Within CXMT's hundred-billion profit lies the first inflection point

[[**CXMT turned profitable in the first three quarters and earned more than 131 billion yuan, but in Q3 revenue rose while net profit attributable to parent barely increased, with net margin falling from 53% to 46%, marking the first inflection point in profit growth; the key is to watch Q3 gross margin, taxes, and minority interest.**

**Key points**

**1. Q3 profit and revenue clearly diverged**

Revenue rose by 15.2 billion yuan quarter-on-quarter to 20.2 billion yuan, while net profit attributable to parent rose by only 550 million yuan to 2.05 billion yuan, and net margin fell from 53.1% to about 46%, showing that profitability has peaked.

**2. Profit leakage may be the result of multiple factors**

Minority interests, taxes, depreciation, expenses, and deviations of actual selling prices from the index could all erode profits; in the Q3 report, focus on gross margin, income tax expense, and minority interest gains/losses to determine whether the issue is accounting or operational.

**3. Price increases narrow, and the stock price peaks before profit**

DRAM contract price quarter-on-quarter growth narrowed from 90% to 95% in Q1 to 13% to 18% in Q3; in the previous cycle, Micron's stock price peaked about a quarter earlier than profit.

**4. ChangXin's capacity expansion and technological shortcomings may end the price increases**

ChangXin's share is about 7.7%, with plans to double capacity before 2030; HBM is still in validation, customers skew toward consumer, and incremental general-purpose DRAM depends on supply shortages. The buyback accounts for only 0.03% to 0.07% of market value, showing priority on capacity. 2026-10-25 00:16 Inside ChangXin's hundred-billion profit lies the first turning point | Wenwu Zhao Quick Take

This article is from the WeChat public account: AD Finance and Economics, author: Shi Yousong

One

On the evening of October 16, ChangXin Technology released its performance forecast for the first three quarters: revenue of approximately 265 billion to 270 billion yuan, and net profit attributable to parent of approximately 131 billion to 132.5 billion yuan. In the same period last year, the company was still losing 5.28 billion yuan.

This is indeed a historic figure: a domestic DRAM manufacturer that was still posting huge losses three years ago earned more than 130 billion yuan in nine months.

But if you look at the third quarter on its own, a set of rather odd numbers emerges: third-quarter revenue was about 15.2 billion to 20.2 billion yuan more than the second quarter, while net profit attributable to parent was only about 550 million to 2.05 billion yuan more.

In other words, for every extra 100 yuan sold, it earned only 4 to 10 yuan more. And just in the previous quarter, the company's net margin attributable to parent was still 53%.

Memory chips are a classic heavy-asset, high-fixed-cost industry. The depreciation of plants and equipment is basically fixed, so for every additional chip sold, most of the proceeds turn directly into profit. That is why, during upswing cycles, memory makers' profit growth always far outpaces revenue growth.

ChangXin in the third quarter was the exact opposite: the extra money from selling more almost never turned into profit.

The forecast only gave the third-quarter range. Only by putting it together with the first-quarter data disclosed in the prospectus and the half-year report data does the quarterly trajectory become complete:

This table shows one thing: the profit margin peaked in the second quarter.

Second-quarter revenue grew 96% and net profit attributable to shareholders grew 113%, with profit outpacing revenue—this is what operating leverage looks like when it is working normally. In the third quarter, revenue still grew by more than 15%, but net profit attributable to shareholders was almost flat, and the net margin attributable to shareholders fell from 53.1% to around 46%, a drop of about 7 percentage points.

This is the first time in this cycle that ChangXin has seen revenue still rising while its profit margin falls.

II

The popular explanation for the slowdown in profit growth is that the pace of DRAM price increases has narrowed. Data from TrendForce shows that the quarter-on-quarter increase in global DRAM contract prices fell from 90% to 95% in the first quarter, to 58% to 63% in the second quarter, and then to 13% to 18% in the third quarter.

But that does not explain why profit has barely grown.

