The Big China Memory Scare, Measured — It Misses The Tier That Matters

CXMT — China's #1 and the world's #4 DRAM maker by capacity, with global share roughly doubling to ~8% over the past year — filed to raise about 29.5 billion yuan (~$4.3B) on Shanghai's STAR Market, the largest mainland IPO since 2022. Its prospectus claims competitiveness in mainstream DRAM, and independent analysis (SemiAnalysis) confirms its DRAM average selling price sat only 5-10% below the three leaders in Q1 2026, with LPDDR5X reportedly matching flagship mobile specs at 10,667 Mbps. But the same analysis shows that near-parity is a pricing artifact of the shortage, not cost or technology: CXMT's per-bit cost on DDR5 is more than 30% higher than the leaders, its yields are low, and analysts peg it two to three years behind on process. Its revenue is ~66% LPDDR and ~32% DDR — commodity and mobile — with overseas sales outside Hong Kong at just 3%. In HBM, where the AI-memory demand actually concentrates, CXMT is under 2% of capacity, running an estimated ~25% yield on HBM3, with HBM3E mass production targeted only for 2027. EUV export restrictions remain the binding bottleneck. The consensus read across Korean and Western analysts is that the impact on Samsung and SK Hynix is limited and the threat is overstated.

This is the most-hyped falsifier of the memory thesis advancing — from lab and testing to IPO, mass production, and apparent ASP parity — and it's exactly the kind of confirming-your-fears headline that deserves the hardest scrutiny. Measured, it bounds rather than breaks the thesis, three ways. The "parity" is shortage-driven pricing, not a closed cost or technology gap. It is concentrated in commodity and mobile DRAM — the low-value tier the Korean leaders are actively vacating. And HBM, where the Charizard/AI-memory thesis lives, is essentially untouched and years away. The deeper point cuts the other way entirely: the reason CXMT has room in commodity DRAM at all is that Samsung, SK Hynix and Micron are shifting capacity into HBM and server DRAM — the same supply defection that tightens the high-value tier the thesis is long. So China filling the commodity gap and the AI-memory squeeze are two sides of one capacity shift. The falsifier's real trigger is HBM, not DDR5.