releases — using an estimate alongside them would imply a precision we do not have.
Monthly wafer capacity. Figures putting CXMT near Micron are analyst estimates. We found no filing to support a specific number.
HBM4 allocation shares between Samsung and SK hynix for any named customer platform. Supply agreements are not public. Percentages we saw attributed to specific accelerator programmes are broker estimates.
The claim that multi-patterning with deep ultraviolet tools requires roughly 30 per cent more wafer passes than an extreme ultraviolet equivalent. Directionally uncontroversial, but the specific figure traces to research-house modelling rather than to a tool vendor or a filing.
Index levels and share prices beyond the individual moves reported above. Korea Exchange operates a public data interface for daily closes; our attempt to authenticate against it for this piece did not succeed, so we have restricted ourselves to price moves reported by named news organisations on named dates.
Corrections
None yet. If you can point to a primary document that contradicts anything above, we will correct it and say what changed.
Disclosure. The author holds no position in Samsung Electronics, SK hynix, Micron Technology or CXMT as of the date of publication. This article is general information and analysis about memory market structure, company disclosures and competitive position. It is not investment advice, not a recommendation to buy or sell any security, and not a forecast of any price or index level. The section headed Editor’s opinion is comment: it argues a view, and we fence it off from the sourced material for that reason. We have not tailored anything here to your circumstances, and only a qualified adviser in your jurisdiction can do that. We are not registered as an investment adviser in any jurisdiction, and this site has no commercial relationship with any company, exchange or research provider named above. See the Editorial & Sourcing Policy and Disclaimer.