After all, prices are still rising. In the third quarter, contract prices still rose 13% to 18%, and ChangXin's revenue also grew 15% to 20%. The two magnitudes are comparable, indicating that the extra revenue mainly came from price increases.

The revenue brought by price increases should have been the "fattest" revenue. Production costs did not change, the product sold for more, and the price difference almost entirely fell to profit. By conventional logic, of the extra 15 billion to 20 billion yuan in revenue, profit should have accounted for the bulk, not 4% to 10%.

So the problem is not on the revenue side. The money must be leaking somewhere in the income statement.

III

Before the third-quarter report is officially disclosed, no one can give a definitive answer. But based on the information ChangXin has made public, at least five areas are worth suspecting.

Suspicion one: the share taken by minority shareholders. This is the most direct piece of evidence. The prospectus shows that ChangXin's net profit in the first quarter of this year was 33.012 billion yuan, with net profit attributable to the parent company at 24.762 billion yuan, leaving about 8.25 billion yuan, or roughly a quarter, to minority shareholders. ChangXin is a typical product of the "Hefei model," and the entities operating its production lines have brought in a large number of external shareholders such as local state-owned capital. The 131 billion to 132.5 billion yuan in the forecast is on an attributable-to-parent basis. If more of the third-quarter profit comes from entities in which external shareholders hold a higher stake, attributable profit will be systematically diluted, and the growth in the company's overall net profit may be far less weak than it appears.

Suspect two: income tax. From 2022 to 2024, ChangXin accumulated losses of more than 30 billion yuan, and these losses can be used to offset future profits for tax purposes. Based on the profit scale in the fourth quarter of 2025 and the first quarter of this year, this offset quota was likely largely used up in the first half of the year. If so, the third quarter may be ChangXin's first quarter of full taxation at the normal rate: pre-tax profit still grows, but after-tax profit has a chunk taken out of it.

Suspect three: depreciation. ChangXin has two 12-inch DRAM fabs in Hefei and one in Beijing, with monthly capacity of about 300,000 wafers, and a new Shanghai plant scheduled to start production in 2027; the 29.5 billion yuan raised in the IPO is also mainly directed to production line upgrades and technology upgrades. New equipment at memory fabs carries the heaviest depreciation right when it is first capitalized, before output has ramped up, so unit costs actually rise. The company stressed in its forecast that "capacity utilization remains high," but that refers to existing production lines; depreciation on the new equipment is a separate matter.

Suspect four: expenses. In the first half, ChangXin's R&D spending was 6.859 billion yuan, up 87% year on year. In addition, before its listing ChangXin implemented two employee stock ownership plans covering about one-third of employees, and the grant price of the second plan was only 0.108 yuan per registered capital. After the listing, the share price surged, and share-based payment expenses may show up more clearly on the income statement. The 1 billion to 2 billion yuan buyback announced the same day is partly intended for employee equity incentives.

Suspect five: actual selling prices and the index are not in sync. The 13% to 18% range is TrendForce's industry contract price measure. ChangXin's product mix (how much each of servers, phones, and PCs accounts for, and the ratio of DDR5 to LPDDR) and its long-term agreements with major customers will all cause its actual selling prices to deviate from the index. According to Mizuho research, ChangXin may have already signed long-term agreements with ByteDance, Alibaba, and Tencent. Long-term contracts lock in sales volume in the early stage of a price increase, and in the later stage of a price increase, they may also lock in prices.

These five suspects are not mutually exclusive, and the real answer is most likely a combination of several. On the day the third-quarter report comes out, focus on three lines: gross margin, income tax expense, and minority interest income.

Gross margin is still rising: the problem lies in the lower part of the income statement, such as taxes, minority interests, and expenses. The operating fundamentals remain strong, and a considerable part of the third-quarter "cliff" is at the accounting level.

Gross margin has already fallen back: depreciation or selling prices have run into trouble. This is the real signal of a cyclical turning point.

Four

The third-quarter report will tell us whether this "cliff" is an accounting problem or an operating problem.

