U.S.-China Relations
NextFin News - ChangXin Memory Technologies, China's largest DRAM maker, swung to a preliminary net profit of 131 billion yuan to 132.5 billion yuan ($19.6 billion to $19.8 billion) in the first nine months of 2026, the company said on Friday, reversing a 5.3 billion yuan loss in the same period a year earlier as a global memory shortage sent chip prices soaring.
Preliminary revenue for the nine-month period is expected to reach 265 billion yuan to 270 billion yuan, up from 32.1 billion yuan in the first nine months of 2025 - a more than eightfold jump. The announcement marks the latest evidence that the AI-driven memory boom has moved beyond a handful of high-end suppliers and into the broader DRAM market, where pricing power has returned with a vengeance. For a company that was still losing money twelve months ago, the swing is not just a recovery; it is a statement that the memory industry's center of gravity is shifting, and that a Chinese challenger has arrived at the profit table while the incumbents are still debating how to split it.
The Numbers: A Loss Reversed, Revenue Multiplied
The magnitude of the reversal is best understood against CXMT's own recent trajectory. The nine-month profit guidance alone exceeds the 77.61 billion yuan in net income the company reported for the first half of 2026, when revenue jumped 874% to 150.31 billion yuan ($22.4 billion) and beat its own sales guidance of 110 billion yuan to 120 billion yuan. In the first quarter, revenue was 50.8 billion yuan, up 719% from a year earlier, with net profit of 25 billion yuan against a 1.6 billion yuan loss in the same quarter of 2025. Full-year 2025 revenue was just 15.4 billion yuan - meaning CXMT is on track to multiply its annual sales many times over in a single year.
The sequence tells a story of acceleration, not a one-off spike. First-quarter revenue of 50.8 billion yuan implies second-quarter revenue of roughly 99.5 billion yuan, and the third quarter - implied by the nine-month guidance - adds another 115 billion yuan to 120 billion yuan. Each quarter has been larger than the last. That is the signature of a company riding a price wave that has not yet crested, and it is why the profit figure, striking as it is, may understate the run-rate the company enters 2027 with.
The market has already rewarded the turn. CXMT's shares closed at 45.02 yuan on the Shanghai STAR Market on Friday, up 3.28% on the day, exchange data show. The stock is down from an intraday high of 61.80 yuan reached earlier in its listing history but remains far above its July 27 issue price of 8.66 yuan. The company raised 57.9 billion yuan in the largest IPO in the STAR Market's history, and closed its debut up 465.8% with a market capitalization of roughly 3.3 trillion yuan - making it China's most valuable listed company, ahead of Tencent Holdings and Alibaba Group. Goldman Sachs has valued CXMT at around 10 times its projected 2027 earnings, a cheaper multiple than the company's international rivals.
Why Prices Rose: The Transmission Mechanism Behind the Surge
CXMT's windfall is not an isolated company story. It is the financial expression of a supply-demand imbalance that has tightened across the entire DRAM industry, and the chain of transmission runs straight through the AI data center. Research firm TrendForce estimates global DRAM revenue hit about $154 billion in the second quarter of 2026, up 59.5% from the previous quarter - the kind of sequential acceleration that usually appears once a decade, not once a cycle. CXMT's own share of that market reached 9.5% in the quarter, up from 7.6% in the prior period, with its DRAM revenue nearly doubling to roughly $14.6 billion.
The mechanism works in three steps, and understanding them explains why CXMT - not just the AI darlings - is capturing the upside. First, hyperscalers building AI clusters pulled forward orders for high-bandwidth memory (HBM) and server DRAM, which consume far more wafer capacity per unit of revenue than standard memory. Second, the three incumbents - Samsung Electronics, SK Hynix, and Micron Technology - responded by reallocating fab capacity toward these higher-margin AI chips rather than maximizing total bit output. Third, that reallocation shrank the supply of standard DRAM available to the rest of the market, and with demand for PCs, phones, and enterprise servers holding up, the price of the leftover supply rose. CXMT, which still sells large volumes of commodity DRAM to customers priced out of or underserved by the Korean and American camp, is the marginal beneficiary of a shortage it did not create.
