The code didn't crack. The 8.7 billion yuan extra raise did.
ChangXin Memory Technologies (CXMT) — China's only real DRAM player — just exercised its IPO over-allotment option in full. That's not a footnote in a prospectus. That's a desperate, beautiful tell.
Here's the setup. CXMT priced at 8.66 yuan, raised roughly 8 billion yuan. Then the underwriter, CICC, triggered the greenshoe. Full exercise. An extra ~870 million yuan. Total shares now at 6.7884 billion. Market cap at ~58.8 billion yuan. And here's the kicker — CICC didn't buy a single share from the secondary market.
The greenshoe mechanism exists for one reason: price support. Underwriters buy back shares to stabilize price. They didn't. Price held above issue. That's not noise, that's conviction.
We didn't need a speech. The absence of intervention is the signal.
But nobody's asking the real question. What's this money actually for?
It's for machines. It's for the tools of fabrication — and it's for the race against time.
The Node Gap That Everyone Ignores
DRAM is not logic. No FinFET. No GAA. It's 1T1C — one transistor, one capacitor. Stack capacitor architecture. Same as Samsung, SK hynix, Micron. The physics is known.
The problem isn't architecture. It's process precision. CXMT's mass production is at 17nm/18nm. Samsung and SK hynix are at 1α/1β — roughly 12-14nm equivalent. That's a 1.5 to 2-node gap. Two to three years of distance.
DDR4 and LPDDR4? Fine. Competitive. But DDR5 and LPDDR5? Those are ramping on a 19nm process. That's not optimal. That's running an older road to a newer destination.
This gap isn't just pride. It's cost. In DRAM, node density is cost. The yield differential tells you everything:
Samsung and SK hynix hit 85-90% on 1α DDR5. CXMT is likely at 70-80% on 17nm DDR4/LPDDR4. That's a 10-15 point gap. And that gap is the real story.
It translates directly to gross margin: CXMT around 15-25%. Samsung DRAM? 40-50%. That's not management failure. That's physics, under sanctions.
The HBM Problem Nobody Wants to Address
Now the elephant — HBM. High Bandwidth Memory. The fuel of AI.
CXMT is not in HBM production. Zero percent market share. SK hynix has ~50%. Samsung ~40%. The gap isn't a couple years. It's a generational leap.
HBM3E requires TSV stacking — through-silicon vias, 8-layer or 12-layer stacking, thin wafer handling, and co-design with the GPU. It's a discipline CXMT hasn't built.
Their equipment access is the bottleneck. DUV immersion tools. ASML NXT:2000i or above — restricted. They've hoarded some, but new orders face 18-24 month delivery delays. And the US Entity List, imposed December 2022, blocks US tools entirely.
So where's the path?
The plan, as far as we can see: DDR5 mass production by 2026-2027. HBM3E by 2028-2030. That's the roadmap. If equipment sanctions tighten further, add 2-3 years.
The Contrarian Read: Greenshoe as Supply-Chain Insurance
Now here's the angle no one's talking about.
The greenshoe money isn't going to DDR5 R&D. The real target is supply chain localization.
CXMT's equipment import dependence is over 90% for lithography, over 80% for etching, over 70% for deposition. The materials story is worse: photoresists are 90% imported. 12-inch wafers 80%.
But here's the beautiful hidden signal — the full exercise of the greenshoe. The extra 870 million yuan. That's not enough to buy a single advanced lithography machine. But it's more than enough for something else: equipment qualification and procurement deposits.
In this industry, deposits get you queue position. It's how you signal to suppliers, local or Japanese, that you're real. And it's how you tell the market that your expansion plans are already oversubscribed.
The message embedded in the price stability isn't just about sentiment. It's about institutional confidence in supply chain resilience. If the market believed the sanctions would break CXMT, the price would be below issue. CICC would have to buy. They didn't. That's a verdict.
The greenshoe is an option. The exercise is a confirmation. The non-purchase is the real alpha.
The Money Problem
Now the uncomfortable numbers. The ones that the IPO narrative conveniently ignores.
CXMT's current valuation: PE 50-60x, PB 3-4x, PS 5-7x. Compare to Samsung: PE 20-30x, PB 1.5-2x, PS 2-3x. CXMT is trading at double to triple the valuation of the global leaders. That's the domestic substitution premium.
The question is: can it deliver?
Gross margins are 15-25% and rising. The company's OCF/net income ratio is healthy — 1.2-1.5, suggesting quality of earnings. But free cash flow is negative — about -2 billion yuan. CapEx runs at 50-60% of revenue, dwarfing TSMC's 35-45%.
That's expansion. That's also dependence on the National Fund III — the 344 billion yuan war chest. CXMT is expected to receive 10-20 billion yuan from that.
The critical metrics:
Current capacity utilization: 80-90%, healthy. The industry is in the early stage of an upcycle. DRAM contract prices rose 10-15% in Q3-Q4 2024. Expected another 10-20% in 2025. But — CXMT is not a beneficiary of HBM demand. Not yet. The AI boom flows to HBM, to Samsung, SK hynix, Micron.
CXMT's role is DDR5 and LPDDR5 — the mid-tier AI inference chips. The 5-10% price premium on those is real, but it's not HBM-level.
The demand is real — smartphones 40-50%, servers 20-25%, automotive growing 20%+. But the real strategic value is geographic: CXMT holds about 50% of the China DRAM market. The number 1 position in the world's largest, most sanctioned economy.
What They're Not Telling You
Now here's the part I think the market is missing. It's about the strategic direction, not the node.
The greenshoe exercise happens when the underwriter is confident the market is willing to absorb more shares at the issue price. That's the technical read.
But there's another read: CXMT is signaling that its expansion plans are already overcommitted. The 870 million is a small addition, but it's a necessary one. The company is running a marathon in shackles.
The current fab expansion — Fab 1 Phase 2 adding 60,000 wafers per month by 2026, new fabs planned for 2027-2028. The capex is on the order of $150 billion for the new fab.
And here's the most important part of the hidden story: The lack of secondary-market purchase means the market is pricing in a DRAM upcycle.
This isn't just about CXMT. This is about the global memory market reading. If the greenshoe is exercised and the price stays above issue — that's a validation of the whole DRAM cycle.
The Takeaway
So what happens next? I'm watching three things:
First, the next ASML shipment news. If the Dutch license approval continues, CXMT's timeline holds. If it stalls, the entire roadmap shifts 2-3 years.
Second, the local supply chain. I want to see what percentage of the new capex goes to domestic tools. If that number rises above 30%, the substitution story is real.
Third — the HBM roadmap. Any hint of TSV capability, any partnership with a Chinese AI chipmaker — that's the trigger that changes the entire valuation.
CXMT isn't just another semiconductor stock. It's a geopolitical narrative wrapped in a DRAM price cycle.
The greenshoe was exercised. The market didn't blink. The machinery is running.
Now watch the machines.