CXMT Fear Is a Mirage. Korean Chips Still Lead

The Market Got the Signal Wrong

On October 13 and 14, Korean stocks sold off sharply after reports emerged that China’s ChangXin Memory Technologies (CXMT) had advanced its next-generation DRAM cell design — a so-called 4F square structure. The narrative that followed was immediate and visceral: Chinese memory is closing the gap, Samsung and SK Hynix face a new rival, and foreign capital might flow east.

It turned out to be a mirage.

In Shanghai, where the story actually broke, no one was alarmed. Liu Yang, deputy director of research at Shanghai Xinwan Hongyuan Securities, told Pulse that CXMT’s 4F square announcement was precisely what analysts had expected months ago. It was not a surprise. It was not even particularly newsworthy.

More striking was his assessment of what this actually means for the Korean giants. “We in this research team view rising Korean memory stocks as a catalyst for Chinese memory stocks,” Liu said. “When Korean prices go up, that becomes a positive signal on our side.”

The logic is straightforward but often missed in Seoul. CXMT’s progress is real, but it is also expensive. Every new process node China unlocks requires more capital expenditure, more fab capacity, and more competitive pressure on pricing — which lifts the entire sector, including the established players in Korea. China’s push does not erase Korean dominance; it funds it.

The market proved the point almost immediately. When Korean stocks fell on the 13th, CXMT itself dropped 2.79 percent the same day and 2.11 percent the next. The correlation ran in the opposite direction of what headlines suggested.

The Real Gap Is Wider Than You Think

The 4F square announcement may sound dramatic, but it operates inside a far tighter constraint than most investors appreciate: U.S. export controls on advanced lithography equipment.

South Korea and Taiwan face no such restrictions. Chinese memory makers are effectively quarantined from the newest EUV (extreme ultraviolet) tools, which means they cannot simply copy the path that Samsung and SK Hynix have followed. Instead, they must invent around it.

Liu’s baseline estimate for the technology gap is telling. He sees roughly three years to catch up on HBM — high bandwidth memory, the critical component for AI training chips. For standard DRAM, two years. For NAND flash, possibly less. Those figures sound compressible. They are not.

“I think we’re in a phase where the gap is being maintained, not closed,” Liu said. “China will find alternative solutions — non-HBM storage architectures, hybrid bonding stacks, multi-layer integration — but these are compensations, not replacements.”

That distinction matters. Chinese engineers are not sitting idle. Firms like YMTC (Yangtze Memory Technologies) and CXMT are actively pursuing alternative process routes, including advanced DUV (deep ultraviolet) multi-patterning and 3D stacking techniques. But each workaround costs more, yields less, and moves slower than the direct path the West controls.

The result is a structural moat, not a temporary delay. And until U.S. policy shifts — a scenario Liu said is unlikely given current bipartisan consensus — the gap will persist at roughly the level he described.

Phase Two Has Already Started

Beyond the chip wars, Liu offered a broader framework for thinking about where Chinese markets actually stand. His view is that the broad-based rally of early 2025 — when AI-related stocks surged 30 to 50 percent across the board — is over. We are now in what he calls Phase Two: a selective market where fundamentals matter more than narratives.

“Phase One was every sector rising together,” he said. “Phase Two is about winners and losers. The market now demands profitability, not just pipeline.”

That shift has immediate implications for how Korean investors should think about Chinese memory stocks. The companies that will survive the transition — CXMT, YMTC, the listed AI players — are already known. The ones that won’t are still raising money and making announcements.

This distinction also applies to the generative AI wave sweeping Chinese exchanges. Several high-profile models, including Moonshot AI’s Kimi and DeepSeek, are preparing for domestic listings rather than pursuing U.S. or Hong Kong venues. Their reasoning is pragmatic: valuation expectations in China are more aligned with current revenue trajectories, and the regulatory environment for financial issuance has shifted favorably toward home-market listings.

But Liu warned that listing alone does not guarantee stability. Zhipu AI, which listed earlier this year, saw its market capitalization collapse from trillions to hundreds of billions of yuan — a reminder that subscription-based software businesses face structural headwinds in China, where users are reluctant to pay for digital services. Hardware plays, by contrast, tend to hold their value better in a down cycle.

What Korean Investors Should Actually Watch

For Seoul-based portfolio managers, the practical takeaway from Liu’s analysis is narrower than the panic suggested. Three items deserve attention:

First, Huawei’s recent disclosure of a new physical circuit design theory — the “Tau Scaling Law” — which could reshape how Chinese fabs approach performance gains without relying on smaller nodes. Related equipment suppliers are the play here, not the model providers.

Second, the clear leaders in each Chinese memory segment. CXMT for DRAM, YMTC for NAND, Alibaba and Tencent for the AI software stack. In Phase Two, concentration matters more than diversification.

Third, Hong Kong-listed Chinese stocks. Liu noted that while mainland A-shares outperformed in 2025, Hong Kong’s H-shares remain far more sensitive to U.S. Treasury yields — and yields are likely to fall. That shift could open a new window for capital rotation into the offshore Chinese market.

The Bigger Picture

The Korean market’s reaction to CXMT reveals more about investor psychology than semiconductor economics. When a Chinese firm announces a technical advance, the first instinct in Seoul is defensive. But the data from Shanghai — the indifference to the 4F square news, the inverse correlation between Korean and Chinese memory prices, the explicit call to buy Korean stocks — points in the opposite direction.

China’s memory ambition is real. It is also constrained, incremental, and increasingly expensive. Every step forward buys Samsung and SK Hynix more time, more margin, and more reason to invest in the next generation of HBM and advanced DRAM. The race is not over. But the starting line has not moved.

What has moved is the market’s ability to read the signals. And right now, Seoul is listening to the wrong frequency.