CXMT Q3 2026: Revenue Up 613% to About $16.7B as AI Memory Prices Lift China's DRAM Maker

China's largest DRAM maker, CXMT (ChangXin Memory Technologies), has reportedly booked third-quarter 2026 revenue of about $16.7 billion, up 613.2% year on year, and a net profit of about $8.17 billion, against a loss in the same period a year earlier, as AI demand lifts memory prices. The figures were relayed on October 30, 2026 by Techmeme, citing Reuters, and match an aggregator summary of Bloomberg's report, which also says growth slowed during the quarter and that CXMT plans to buy back 1 to 2 billion yuan of its own shares.

We could not open the original Reuters or Bloomberg articles, and the full CXMT report had been scheduled for October 31. So this post is a "what is reported versus what is on record" read: the headline numbers, how they square with the company's earlier guidance, and why a slowing growth rate may matter more than the giant percentage.

TL;DR: reported numbers and their status

The reported revenue lands inside the range CXMT itself floated to shareholders on October 19, which is a reason to trust the headline: it is not a surprise number, it is the guided number arriving.

What the earlier guidance told us

TrendForce's October 20 write-up, which compiles Star Market Daily, Eastmoney, SCMP, Ijiwei and 36Kr, gives the best picture of the run-up. CXMT forecast nine-month revenue of 265 to 270 billion yuan, more than sevenfold year on year, and nine-month net profit attributable to shareholders of 131 to 132.5 billion yuan, against a loss of 5.3 billion yuan a year earlier. Management said its fabs in Hefei and Beijing continue to run at high utilization with full order books.

The same report flagged what was already visible inside the quarter. Eastmoney's arithmetic put Q3 revenue up 15.25 to 20.28% from Q2, but Q3 net profit up only 1 to 4% to 53.40 to 54.90 billion yuan, after Q2 net profit of 52.84 billion yuan, more than double the quarter before. Eastmoney attributes the cooling to moderating DRAM price gains. That matches the Bloomberg line that growth slowed during the quarter.

One caution on currencies: the dollar figures in headlines are conversions of yuan results, and different outlets round differently. The yuan figures are the ones to track against the filing.

Why AI is behind CXMT's windfall

Memory makers have spent 2026 in a supply squeeze created by AI. Training and inference servers consume large amounts of both high-bandwidth memory (HBM) and ordinary DRAM, and the big producers have shifted wafer capacity toward the higher-margin AI parts. That leaves less capacity for the DDR and LPDDR chips that go into servers, PCs and phones, so prices for those rose too.

36Kr, cited by TrendForce, says CXMT's DDR and LPDDR average selling prices climbed 61% and 24.46% year on year in 2025, and the effect carried into the first half of 2026, when DDR and LPDDR revenue reached 69.47 billion yuan and 78.19 billion yuan. CXMT is a beneficiary, not the cause: it is a conventional DRAM supplier selling into a market where its rivals' capacity is being pulled elsewhere.

We have covered the same chain from the supplier side: Samsung's record quarter and HBM4 contracts and Kioxia's sold-out-through-2027 guidance tell the same story from Korea and Japan. CXMT is the Chinese leg of it.

The slowdown is the real story

A 613% year-on-year jump compares against a trough, because CXMT was loss-making in the same period of 2025. The quarter-on-quarter picture is different. TrendForce's research team projected conventional DRAM contract price increases of 13 to 18% quarter on quarter in Q3 2026 and 10 to 15% for NAND, and noted that record memory prices are pushing PC and smartphone makers toward their affordability limits.

A peer shows the same turn. Chinese module maker Longsys expected Q3 profit of 3.42 to 4.32 billion yuan after 6.72 billion yuan in Q2, a sequential fall of up to 49%, according to Ijiwei as cited by TrendForce. Price gains are still positive, but the second derivative has turned.

There is also a cost side. CXMT's first-half depreciation expense rose 27.85% to 14.51 billion yuan, per its half-year report as cited by 36Kr, and it is expanding with two 12-inch fabs in Hefei and one in Beijing, each up to 100,000 wafers a month, plus new DRAM and HBM lines planned in Shanghai. More capacity brought on while prices flatten is how memory cycles turn.

