ChangXin Technology earns a net profit of 77.6 billion yuan in half a year, and invests another 34.9 billion yuan to bet on next-generation DRAM

[[**CXMT earned a net profit of 77.6 billion yuan in half a year and simultaneously announced 34.9 billion yuan in investment (24.1 billion for R&D + 10.8 billion for testing), using profits and IPO over-raised funds to expand capacity counter-cyclically, consolidate its position as the world's fourth-largest DRAM maker, and bet on next-generation processes and in-house testing capabilities.**

**Key points**

**1. Half-year profit surged, with gross margin as high as 84.84%**

In the first half of 2026, revenue was 150.3 billion yuan and net profit attributable to shareholders was 77.6 billion yuan, turning profitable year-on-year; the second quarter grew 113% quarter-on-quarter, and profit is still accelerating. AI computing demand has pushed up DRAM prices, and CXMT filled the general-purpose memory gap left after the giants shifted toward HBM.

**2. 34.9 billion invested in R&D and testing to shore up IDM shortcomings**

24.1 billion is for DRAM process iteration (the fifth-generation 11.95nm process has just entered mass production), and 10.8 billion is for building a back-end testing base to resolve the bottleneck in external packaging and testing capacity. Of this, 18 billion comes from IPO over-raised funds, adding no new debt and not diluting shareholders.

**3. Downstream GigaDevice is paying more to grab capacity, a clear seller's-market signal**

GigaDevice raised its 2026 related-party transaction quota from 5.7 billion to 7.85 billion yuan, and expects 4.4 billion yuan for January-April 2027, already exceeding its total revenue in the first quarter of 2026. An asset-light design company sharply raising prices to grab foundry capacity confirms that DRAM capacity is extremely tight.

**4. Overseas memory stocks fall while CXMT expands counter-cyclically, creating a time gap**

U.S. stocks such as Micron and SK Hynix fell under pressure from high interest rates, while CXMT invested 34.9 billion yuan to expand capacity. Its testing base will start production in 2029, coinciding with a cyclical trough; counter-cyclical investment is expected to replicate Samsung's history—taking the largest share in the next upcycle. 2026-09-30 09:58 Earned 77.6 billion in half a year, CXMT again puts up 34.9 billion to bet on next-generation DRAM | Laika Think Tank Quick Overview

This article is from the WeChat public account: Laika Think Tank, author: AA, editor: G3007

On the evening of September 28, CXMT (688825.SH) issued two investment announcements in succession: a 24.1 billion yuan technology R&D project, plus a 10.8 billion yuan second phase of a memory wafer back-end testing base, totaling 34.9 billion yuan in a single day. Of this, 18 billion comes directly from IPO over-raised funds, and the rest is covered through capital increases and loans.

A company had just recorded a rare profit figure for A-shares in its half-year report, and then turned around and plowed profits and over-raised funds back into the production line. How should this move be read?

01

What is the quality of the 77.6 billion profit?

In the first half of 2026, CXMT achieved operating revenue of 150.31 billion yuan, up 873.64% year-on-year; net profit attributable to shareholders was 77.605 billion yuan, while in the same period last year the company was still losing 2.332 billion yuan. Gross margin of the main business was 84.84%, and net operating cash flow was 131.156 billion yuan.

It is simply too profitable.

An 84.84% gross margin for the main business is quite outstanding even among large manufacturing enterprises across the entire A-share market. AI computing demand has driven global DRAM supply short of demand; Samsung, SK Hynix, and Micron have shifted capacity toward HBM and server DDR5, leaving a structural gap in general-purpose memory products, and CXMT has precisely captured this market.

More noteworthy is the quarter-on-quarter curve. First-quarter net profit attributable to shareholders was 24.762 billion yuan, and second-quarter was 52.843 billion yuan, up 113% quarter-on-quarter. That is, by mid-year this year, the profit brought by price increases was still accelerating rather than peaking and falling back.

Market share also rose in step. According to CFM flash memory market data, in the second quarter of 2026 the global DRAM market was 147.024 billion U.S. dollars, up 55.9% quarter-on-quarter and a record high; after Samsung, SK Hynix, and Micron, CXMT's share has approached 10%, firmly holding fourth place globally, and its quarterly sales revenue quarter-on-quarter growth ranked first globally.

The R&D project has total investment of 24.1 billion yuan, a construction period of 30 months, and focuses on DRAM process iteration. The background for this investment is that CXMT has just completed a key leap: in September, the company's fifth-generation DRAM process technology platform officially entered mass production, relying on quadruple patterning technology to shrink the half-pitch of the active area of the memory array to 11.95 nanometers, placing it among the world's top mass-production processes. In the first half of this year, R&D investment was 6.859 billion yuan, up 87.38% year-on-year; 24.1 billion is equivalent to 3.5 times the first-half R&D investment.

The second phase of the testing base answers a more practical question. The wording in the announcement is quite direct: as capacity continues to be released and the product matrix keeps enriching, insufficient external packaging and testing capacity and limited supporting resources have begun to constrain capacity release. This testing plant, with total investment of 10.8 billion yuan and a construction period of 37 months, will handle DRAM wafer testing and module assembly after completion, further strengthening back-end testing capabilities under the IDM model.