But one thing is already clear: the second derivative of price increases has long since turned negative.

The memory industry has an inertia: compared with price reports, the market pays more attention to changes in the rate of increase. Stock prices tend to peak when the rate of increase peaks, not when prices peak, and certainly not when profits peak.

The previous super-cycle played out in exactly this order:

In short, the rate of increase narrowed first, the stock price peaked first, profits peaked next, and prices fell last.

Micron's stock price peaked about one quarter earlier than profits, and about two quarters before prices turned down. After prices turned from rising to falling, Micron's average DRAM price had cumulatively fallen more than 55% from its peak.

What triggered the stock price peak was the rate of increase dropping to single digits. That is exactly what happened in the first quarter of 2018.

Compare with 2026: contract price increases narrowed from more than 90% in the first quarter to 13% to 18% in the third quarter. The capital market's reaction has in fact already appeared.

In July of this year, Micron's average DRAM selling price hit a ten-year high of about $1.17, far above the 2018 peak of about $0.87, yet its stock price has already fallen about 28% from its June high. One of the reasons cited by overseas analysts is precisely that CXMT's listing will accelerate the expansion of China's DRAM capacity.

On its first day of trading on July 27, CXMT closed with a market value of 3.28 trillion yuan; since then the company has delivered results showing second-quarter net profit doubled and first-three-quarter net profit exceeding 130 billion yuan, yet its market value on October 16 was 3.06 trillion yuan, lower than on its first trading day.

Of course, there is one difference between this cycle and 2018. In the third quarter of 2018, SK Hynix's average DRAM price barely rose, yet its operating profit still grew 16% quarter on quarter, driven by shipment volume. In the third quarter of 2026, by contrast, CXMT's contract prices were still rising by double digits, yet its net profit attributable to shareholders was almost flat.

If this reflects operational problems, then CXMT's profit pressure has arrived earlier than it did for the Korean manufacturers in 2018; if it is mainly due to accounting factors such as taxes and minority interests, then the timing of the two is still roughly the same.

Five

Over the past year, the price increases were a rising-tide market. AI servers consumed a large amount of capacity, and the three major manufacturers prioritized allocating wafers to high-margin products, squeezing the supply of general-purpose DRAM. As a result, almost all DRAM was rising in price, from server memory to phone and PC memory. CXMT made its money in this kind of market.

The so-called "technology cycle" means that pricing power in the next stage must be won through the products themselves: whoever can supply HBM, whoever can make high-capacity server memory modules, and whoever can pass validation by AI server manufacturers will be the ones able to continue commanding high prices. For general-purpose, replaceable products, the momentum for price increases will fade first.

For CXMT, there are three realities:

HBM is not yet on the table.

CXMT's HBM3 samples have been sent to domestic AI chip manufacturers and are still in the pre-mass-production validation stage; the mass-production target for 12-layer HBM3E is 2027. Even if progress goes smoothly, there is still a two-to-three-year gap with Samsung and SK Hynix.

The customer structure leans toward consumer.

CXMT's core customers include terminal manufacturers such as Lenovo, Xiaomi, Transsion, Honor, OPPO, and vivo, and the company also acknowledged in its semiannual report that customer concentration is relatively high. Consumer electronics is precisely where high prices are felt first.

Its bargaining chip lies in the incremental growth of general-purpose DRAM.

Mizuho believes that CXMT may be the only supplier able to provide incremental DDR5 in 2027; CXMT has also captured most of the DRAM share in national-level project procurement. This is CXMT's moat, but what it is betting on is the "shortage" itself, and shortages will eventually pass.

There is another detail here. TrendForce data shows that since the first quarter of this year, the revenue per wafer of HBM has been overtaken by general-purpose server memory such as DDR5 64GB RDIMMs, and HBM's profit margin has accordingly fallen below the latter's.

In other words, the most profitable product right now is not HBM, but precisely the general-purpose server memory that CXMT can also make.

This sounds like good news for CXMT, but it is actually a double-edged sword.

