The crowd sees a Chinese DRAM maker grabbing 8% global share at 60% discount. I see a capital incinerator running on state subsidies and dying equipment.
Context: ChangXin Memory Technologies (CXMT) holds 8% of the global DRAM market. Apple is testing their chips for China-only iPhones. The narrative: Chinese semiconductor independence, a rising competitor. The reality: a stranded asset propped by political will.
The 8% share is a debt, not an asset. CXMT sells DDR4 at 60% below Samsung’s price. That’s not market competition — that’s dumping. And dumping requires a funding source: perpetual government bailouts.
Core: Let’s break down the order flow. CXMT’s revenue per wafer is negative. Their DDR4 yields hover around 65-70% vs 90%+ for incumbents. Every chip shipped burns cash. The only reason they survive is that the Chinese government treats them as a strategic option, not a P&L center.
Technology gap? They’re 2-3 nodes behind. Samsung and SK Hynix are shipping 1a nm DDR5 and HBM3E. CXMT struggles at 17nm DDR4. Their roadmap to DDR5 is blocked by Dutch and Japanese equipment export controls. ASML has stopped delivering DUV immersion lithography machines since 2023. CXMT relies on a finite stockpile of refurbished tools. When those break, capacity dies.
Their fab expansion — Hefei Phase II — is effectively dead. Equipment orders canceled. No new machines. The 8% share is the ceiling, not the floor.
Apple’s testing is a hedge, not a validation. Apple needs a second source for China-market memory to buffer against trade war disruptions. They don’t need CXMT’s performance — they need CXMT’s pricing and political convenience. If US export controls tighten (BIS can block Apple from using Entity List components), the deal vaporizes.
Contrarian: The market-oracle crowd sees CXMT as a bullish sign of China’s tech self-reliance. That’s emotional coverage. I see a leveraged liability: massive CapEx, negative margins, zero ability to capture AI-driven HBM demand. AI needs high-bandwidth memory — HBM is 50% of DRAM profit pool by 2025. CXMT has zero HBM capability. They’re stuck in the dying DDR4 pool.
The crowd sees art; I see a leveraged liability. Every 1% share gain costs billions in wasted capex. The unit economics don’t work without government subsidy. When the subsidy slows (and it will — local governments are running out of cash), CXMT defaults to a negative equity state.
Floor prices are illusions sold by desperate hope. The price floor for DDR4 is falling below cash cost because CXMT needs to move inventory to service debt. That’s not a competitive moat — that’s a fire sale.
Takeaway: Smart money doesn’t chase negative margin share. The real question: how long can Beijing sustain this black-swan drain? Optionality is the shield — short CXMT proxies, long Samsung. The asymmetry is clear: if CXMT dies, no impact on global supply. If it somehow survives, it still can’t compete on HBM.
