China's DRAM Surge Is Rewriting the Chip Rules
Chinese memory maker CXMT is seizing DRAM market share as AI demand strains global supply — and US export controls are unintentionally clearing the path. The implications for Samsung, SK Hynix, and the semiconductor order are profound.
The numbers no one expected
CXMT, China’s state-backed memory chip champion, reported something startling about its third quarter this year. Revenue will come in between 114.7 billion and 119.7 billion yuan — roughly $16 billion. That is up 589 to 619 percent from the same period a year earlier. Net profit will land at 53.4 billion to 54.9 billion yuan. Last year, the company lost money. This year, it is projecting a surplus roughly equivalent to what Samsung Electronics spends on R&D every nine months.
The full-year picture is steeper still. Through the first three quarters, CXMT expects cumulative revenue of 265 billion to 270 billion yuan, up 726 to 742 percent from last year. The company went public on the STAR Market last July and raised 66.6 billion yuan — a sum that immediately made it China’s largest listed semiconductor company by market capitalization. As of mid-October, its market cap sat at roughly 3.36 trillion yuan, about $465 billion.
The market is pricing in acceleration. CXMT has also announced a share buyback of up to 8 billion yuan — enormous for a Chinese-listed firm, where buybacks typically top out at 200 million yuan.
Share gains that rewire the map
Counterpoint Research places CXMT’s DRAM market share at 10 percent in the second quarter of this year, up from just 4 percent a year ago. That is a six percentage point jump in twelve months. Samsung holds 38 percent. SK Hynix commands 25 percent. Micron sits at 24 percent. CXMT is fourth. The gap to the leaders remains wide, but the trajectory is what matters.
This is not normal for a market that has historically moved in single-digit increments per quarter. It is also not normal for a Chinese company to be expanding in a sector where US export controls were supposed to freeze technological progress.
How export controls became a two-way street
The conventional wisdom in semiconductor policy circles is that export controls on advanced lithography and equipment keep China stuck at older process nodes. That has held true for logic chips. But memory is different. DRAM manufacturing, especially for commodity grades used in consumer electronics and industrial applications, does not depend on the most extreme EUV lithography. Older nodes can produce functional, cost-effective memory at scale. And CXMT has been pouring state-capital infrastructure into precisely those capabilities since Beijing identified memory chips as strategic enough to subsidize beyond reason.
What the controls did instead was force China to build domestic supply chains for products it previously imported. Now CXMT is producing DRAM that meets the needs of a vast downstream market — servers, networking gear, automotive electronics, consumer devices — much of it in countries that cannot or will not follow Washington’s procurement restrictions.
The irony is structural. Export controls were designed to slow China’s chip industry. They may have done the opposite for memory, by closing off alternative sources and forcing buyers in the Global South toward the only viable alternative: Chinese-made DRAM. That is a market Samsung and SK Hynix assumed would remain theirs by default.
AI demand is the accelerant
The CXMT earnings narrative points directly to the AI boom. Demand for high-bandwidth memory and general-purpose DRAM has surged beyond what the established players can fill. Supply has not caught up with capacity expansion, and prices have followed.
CXMT’s management acknowledged the imbalance publicly, saying AI computing demand drove prices higher while output lagged. The company has been ramping wafer capacity and improving yield, which is now showing up in margin expansion. A company that posted a 5.3 billion yuan loss over the same period last year is projecting over 130 billion yuan in cumulative profit this year. That is not incremental improvement. That is a regime shift.
Who wins, who loses
Samsung and SK Hynix still dominate. But dominance is not permanence. Every percentage point CXMT takes from the commodity DRAM segment erodes the volume advantage that lets the Korean firms invest in next-generation memory — HBM, DDR5, the products that actually power AI data centers. If CXMT captures enough volume at the lower end, the economics of the Korean strategy tighten. They can still lead in advanced memory. But margins compress faster when a competitor absorbs the base-load demand at lower cost.
Micron faces the same dynamic, though with an added complication: it is an American company operating under stricter export compliance constraints. CXMT does not carry that burden in markets where US rules do not apply. That is a competitive asymmetry that Beijing will exploit.
The buyers in India, Southeast Asia, the Middle East, Latin America, and Africa are the real winners in the near term. They get DRAM at prices that fall competitors’ margins. For now, they gain access to capacity that Western controls did not deny them.
What happens next
CXMT’s buyback announcement signals confidence, but also the need to sustain investor interest in a star-market environment where expectations run ahead of fundamentals. The company is trading on momentum generated by state backing, policy tailwinds, and a favorable demand cycle. Any of those three could shift.
The more consequential question is whether the export control architecture can adapt. If Washington expands restrictions to cover mature-node memory equipment — not just leading-edge logic — CXMT’s path forward narrows. If it does not, the company will continue gaining share in the exact segment where volume most protects the incumbents’ ability to fund innovation.
Samsung and SK Hynix are responding. Both have accelerated capacity investment in HBM and advanced DRAM precisely to stay ahead of the commodity flood. The strategy is sound, but it assumes CXMT stays confined to older nodes. The data from this quarter suggests that assumption is already under pressure.
The global DRAM market is no longer a three-firm club. It is becoming a contest between an entrenched Korean-American oligopoly and a state-subsidized Chinese entrant with accelerating momentum. The side that loses volume loses the funds to win the next generation. CXMT has just proven it can take volume.