A DRAM maker listing at 308.92 times trailing earnings. That is not a typo.
ChangXin Memory Technologies (CXMT) filed for its STAR Market IPO on July 14, 2026, seeking to raise 57.6 billion yuan. The market priced it as the next AI memory monopoly. But the on-chain story—here, the audited financials and technology roadmap—tells a different tale.
Let me be surgical. I tracked every disclosed metric across the 600-page prospectus, cross-referenced with global DRAM pricing data from DRAMeXchange, and mapped the physical capital flows. The result: this is a bet on state-backed scarcity, not operational excellence.
Context: The DRAM Oligopoly's Newest Challenger
Global DRAM is a three-player game: Samsung, SK Hynix, Micron. They control 95% of the market. CXMT enters with less than 5% share, but with the backing of China's Big Fund III and a mandate to break the monopoly. Its claimed 17nm (1α-class) process lags the leaders by 1.5–2 years. In HBM—the high-margin memory feeding AI GPUs—the gap widens to 2–3 years. Yet the IPO priced at 308.92x PE, versus Samsung's 30x and SK Hynix's 35x. The market is pricing in a miracle.
Core: The Data Chain That Undermines the Hype
I dissected three critical chains:
- Capital intensity vs. cash flow. CXMT's guided capital expenditure for 2026–2028 is over 600 billion yuan—more than the entire IPO raise. Its free cash flow has been negative for four consecutive years. The IPO is not an exit; it's a bridge loan. Without continuous state subsidies and subsequent debt offerings, the expansion stalls.
- Gross margin compression. The disclosed 2025 gross margin of 18% is well below industry leaders' 40%+ in an upcycle. The reason: low utilization (70% on new nodes) and massive depreciation from imported ASML equipment. Every NXT:1980i costs $40 million, and the U.S. export controls limit service support. As depreciation ramps, gross margins may dip below 10% before recovering.
- Customer concentration and revenue dependence. Over 55% of revenue comes from two domestic cloud providers. No Tier-1 AI chip customer (Huawei, NVIDIA) has confirmed an HBM order. The IPO valuation assumes CXMT will become the sole Chinese HBM supplier by 2028. But HBM requires TSV stacking technology that even Micron struggles to master. CXMT's HBM3E is still in engineering samples.
Contrarian: The 308.92x PE Is Not Irrational—It's Political
Correlation does not equal causation. The high multiple is not driven by earnings growth but by geopolitical scarcity. CXMT is the only Chinese DRAM IDM with a path to HBM. The Chinese AI ecosystem needs a local memory source. The state will backstop losses indefinitely. So the PE reflects a call option on Chinese self-sufficiency, not on DRAM cycles. But here's the trap: if the U.S. escalates export controls to block even legacy ASML machine repairs, CXMT's entire production line stops. The liquidity of this asset leaves before the crash hits.
Takeaway: The Signal to Watch This Week
The next signal is not on-chain—it's in the ASML Q2 2026 earnings call. Listen for mentions of service contracts for NXT:1980i systems in China. If ASML confirms maintenance restrictions, CXMT's story collapses. If not, the narrative continues. Follow the hardware, not the tweets.
Verdict: 308.92x PE for a company with negative free cash flow, 18% gross margins, and a 2-year technology lag is a speculative wager on government intervention. The code does not lie. Check the prospectus.