SK Hynix told investors on its October earnings call that its HBM, DRAM and NAND capacity is "essentially sold out" for 2026.
The price data confirm the squeeze. Contract prices for server DRAM rose 64% in the second half of 2025, according to TrendForce. Samsung raised the price of its 32GB DDR5 modules to $239 from $149 in September, a 60% increase, while DDR5 contract pricing more than doubled to $19.50 per unit from around $7 earlier in 2025. Counterpoint Research expects DRAM prices to climb another 30% in the fourth quarter and a further 20% in early 2026, which would put DDR5 64GB server modules at roughly twice their early-2025 cost by the end of next year. CXMT sells its 32GB DDR4 modules for about $138 - less than half what Samsung and Micron charge - which is exactly the price point that wins share when procurement budgets are stretched and when export controls steer Chinese buyers toward domestic suppliers.
The Second-Order Effect: Who Really Pays for the Shortage
The first-order effect of the memory shortage is obvious: chipmakers make more money. The second-order effect is who absorbs the cost, and that is where the story gets less comfortable for the rest of the technology chain. Memory is a component that shows up in everything from smartphones to servers to cars, so a doubling of DRAM prices is a tax on the entire electronics supply chain. Device makers facing $239 for a 32GB DDR5 module that cost $149 three months ago have three bad options: pass the cost to consumers and risk demand destruction, absorb it and compress their own margins, or redesign products with less memory and weaker performance.
That dynamic is already visible in the equity market's split reaction. Micron Technology shares fell 3.76% on Friday while Samsung Electronics rose 2.62%, a divergence that suggests investors are weighing two competing reads: the incumbents are enjoying record pricing, but they are also the most exposed if the shortage triggers a demand break or accelerates the shift of customers toward cheaper alternatives like CXMT. The memory rally, in other words, contains its own contradiction - the higher prices go, the more incentive customers have to qualify a second source, and the more runway CXMT has to convert a cyclical windfall into permanent share.
Cyclical Windfall or Structural Breakthrough?
Here is the question investors should ask: is CXMT's profitability a cyclical gift from a tight market, or has the company crossed a structural threshold that will outlast the cycle? Getting this wrong flips the conclusion, because cyclical profits revert and structural gains compound. The honest answer is that both forces are at work, and they point in opposite directions.
The profit itself is overwhelmingly cyclical. CXMT's revenue multiple came from price as much as volume, and DRAM is a notoriously cyclical industry where today's shortage becomes tomorrow's glut. The industry has lived through this movie before: the 2017 memory supercycle lifted all three incumbents to record margins, only for prices to collapse when new capacity arrived in unison. History says the current margin expansion will revert - and the reversion, when it comes, will be sharp, because DRAM is a homogeneous product and price is the only differentiator that matters in a glut.
But the capacity story underneath is structural, and it is the part the market is underweighting. SemiAnalysis estimates CXMT will end 2026 with roughly 350,000 wafer starts per month, only about 25,000 below Micron's estimated 385,000. By wafer capacity alone, CXMT is approaching the position of the world's third-largest memory supplier. The company currently operates three 12-inch DRAM fabs across Hefei and Beijing, and its new Shanghai fab is expected to roughly double output to around 600,000 wafers per month as it ramps. SemiAnalysis projects CXMT will reach 500,000 wafer starts per month by the end of 2028, accounting for about 17% of global DRAM supply, up from 11% in 2025.
Capacity without technology is a trap, which is why the node progression matters. CXMT's fifth-generation DRAM platform entered mass production in September, and the company has said it plans to introduce DDR5 server-memory products by the end of 2026. Its 24Gb LPDDR5X chips are already in mass production, packing 50% more capacity per die than the previous generation. That is a technological narrowing that did not exist five years ago, when CXMT was dismissed as a producer of low-end commodity parts sold at a discount. The valuation gap reflects this: at roughly 10 times projected 2027 earnings, CXMT trades at a discount to rivals whose growth is capped by the same capacity constraints that are lifting CXMT.