It should be noted that 34.9 billion yuan is the planned total investment for the two projects, of which 18 billion yuan is intended to use over-raised funds. The money the market gave CXMT at the time of listing has still not been fully spent, and now it is being turned around and invested in the production line, with no new debt and no shareholder dilution. Profit for capacity, cash flow for time—the approach to this calculation is quite traditional, and quite solid.

On the evening of September 29, GigaDevice announced that it would raise its estimated quota for routine related-party transactions with CXMT Group in 2026 from 5.711 billion yuan to 7.854 billion yuan. The reason for the increase was stated clearly: DRAM product foundry prices are rising with market conditions. At the same time, the company expects the related-party transaction quota for January to April 2027 to reach 4.411 billion yuan.

The weight of these figures needs to be converted for perspective: 4.411 billion is only a four-month projected quota, yet it already exceeds GigaDevice's entire revenue for the first quarter of 2026 (4.188 billion yuan) and approaches half of its full-year 2025 revenue (9.203 billion yuan). Yet the actual related-party transactions between the two sides in 2025 amounted to only about 1.18 billion yuan—in one year, the quota has swelled 6.7-fold.

Another layer of signal is hidden in the execution progress. From January to August 2026, 3.596 billion yuan was actually incurred, leaving 4.258 billion yuan of headroom below the 7.854 billion cap—meaning that in the second half of the fourth quarter, the procurement pace is highly likely to accelerate further.

An asset-light design company paying a premium to grab foundry capacity—this is the most typical micro-level form of a seller's market. A fabless manufacturer can do without a wafer fab, but it cannot do without DRAM capacity, and at present, the capacity worldwide that can undertake foundry production of specialty DRAM at scale consists of just a few lines.

04

The Other Side's Price: US Memory Stocks Plunge Overnight

The other side of the bustle is another quote the global capital market has given this industry.

Around the time of CXMT's announcement, US memory stocks fell collectively: Micron closed down 2.61%, SK Hynix closed down 5.03%, SanDisk closed down 3.65%, and the Philadelphia Semiconductor Index closed down 1.61% (at one point down 3% intraday). The backdrop: the 10-year US Treasury yield rose intraday to 5.27%, a new high since 2007, and the 30-year rose to 5.57%, a new high since 2002; under a high-rate environment, high-valuation growth assets came under broad pressure.

On the A-share side, CXMT Technology closed down 4.27% on September 28, with a total market capitalization of 3.65 trillion yuan, still firmly atop the A-share market cap rankings. On one side, the domestic leader is bucking the trend with heavy investment to expand capacity; on the other, overseas memory assets are seeing valuation contraction—the same industry, two markets giving pricing in opposite directions.

Signals on the fundamentals side are also worth recording: because memory manufacturers and some server customers have locked in long-term agreement price ceilings, the room for price increases on non-long-term orders is shrinking quarter by quarter, and institutions generally expect DRAM prices to continue rising in the third quarter but at a narrowing pace. The explosive surge phase is most likely passing.

So, with price-increase momentum slowing, is spending 34.9 billion yuan on capacity expansion a contrarian gamble or a textbook move?

The history of the DRAM industry offers the latter as the reference. This is a track known for deep cycles, where the amplitude of price surges and crashes far exceeds that of general manufacturing, and the reshuffling of the industry landscape has almost always occurred in capacity-expansion decisions during downturns: whoever keeps building production lines when money is scarcest takes the largest share in the next upcycle. Samsung back then squeezed Japanese rivals out of the memory market precisely through countercyclical investment across multiple consecutive downturns.

Place CXMT's timetable into this framework. The second phase of the test base has a construction period of 37 months, with the production start node falling around 2029—exactly corresponding to the window when this round of shortage eases and supply and demand seek a new bottom. By then, a company with new capacity, fifth- and even sixth-generation processes, and a self-built testing system will be in a completely different position from a company that has only earned one round of price-increase money.

Moreover, CXMT's expansion is not homogeneous volume stacking. The structural gap in general-purpose DRAM left by the giants, the world's first commercial mass production of LPDDR6 (debuting in the Xiaomi 18 Fold, with a peak rate of 12,800 Mbps and a maximum capacity of 16GB), and CXMT's 10,667 Mbps LPDDR5X landing on the Nubia NaviX Ultra (the first mass-production application of a domestic product at this rate level)—these product-side positionings show that the 34.9 billion yuan buys not just capacity, but a differentiated position.

Conclusion

The mid-game judgment on the memory super-cycle remains unchanged; what has changed is the focus of observation. What is worth watching next is no longer how many points DRAM rises each quarter, but three slower variables: when the sixth-generation process beyond 11.95 nanometers lights up, at what pace the 7.854 billion yuan related-party quota will be executed next year, and where in the cycle the industry will be when the test base starts production in 2029.

For the secondary market, a measure of sobriety must be retained: the valuation peak of cyclical stocks has historically appeared before the profit margin peak. The 34.9 billion yuan is industrial capital's bet on the future, while financial capital prices the slope of current-period profits—the time gap between the two is precisely the entire game space of this business. Channel: Finance and Economics

10 articles published on Huxiu. This content is published with the author's authorization; the views represent only the author and not Huxiu's position. If you have objections or complaints about this manuscript, please contact tougao@huxiu.com. Everyone who is changing and wants to change the world is on the Huxiu APP[ ]