CXMT Opens at $487B: China's Memory Champion Gets a Public Balance Sheet
Source: The Information, 2026-07-27 (Qianer Liu) | Link: The Information | Raw: raw file
TL;DR
ChangXin Memory Technologies, China's leading DRAM maker, rose 472% on its Shanghai debut on Monday. The opening price valued CXMT at roughly 3.3 trillion yuan, about $487 billion. Investors are pricing it as an AI-memory beneficiary. Two days ago the wiki recorded Tim Cook lobbying Washington for permission to buy memory from CXMT and YMTC for devices sold outside the US. The company he wants a carve-out to buy from is now one of the most valuable semiconductor firms on earth, funded by a domestic capital market that does not need the carve-out to exist.
Why this is a memory-hierarchy story, not a markets story
A 472% first-day move is not a considered valuation. It is a supply-constrained listing in a market with capital controls and heavy state participation, and the wiki's own sources have flagged the structural version of this: the Semiconductor Newsletter's Week 30 issue notes that state-affiliated capital supplies more than 90% of China private equity funding. Treat $487B as a policy signal rather than a price.
The signal that survives that discount is about capacity, and capacity is what the memory-hierarchy page has argued is the binding constraint on inference cost through the end of the decade. An IPO of this size funds fab construction. DRAM fabs take years and enormous capital, and the single most important open question in AI infrastructure economics right now is whether HBM and conventional DRAM supply expands fast enough to break the allocation regime. CXMT just raised against that question.
It partially resolves a prediction from two days ago
The 07-26 digest predicted: "The administration denies Apple's CXMT and YMTC carve-out within 90 days, keeping memory contention fully priced into HBM." The signal named was any Commerce or BIS determination on Apple sourcing from entity-listed Chinese memory makers.
No determination has arrived, so the prediction is unresolved on its own terms. But the framing needs an update, and this is the useful kind of prediction failure. The digest treated the carve-out as the main channel by which CXMT capacity could reach the market. Today shows a second channel that does not route through Washington at all: CXMT can fund expansion domestically at an enormous valuation regardless of whether Apple is permitted to buy from it. Apple's carve-out determines who CXMT sells to, not whether CXMT builds. The capacity question and the access question have come apart, and the wiki was conflating them.
Relation to prior wiki state
This is the fifth independent vantage point on the same memory shortage in roughly two weeks, and the first from the supply side:
- AMD deleted the LPDDR tier from MI455X, a product decision made under allocation pressure.
- NVIDIA committed $500B with SK hynix's parent (07-25), locking in supply through equity rather than purchase orders.
- Micron ran up 213% year to date against NVIDIA's 10%, which is the market pricing memory scarcity above compute scarcity.
- Apple asked for an entity-list carve-out while simultaneously accusing Micron of gouging at 80% margins and calling the shortage a "100-year flood," two incompatible complaints.
- CXMT lists at $487B, today.
Every one of the first four is a demand-side actor scrambling for allocation. CXMT is the first datapoint about new supply arriving, and it arrives on the wrong side of an export-control line from most of the buyers who need it. That is the whole story: the shortage is now generating capacity, and the capacity is being built inside a jurisdiction that the largest buyers are legally restricted from purchasing from.
What to watch
- Announced capex and fab timelines from the IPO proceeds. The valuation is meaningless; the wafer starts are not. Any CXMT capacity announcement with a date is the number that matters.
- Whether CXMT is doing HBM or only conventional DRAM. These compete for wafer starts but serve different markets. HBM capability would be the far more consequential development, and the reporting here does not say.
- The BIS determination on Apple. Still the cleanest test of whether the allocation regime holds, now with the added context that denial does not stop CXMT from expanding.