ChangXin Memory Technologies (CXMT, 688825.SH) closed its first day of trading on July 27 at RMB49.00, up 465.8% from the RMB8.66 offer price, for a market capitalization of RMB3.277 trillion. On an A-share basis — the A-share close multiplied by total shares outstanding — that puts it ahead of ICBC at RMB2.755 trillion, China Construction Bank at RMB2.71 trillion, Agricultural Bank of China at RMB2.352 trillion and China Mobile at RMB2.047 trillion, making it the largest company listed on China’s domestic market. ICBC and its peers also carry Hong Kong lines, so this is a single-market comparison rather than a global one.
Set the headline aside. The session threw off four mismatches that matter more.
One Stock Absorbed More Money Than the Entire STAR Board
CXMT traded RMB141.2 billion on debut. On the same day, all 50 constituents of the STAR 50 index combined traded RMB123.1 billion. CXMT is newly listed and not yet an index constituent, so the two figures do not overlap: one IPO out-traded the entire benchmark basket. Against the RMB2.076 trillion of combined Shanghai and Shenzhen turnover (RMB1.031 trillion and RMB1.045 trillion respectively), a single name accounted for roughly 6.8% of all activity.
Yet the STAR 50 rose only 1.16%, well behind ChiNext’s 3.16%. Semiconductors did not get bid up as a group. The money concentrated into one ticker.
A RMB3.28 Trillion Price Tag Set by a RMB220 Billion Float
Total market capitalization is RMB3.277 trillion. Free-float market capitalization is RMB220.6 billion. Tradable shares number 4.503 billion against 66.88 billion total — a float of 6.7%.
A RMB3.28 trillion valuation was therefore discovered on RMB220.6 billion of tradable paper. First-day turnover was 66.4%, so two-thirds of that 6.7% changed hands in a single session. The scarcity here is in the shares as much as in the company.
The variable ahead is the lock-up schedule. When the remaining 93.3% comes off in tranches under the terms of the offering, the market gets its first real test of what is holding today’s price up. That is a point about the pricing base, not about the business.
The New Leader Rose 466%. The Domestic Memory Chain Did Not Follow.
If investors read the listing as a catalyst for China’s memory complex, the most obvious proxies should have moved with it. They did not. GigaDevice fell 5.03%. Longsys rose 1.05%, Biwin Storage 2.01%, Techwinsemi 0.43% and Shannon Semiconductor 2.63% — every one of them lagged ChiNext’s 3.16%, and the name most often used as shorthand for “domestic memory,” GigaDevice, dropped outright.
Global memory was firm that day. KOSPI closed at 6,755.75, up 0.97% after a 5.72% drop in the prior session; SK Hynix rose 3.24% and Samsung Electronics 1.80%. So this was not a memory-cycle problem showing up on the tape.
One candidate explanation: buyers treated CXMT as a scarce asset rather than a sector catalyst, funding some of it by rotating out of the incumbent memory names. That is a hypothesis, not a conclusion — a single cross-section cannot establish flows. But GigaDevice’s decline is enough to say the rising tide did not lift the group.
A 22x Forward P/E May Be a Warning, Not a Margin of Safety
Valuation is the part most easily read backwards. The IPO priced at 308.92x trailing earnings, which looks absurd. Divide the RMB3.277 trillion first-day capitalization by the consensus median 2026 net income estimate of RMB148.5 billion, and the forward multiple is roughly 22x. After a rise to 5.66 times the offer price, the stock looks cheaper on forward numbers than it did at the offer.
The problem is the denominator. First-quarter 2026 revenue was RMB50.8 billion with net income of RMB24.76 billion — a 48.7% net margin. The first-half pre-announcement guides revenue growth of 612.53% to 677.31% year over year and net income of RMB50 billion to RMB57 billion, up 2,244.03% to 2,544.19%, which the company attributes to rapidly rising DRAM product prices, continued growth in production and sales volume and an improving product mix. In the year-ago half, the company still lost RMB2.387 billion excluding non-recurring items.
So the 22x is market cap divided by earnings generated at peak-cycle pricing and peak-cycle margins. DRAM is a deeply cyclical product, and low multiples at cycle highs are the norm for this asset class, not a discount signal. The question is not whether 22x is cheap. It is how many quarters that RMB148.5 billion denominator survives.
Capacity and technology cut both ways too. Media reports put CXMT’s 2026 monthly capacity at roughly 300,000 wafers; third-party research firm Citrini Research estimates that if expansion proceeds on plan, capacity could reach about 350,000 wafers per month by end-2026, in the same range as Micron’s roughly 375,000. DDR5 runs on a 16nm node with reported average yields above 80%, though accounts differ on when exactly that level was reached, and DDR4 capacity is reportedly being wound down. HBM3 is where it gets thinner. By the same reports it has only been sampled to domestic customers, front-end yields run 35% to 50% and overall yield roughly 25% against above 90% at the international leaders, and trial production is targeted for the end of 2026. That is the concrete gap with SK Hynix and Samsung: close to parity in mainstream DRAM, not yet there in the product where AI demand is most concentrated.
What Would Falsify the Above
Lock-up absorption. If tranches of restricted stock unlock without a visible drop in turnover or sustained price pressure, the thin-float argument weakens and should carry less weight.
DRAM spot pricing. If spot prices keep rising or hold at current levels for several more quarters, the RMB148.5 billion denominator holds up better and the peak-cycle-trap reading gets weaker. A rollover would confirm it.
HBM execution. If HBM3 yields and trial production land on the end-2026 target, the capability-gap framing needs rewriting. Slippage means repricing whatever the market has already paid for.
A delayed sector move. If GigaDevice, Longsys and the rest catch up over the next few sessions, the “scarce asset, not sector catalyst” explanation is undercut — that would look more like debut-day liquidity absorption than a difference in how the market positions these names.
One session establishes one session. The debut settled nothing about domestic substitution. It produced four mismatches, each of which the data from here can falsify.
Sources: Shanghai Stock Exchange announcements, Eastmoney Choice, and the Korea Exchange (KRX). Prices and market capitalizations are as of the July 27, 2026 close; the 2026 consensus figure is the market median as of July 26, 2026. Market-cap comparisons use an A-share basis (A-share close × total shares outstanding); companies such as ICBC also have Hong Kong listings and are not directly comparable on a cross-market basis. Monthly capacity, yield and expansion figures come from media reports and third-party estimates by Citrini Research rather than company disclosure, and definitions may differ. The discussion of rotation flows and peak-cycle earnings is a research observation, not established causation. The individual securities named here (CXMT, GigaDevice, Longsys, Biwin Storage, Techwinsemi, Shannon Semiconductor) appear as objects of analysis and do not represent a recommendation by KSINQ to buy, sell or hold them; first-day and offer-price-relative returns reflect past performance and do not indicate future results. This piece is a personal observation and does not constitute investment advice.