The cyclical leg says profits will fall when the shortage ends. The structural leg says CXMT will keep more of the market even after prices normalize, because it will have the capacity and the process node to serve customers - particularly in China - who previously had no acceptable alternative to the Korean and American incumbents. The profit is cyclical. The market share is structural. Both are true, and the investment question is which one the market is paying for.
The Counter-Thesis: Commodity Trap, Geopolitics, and the Margin Test
The strongest argument against the structural view is simple and uncomfortable: CXMT still makes money the old-fashioned way, by selling cheaper commodity DRAM into a tight market. Samsung, SK Hynix, and Micron collectively command more than 90% of global DRAM supply, and CXMT's 9.5% share - while rising - leaves it a distant fourth. Its most advanced products, including LPDDR6 memory for smartphones, are still in client evaluation rather than volume production. If the company cannot move up the value chain before the cycle turns, today's profits will look like a one-time price spike rather than a new baseline, and the valuation premium will evaporate with the cycle.
There is also the geopolitical overhang. The U.S. Department of Defense has placed CXMT on a blacklist, and while the company has said the designation will not affect its day-to-day operations, export controls on advanced semiconductor equipment remain a binding constraint on how quickly CXMT can close the technology gap. A company racing to add leading-edge capacity while restricted from buying the most advanced lithography and deposition tools is running with one hand tied behind its back. The incumbents, by contrast, are not standing still: Samsung plans to add 110,000 wafer starts per month in 2028, and Micron plans 115,000, according to SemiAnalysis - more than CXMT's projected additions in the same year, which means the capacity race is a race the leaders are still winning on the margin.
The falsifying signal for the structural-breakthrough view is concrete and observable. If CXMT's gross margin falls back toward pre-cycle levels - below 40% - within two quarters of DRAM contract prices normalizing, the profitability was purely cyclical and the structural case fails. Conversely, if the company holds gross margins above 60% while continuing to gain share after the shortage eases, the structural case is confirmed. A secondary signal: if CXMT's DDR5 server products win meaningful design wins with major Chinese cloud providers by mid-2027, the technology-narrowing thesis is validated; if they remain stuck in evaluation, the commodity-trap thesis wins.
What Comes Next: Three Horizons
In the short term, the stock will trade on sentiment around the preliminary numbers and the broader memory complex. CXMT is due to report fuller results later this month, and any confirmation or revision of the preliminary guidance will move the shares. The broader complex remains split - Micron down, Samsung up - which tells you the market has not settled on who wins the next leg of the cycle.
Over the medium term, three things matter. First, whether CXMT's full-year results confirm the preliminary guidance and whether management's outlook for 2027 assumes prices holding or rolling over. Second, whether its LPDDR6 and DDR5 server products reach volume production on schedule, because revenue from advanced nodes is the only durable hedge against a commodity price decline. Third, whether DRAM contract prices hold through the first half of 2027, as suppliers including Samsung have guided with price increases already requested for mobile DRAM of more than 80% versus the first quarter.
In the long term, the question is whether China becomes a second global DRAM production base. Citrini Research projects that China's total DRAM capacity will reach about 1.41 million wafer starts per month by 2030, with CXMT alone at roughly 950,000. If that build-out happens on schedule, the industry's pricing power - and the incumbents' margins - face a permanent reset, because a market that was an effective oligopoly becomes a contested four-player field. That is the real stake in CXMT's profit swing: not the 131 billion yuan, but the signal that the oligopoly is no longer unchallenged.
The base case is that CXMT reports record full-year profit and the memory rally extends into early 2027, but with rising volatility as investors debate how much of the gain is cyclical versus structural. The upside case is that CXMT's new server products win design wins, its capacity build-out continues without equipment restrictions tightening further, and the shortage lasts longer than the consensus expects. The downside case is that the DRAM shortage breaks sooner than expected - perhaps as soon as new capacity from all four players arrives in 2027 - prices roll over, and CXMT's valuation, at roughly 10 times projected 2027 earnings, compresses along with the cycle.
The memory cycle has made many companies rich and then poor. CXMT's profit is real, but the test is not the profit - it is whether the profit survives the turn.
Explore more exclusive insights at nextfin.ai.
