China's Memory Chip Sector Extends Q3 Boom: CXMT Posts 54.3B Yuan Net Profit as Capacity Expansion and Policy Risks Accumulate
nashnova research
CXMT posted ¥54.3 billion in net profit for Q3, swinging from a ¥5.28 billion loss a year earlier to ¥132 billion in cumulative profit for the first three quarters — but capacity expansion and tightening U.S. policy are converging, putting the durability of this windfall under a double test.
How much did CXMT earn, and where did the money come from?
Q3 revenue hit ¥118.7 billion, up 613% year-on-year. Gross margin rose to 89.53% — nearly ¥0.90 of every yuan sold was gross profit.
This means → CXMT captured the commodity-DRAM supply gap left behind as Samsung and SK hynix shifted capacity toward HBM — high-bandwidth memory, the premium storage built for AI chips. Volume and prices both rose while scale effects pushed unit costs down.
Yet net profit attributable to the parent grew only 3% quarter-on-quarter, while total net profit rose about 19%. In plain terms = most of the incremental earnings flowed to minority shareholders, not to the listed entity's own bottom line.
What signals are inventory and R&D spending sending?
By the end of Q3, CXMT's inventory had grown roughly 37% from year-end 2025 to ¥39.2 billion. R&D spending for the first three quarters jumped 73% year-on-year to ¥11.1 billion.
This means → CXMT is running two tracks at once: stockpiling to grab market share and spending aggressively to close the technology gap — both burning cash at an accelerating rate.
Goldman Sachs forecasts Q4 revenue up another 4% sequentially, with gross margin at about 86% — growth slowing and margin dipping, early signs that the cycle may be approaching its peak.
Design houses are riding the wave — whose inventory is flashing red?
All seven major Chinese memory-chip design firms posted year-on-year revenue growth in Q3. The largest, GigaDevice (兆易创新), reported revenue of ¥8.73 billion, up 226% year-on-year.
GigaDevice's attributable net profit fell about 22% quarter-on-quarter, but the drop came from investment-valuation swings, not operations — stripping those out, core net profit rose 31% to ¥4.55 billion. GigaDevice holds a stake in CXMT; fair-value gains on that stake totaled ¥43 billion for the first three quarters.
GigaDevice's inventory, however, more than doubled from the start of the year to ¥6.29 billion — the starkest stockpiling warning among design houses. This reflects a channel that is building buffer stock while demand is hot; if prices turn, that inventory becomes a burden.
How long can the supply tightness last?
GigaDevice CEO He Wei said DRAM supply tightness is unlikely to ease significantly in 2027, and the SLC NAND flash upcycle — SLC NAND is a simple, fast type of flash memory — will last at least through 2027.
TrendForce projects Q4 commodity-DRAM contract prices up 10%–15% quarter-on-quarter and NAND flash up 15%–20%, supporting near-term momentum.
But Morningstar analyst Jing Jie Yu warned: if CXMT is first to commercialize the next-generation 4F² memory-cell architecture — a new design that can roughly double storage density per chip — "the risk of long-term oversupply will rise, given the rapid expansion of China's DRAM capacity."
Where does the U.S. policy shoe hang?
The U.S. Commerce Department reportedly suspended a proposal in June to place CXMT on the export-control Entity List — a short-term reprieve.
But the U.S. Congress's MATCH Act cleared the House Foreign Affairs Committee on April 22. This means → if enacted, the law would not only ban selling new equipment to CXMT but also ban servicing and providing technical support for equipment already installed — cutting off the foundation, not just the supply line.
In plain terms = the real risk is no longer "can't buy new machines" but "existing machines break down and no one can fix them" — a far deeper threat to production capacity than a simple sales ban.
What is CXMT's biggest absence?
CXMT's quarterly report mentioned progress on engineering-sample validation for its sixth-generation DRAM technology platform but made no mention of HBM.
This means → CXMT remains absent from the product segment most central to AI computing demand. Its profits come from the commodity market that Samsung and SK hynix vacated by pivoting to HBM — not from competing directly in AI computing.
This reflects the nature of the current profit explosion: CXMT has no pricing power born of a technology breakthrough — it has a cyclical windfall from a structural supply gap. Once Samsung and SK hynix refill commodity capacity, or CXMT's own expansion outpaces demand, the windfall narrows.
市场有风险，内容仅供研究参考，不构成投资建议。