Can China's CXMT End the RAM Crisis?

Charles Kau, often called the godfather of Taiwan's DRAM industry, says Chinese memory maker CXMT could flood the market with memory products within two years. That is not the same as saying the RAM crisis is about to end. Speaking at an Ion Electronic Materials pre-IPO media briefing on November 10, 2026, Kau said CXMT would still lag in technology, but a sharp rise in output could become a serious threat to the supply-demand balance. CXMT is a DRAM manufacturer. TSMC is a logic foundry. They are not rivals on the same line.

Why is CXMT in the spotlight?

AI data centers have pulled memory supply toward HBM and server DRAM, and that shift has helped push consumer DDR4 and DDR5 prices higher. Against that backdrop, a large Chinese capacity build is being read as a possible path to cheaper RAM. According to TechNews on November 10, 2026, Kau said memory manufacturing and foundry processes are fundamentally different, and that TSMC will not enter conventional DRAM production. In the same briefing, he said CXMT has put up 10 fabs in Beijing, Shanghai and other locations. That figure should be treated as his assessment, not a verified corporate investment list.

Can extra output bring prices down?

Mirror Daily quoted Kau directly on November 10, 2026. "Two years. In two years, it will come out in large volume," he said. He added that the technology would still lag, but "once the volume gets large, it becomes frightening." He also said that once output surges, oversupply can follow. He did not give a firm date for when the price rally would reverse; the two-year horizon was an industry observation based on the scale of current expansion.

Kau did not cite a 600,000 wafers-per-month figure. A Bank of Korea report, as relayed by Bloomberg on September 4, 2026, said CXMT's capacity is about 300,000 wafers a month today and could rise to as much as 600,000 in 2028, an increase of about 300,000. In the same period, the report said South Korean capacity from Samsung and SK Hynix could increase by about 600,000 wafers a month. The bank also said CXMT accounts for about 15% of nominal global memory capacity but only about 8% of actual shipments, and that yield and technology constraints could keep real output below nameplate capacity. These are research estimates, not a confirmed official plan.

Even if Chinese domestic demand absorbs part of the new supply into 2027, leftover volume could still free up chips from other suppliers and weigh on global prices. That still does not prove prices will fall on a two-year clock. Customer qualification remains a hurdle, and some PC makers have already been testing a wider mix of memory sources, including Chinese suppliers.

How serious are the technology gap and sanctions?

For Kau, volume is not the only question. TechNews reported him saying DRAM has reached around 6F², but no company has truly mass-produced 4F², in part because capacitors and leakage become harder at that density. Even Samsung and Micron have not gotten there yet, he said. CXMT said in early November that it had validated engineering samples of its sixth-generation 4F² platform and planned to send customer samples by the end of 2026. That is not proof of mass production. Chinese firms also face limited access to advanced EUV tools. A large mature-node build can still shake the market, but bits per wafer may lag the leaders.

Quality and customer approval are a separate gate. Big brands usually run long qualification cycles before putting Chinese DRAM into their products, and export controls can complicate that path. CXMT's recent results show how much the current upcycle has helped China's largest DRAM maker, but they do not settle the yield and technology questions.

What about HBM and demand?

Mirror Daily reported that Kau said CXMT has started trying HBM-related technology, but it is still unclear whether it can enter the market. That matters because much of AI demand is concentrated in HBM and server memory, and the consumer DDR4 and DDR5 squeeze is partly a side effect of that shift. At the Nikkei Forum Global Management Dialogue 2026 in Tokyo on November 11, 2026, TSMC senior vice president and deputy co-COO Kevin Zhang said the company is seeing insatiable demand for any chips it can make to power AI. That leaves open the possibility that demand could still absorb new supply two years from now; it does not, by itself, prove where memory prices will go. Apple has also warned that memory costs may keep product prices elevated.

Where does TSMC fit in?

CXMT is not TSMC's foundry rival. In reports published on November 7, 2026 by The Information and Economic Daily, TSMC said it does not comment on market rumors and that its core business is foundry work, meaning the supply of logic ICs. That is not a pledge to build a standalone DRAM factory. A 2025 TSMC symposium paper described a dense, SRAM-like cache embedded in the upper layers of a logic chip. That research is not a plan to replace commodity DDR5 modules.

Point	The bullish case	The caveat
Capacity	A large volume within two years is possible	This is Kau's view; the 10-fab figure is not a verified corporate list
2028 estimate	CXMT could rise from 300,000 to as much as 600,000 wafers a month	Bank of Korea estimate; actual output may lag nameplate capacity
Pricing	Oversupply could pressure prices	Demand, the China market, and customer approval will decide the outcome
Technology	4F² and HBM remain unproven in volume	Sanctions and a yield gap still stand in the way

In short, CXMT is a plausible capacity candidate for easing the RAM crunch, but the technology gap, sanctions, qualification, and AI demand mean the story is far from settled. The answer to whether it can end the crisis is not a clean yes or no. It is a two-year stress test.