Where HBM fits

Conventional DRAM profits do not make CXMT an AI-accelerator memory supplier. The HBM question is separate, and still unresolved: our earlier post on CXMT's reported HBM3E yields of 20 to 30% lays out how far CXMT is from the roughly 70% yields attributed to SK Hynix and why customer qualification has not been shown. Strong cash generation does help fund that push, which is the strategic reason the profit swing matters beyond the earnings line.

What this means for people building with AI

1. Hardware costs stay elevated. Memory is a large share of server and device bills of materials. Cloud GPU and CPU instance pricing, workstation builds and phone prices all carry it. Do not budget for cheaper RAM in the near term.
2. A second large supplier adds conventional supply, slowly. Capacity at CXMT, plus expansion by YMTC and others in a China market that is also absorbing AI and cloud demand, should ease DDR and LPDDR tightness only if demand stops growing as fast.
3. Watch for the turn. If Q4 contract prices flatten as TrendForce's tempering suggests, profits at memory makers will compress faster than revenue. That is a signal about AI infrastructure spending, not just chips; see our read on whether the AI bubble has burst and the supply-chain risk factors in Anthropic's S-1.
4. Local-AI buyers: high RAM and VRAM prices make small, efficient models more attractive, a theme in our small models comparison.

Why memory is a cyclical business, and what that implies

DRAM has always moved in boom and bust cycles. Demand jumps, producers add capacity, supply catches up with a lag of a year or more, and prices fall, often sharply. The swing in CXMT's results, from a 5.3 billion yuan nine-month loss in 2025 to guided nine-month profit above 131 billion yuan in 2026, is what the up leg of that cycle looks like when a producer starts from a low base and sells into tight supply.

Three features make this cycle different from earlier ones, and each is visible in the reporting above. First, the demand driver is AI infrastructure, which buys memory in very large, contract-backed volumes, so utilization stays high and order books stay full, as CXMT told shareholders. Second, capacity is being steered toward HBM at the leading producers, which tightens the conventional market in a way past cycles did not. Third, the buyers of conventional DRAM, PC and phone makers, have a price ceiling. TrendForce says record prices are pushing them toward affordability limits, which is how demand destruction starts.

For a producer like CXMT the practical question is therefore not whether prices are high, but whether volume growth from new fabs and new products (LPDDR6, LPDDR5X and 4F2 DRAM designs, which TrendForce lists as potential growth drivers) can offset flattening prices and rising depreciation. The Q3 numbers show the company at or near peak margin; they do not tell us how long it lasts.

Claimed versus verified: what we can and cannot say

• Verified by more than one route: the October 19 guidance ranges for Q3 revenue and nine-month profit, reported by Eastmoney and SCMP and compiled by TrendForce, and the Q3 revenue headline of about 118.7 billion yuan that sits inside that range.
• Reported, not independently confirmed here: the exact net profit of about $8.17B, the 613.2% growth rate, the buyback size and the "growth slowed" characterization, which come from Reuters and Bloomberg via summaries.
• Not yet known: the product mix, average selling prices, HBM revenue, inventory levels and Q4 guidance. Those will come from the formal filing and the call.
• Context to avoid misreading: a 613% rise is a year-on-year rate against a weak base, while the sequential profit rise guided earlier was only 1 to 4%.

How to check the numbers yourself

• Read CXMT's own disclosure when it posts on its listing venue; compare revenue and net profit in yuan, not dollars.
• Compare against the October 19 guidance ranges above.
• Look at TrendForce's quarterly contract-price tracking for DRAM and NAND to see whether price growth is still positive.
• Treat single-outlet percentages with care: 613.2% is arithmetic against a weak 2025 base.

Related reading

• Iluvatar CoreX says its Tiangai 300 GPU matches Nvidia H200: claimed vs verified
• CXMT HBM3E yields: what Reuters reports and what is verified
• Samsung Q3 2026 record profit and HBM4 contracts
• Kioxia Q2: sold out into 2027
• China's STAR50 selloff and what it means for AI builders
• Has the AI bubble burst?
• Anthropic S-1: Taiwan and HBM supplier risk factors

Figures are as reported on October 30, 2026 and may be revised in CXMT's formal report.