🟡
Monitoring
Jul 15, 2026
The Big China Memory Scare, Measured — It Misses The Tier That Matters
China's largest memory maker, CXMT, is going public in a ~$4.3B Shanghai listing and telling investors it has reached competitiveness with Samsung and SK Hynix in standard DRAM. That headline is the one the memory bears have been waiting for. But when you measure the claim instead of reading it, it hits the exact tier of memory the thesis was never long — and quietly reinforces the part that matters.
THE SIGNAL
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.
View source ↗
2026-07-15
THESIS CONNECTION
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.
WHAT TO WATCH
- CXMT's HBM3E timeline into 2027 — the genuine falsifier for the AI-memory tier, versus DDR5 catch-up which is not
- Whether the DDR5 cost gap (>30%) and yield close, or stay wide and turn "destructive when the cycle reverses"
- EUV export-control status — the bottleneck gating any move up the stack
- Whether commodity-DRAM parity pressures the low-value ASPs the Koreans have already deprioritized (minimal
⚡ ) versus any HBM progress (high thesis impact)
THESIS IMPACT: Advances the China falsifier on the commodity tier while confirming it does not reach HBM — on balance, bounds the falsifier and reinforces the server/HBM moat as the actual battleground.
FOLLOW-UPS
🟡
CXMT Hit The Tier That Matters — And The Tier Moved
🔄 Follow-up
Aug 31, 2026
That card argued the China memory scare landed in standard DRAM, not in the memory the thesis is actually long. CXMT has now begun small-scale production of HBM3E. That is the tier that matters. The card was right about where the scare was aimed and it is now overtaken on the one axis it drew the line. What makes this worth writing rather than quietly dropping is that the timeline moved, not just the product. The published roadmap had CXMT mass-producing HBM3 by the end of 2026 and reaching HBM3E in 2027, with HBM3 8-high yields reported around 25%. Producing HBM3E now skips a rung and puts them roughly a year ahead of what analysts modeled. A company that was two full tiers behind is one tier behind. But the frontier moved too, and it moved further. Samsung has scaled HBM4 sales and shipped the industry's first HBM4E samples. NVIDIA's Rubin Ultra ships with HBM4 8-Hi. The memory that matters for frontier training is no longer HBM3E — it is HBM4, and HBM4 requires 2nm or 3nm logic base dies. That is exactly where lithography export controls bite hardest and where CXMT's constraint is physical rather than a matter of engineering time. So the honest scorecard: the gap narrowed from two generations to one, faster than anyone modeled, and the remaining generation is the one China cannot buy the tools for. The original card said the scare missed the tier that matters. It no longer misses. It is one tier short of the tier that matters, and closing. Caveats that stay attached: small-scale production, single-sourced to The Information, no yield figure disclosed. Small-scale HBM3E at 25%-class yields does not move global HBM pricing. It does change what CXMT is. What would settle it: a disclosed HBM3E yield above 50%, a named Chinese AI accelerator shipping CXMT HBM3E in volume, or evidence of CXMT taping out an HBM4 base die.