CXMT forecasts $19.5B profit on AI DRAM demand — shares fall 6.5%
The AI memory shortage just produced its largest profit number on record — and China's DRAM champion still couldn't hold a rally on it.
ChangXin Memory Technologies told the Shanghai Stock Exchange it expects net profit of 131 billion to 132.5 billion yuan for the first nine months of 2026 — roughly $19.58 billion — reversing a net loss of 5.28 billion yuan a year earlier, on revenue of about 265 billion to 270 billion yuan, and the market sold the news anyway: CXMT shares fell 6.5% on Monday, according to CNBC, which translated the filing. The company attributed the swing to demand for AI computing power that has surged since the second half of 2025, tightening the global DRAM market into supply shortages and significant price increases. That is the whole AI hardware cycle in one sentence: the models need memory, memory got scarce, and the one Chinese fab able to ship standard DRAM at scale converted scarcity into a nine-month profit that exceeds what most profitable chipmakers earn in several years.
Strip the nine-month guidance down to the third quarter and the pace is even steeper. Seoul Economic Daily calculated the quarter on its own: revenue of 114.7 billion to 119.7 billion yuan, a jump of 589% to 619% from a year earlier, with quarterly net profit of 53.4 billion to 54.9 billion yuan. Implied net margins near 50% for a company that was posting losses inside the last two years — this is what a genuine price upcycle looks like on a balance sheet, not a forecast about future AI demand but cash from DRAM that datacenter buyers already paid for. The scale matters beyond the yuan figures: Counterpoint Research data cited by CNBC puts CXMT at 10% of global DRAM revenue in the third quarter, double the 5% it held a year ago, which makes the former sanctions-list newcomer the fourth-largest memory supplier in the world, sitting alongside Samsung, SK Hynix and Micron rather than behind them. We covered the first leg of this in August — China's CXMT flips to $11.5B profit as memory prices ride the AI wave — and the third quarter says the upcycle has accelerated rather than cooled. For scale: Chinese trade publication AET noted that CXMT held less than 1% of global DRAM revenue as recently as 2023, and the capacity behind the climb comes from its expanded Hefei and Beijing production lines, according to CNBC's reporting.
The part of the story that scared the market is not the earnings; it is the technology claim attached to them. Earlier this month, CXMT said it will release server DRAM this year using a 4F² cell structure that cuts cell area by about a third compared with the 6F² design the industry runs on, according to Seoul Economic Daily. Denser cells mean more bits per wafer, which means lower cost per gigabyte at exactly the moment buyers are desperate — and Korean memory stocks reacted accordingly: Samsung fell 17.8% and SK Hynix 24.3% over the past week, their worst stretch since the 2008 financial crisis, in Seoul Economic Daily's framing. A profitable CXMT printing fat margins on this quarter's prices is one thing; a profitable CXMT telling the market its next process node undercuts the incumbents' cost structure is a repricing.
The demand behind the numbers is physical, not speculative. Seoul Economic Daily's investor briefing, summarizing foreign reporting, said Apple's John Ternus has called the memory shortage an absolute crisis and raised the possibility of buying Chinese-made parts, while Elon Musk wrote on X on October 15 that China would produce enormous volumes of memory by 2028. Both statements point at the same bottleneck: server DRAM and high-density memory for inference racks, the layer of the AI buildout that cannot wait for anyone's frontier roadmap. The geopolitical read is where this gets interesting — a CXMT clearing net margins near 50% on standard DRAM shows that China's memory self-sufficiency no longer requires reaching the frontier, only the frontier's hunger for memory outpacing the incumbents' expansion. Our August coverage put the ceiling on the story: US export controls still wall off the most advanced high-bandwidth memory from CXMT. This quarter says the floor beneath that ceiling keeps rising fast.
The skepticism deserves equal space, because the 4F² claim is exactly where this story could break. Rosenblatt analyst Kevin Cassidy argued that given CXMT's actual process capabilities, its 4F² cell could in practice be around 34% larger than rivals' existing products — the geometry on paper does not automatically survive the process node it is manufactured on — and said the company may prove to be a paper tiger. That is the right test to hold this quarter's numbers against: price-driven margins flatter any fab with wafers to sell, but a durable CXMT requires the cell-size advantage to show up in real cost-per-bit when the DRAM cycle turns. The shares falling 6.5% on a record forecast is the market saying it wants to see that proof before paying up — the same pattern we flagged when TSMC's record quarter fell short of the market despite being the foundry's best ever.
What to watch: whether the 4F² server DRAM actually ships this year as announced, where contract DRAM prices settle after the current shortage peak, and whether CXMT's HBM efforts — the company is pushing into advanced processes and high-bandwidth memory — get it back into the conversations US export controls still cut it out of.
Is a massively profitable CXMT good news for AI supply, or the signal that the memory cycle is at its top? Write to us via our contact page.