If general-purpose DRAM is more profitable than HBM, the three major manufacturers will have ample incentive to shift part of their capacity from HBM back to general-purpose DRAM. TrendForce also noted that the manufacturers will adjust capacity allocation between the two based on the outcome of HBM price negotiations.

This round of supply-side correction does not necessarily require waiting for new fabs to come online; a single reallocation of capacity by the three major manufacturers could accomplish it. And the capacity that gets shifted back will first hit the very market that CXMT relies on to make money.

That is the supply coming from the major manufacturers. The other source of supply comes from CXMT itself.

Six

"The supply-demand tightness will persist beyond 2028" is the common judgment among the major manufacturers. But there is one big variable in that judgment: CXMT itself.

In the first quarter of this year, CXMT accounted for about 7.7% of global DRAM sales, ranking fourth worldwide.

Its current monthly capacity is about 300,000 wafers, and its new Shanghai fab will begin production in 2027; the company plans to double capacity by 2030 and triple it by 2035.

Counterpoint expects CXMT's bit shipment share to rise from about 9% currently to about 11% by 2028.

Goldman Sachs estimates a global DRAM shortfall of 4.9% in 2026. With a shortfall of less than 5%, prices can nearly double in a single quarter, which shows that demand is almost completely inelastic.

The same low elasticity works in reverse: as long as supply increases by just a few percentage points, prices can fall just as quickly. And among the new global DRAM supply, the one most certain to deliver those few percentage points is CXMT.

CXMT is both one of the biggest beneficiaries of this round of price increases and the player most likely to end it.

This is not to say that CXMT will cut off its own source of profit.

SemiAnalysis estimates that in the first quarter of this year, CXMT's average DRAM selling price was only 5% to 10% lower than those of the three major manufacturers. It did not rely on low prices to grab market share; instead, it enjoyed the price increases along with the rest of the industry. As long as the shortfall persists, capacity expansion and high prices can coexist.

The test comes at the moment the shortfall begins to narrow.

Samsung, SK Hynix, and Micron have been through several cycles and developed a tacit understanding of sorts: "curb output to defend prices." CXMT's capacity-expansion logic, however, is not entirely a commercial one: it carries the strategic mission of domestic substitution. According to Mizuho research, CXMT has already captured 80% to 90% of DRAM share in national-level project procurement channels; among its shareholders are the Big Fund and state capital from multiple regions, and the company itself has no actual controller. For a company like this, when prices begin to loosen, whether it will choose to defend prices or defend volume has no precedent to follow.

Seven

On the same night the earnings forecast was released, CXMT also announced a buyback proposed by Chairman Zhu Yiming: 1 billion to 2 billion yuan, to be used for employee equity incentives, an employee stock ownership plan, or a reduction of registered capital.

Based on the market value of 3.06 trillion yuan on October 16, this amounts to only 0.03% to 0.07% of market value, and about 0.8% to 1.5% of net profit attributable to shareholders for the first three quarters.

A Securities Times commentary argues that CXMT is at the peak of capacity expansion, and that putting up real money for a buyback at this time carries more weight than a company sitting on excess cash.

That argument makes sense, but it actually illustrates another point: CXMT's money must first go into factories, with shareholder returns taking a back seat.

By contrast, in July this year, after the share price of GigaDevice, which Zhu Yiming heads, fell sharply, he also proposed a buyback of the same size; that one was explicitly designated for cancellation, and he personally pledged to increase his holdings by no less than 1 billion yuan. Changxin has neither this time.

The buyback happens to fall in the quarter when quarter-on-quarter profit growth dropped from 113% to the single digits and the share price had already fallen below its first-day listing price. It shows that management has noticed the market's concerns, but the order of "capacity first" will not change. Channel: Finance and Economics. This content is published with the author's authorization; the views represent only the author and do not represent Huxiu's position. If you have any objections or complaints about this article, please contact tougao@huxiu.com. Everyone who is changing the world and wants to change the world is on the Huxiu APP[ ]]]