CXMT's Revenue Went Global. Its Fabs Did Not.
Most coverage of Chinese memory asks whether CXMT can build capacity. The H1 2026 filing raises a different question: it already sells most of its output abroad, and it makes essentially none of it abroad.
Overseas revenue went from 36.9% of the total to 63.8% in twelve months. Non-current assets outside mainland China: 0.04%. That asymmetry is the most structural thing in the report.
Part 3 of a series on CXMT's H1 2026 interim report. All figures in Chinese yuan unless marked otherwise.
KEY TAKEAWAYS
1. Revenue booked outside mainland China rose from 5.61B to 95.79B yuan — 17.1x — taking the overseas share from 36.9% to 63.8%. Domestic revenue grew 5.6x over the same period.
2. Of 231.18B yuan in non-current assets, 0.097B sits outside mainland China. The selling moved; the making did not.
3. Customers each accounting for over 10% of revenue total 80.15B yuan, or 53.32% of the half.
What the filing says, and what it cannot say
Domestic revenue went from 9.61B to 54.25B yuan. Overseas revenue went from 5.61B to 95.79B. The split moved from 63.1/36.9 to 36.2/63.8.
Two caveats before going further. Regional revenue sums to 150.04B yuan against a reported total of 150.31B — a 0.27B gap, presumably other and adjustment items, immaterial to the shares.
The second matters more. "Outside China" is where revenue is booked, not necessarily where the chips end up. Chinese semiconductor exports frequently route through Hong Kong and Singapore trading intermediaries, and an interim report does not separate that. This post treats only "revenue booked as export rose sharply" as established fact.
Why overseas grew three times faster
Price alone does not explain the gap. Higher prices apply to domestic customers too. What is left is that more volume actually went abroad.
Two separate reports point that way. TrendForce reported on August 4, 2026 that HP, Asus and Acer had begun limited use of CXMT memory in products sold outside the United States. The Wall Street Journal reported on August 9 that Apple was in early-stage supply discussions about using CXMT memory in some iPhones and MacBooks sold in China.
Both postdate the June 30 cutoff, and both describe evaluation and early discussion, not adoption. Neither explains the half's numbers. They do suggest the export growth is not purely intermediary flow.
The selling left. The making stayed.
Two thirds of revenue now comes from outside mainland China. Of 231.18B yuan in non-current assets, 0.097B yuan — 0.04% — sits outside it. In practical terms the entire production base is in one country.
Compare that with Samsung and SK hynix operating fabs in both the United States and China, or Micron splitting production across Japan, Taiwan and Singapore. Concentration cuts two ways.
One way is efficiency. Everything in one place expands faster and staffs more easily. The capacity ramp discussed in Part 1 is partly a product of this.
The other is single-point exposure. Revenue now spans many regulatory jurisdictions while the assets producing it sit inside one. That is a structural fact rather than a good or bad one.
June 8, and two common misreadings
The geopolitics item in the risk section. On June 8, 2026 the US Department of Defense updated its list of Chinese military companies under Section 1260H, and subsidiary CXMT Storage appears on it. The company states there is no material adverse effect on ordinary operations or going concern.
It was not a new listing. The June update added names such as Alibaba, Baidu and Tencent. CXMT and YMTC had their existing entries maintained. The list runs to 188 companies.
Listing is not sanction or export control. It raises the risk of disadvantage in Department of Defense procurement and federal contracting, and the department has said it reserves other authorities for further action. Reading the listing as "cut off from export markets" is wrong.
The listing matters more when set beside the asset map. If regulation tightens, a company with essentially all of its plant in one jurisdiction has very little room to relocate around it.
How many customers is this
The company discloses the figure directly: customers individually accounting for more than 10% of revenue total 80.15B yuan, or 53.32% of the half.
Elsewhere the report names Alibaba Cloud, ByteDance, Tencent, Lenovo, Xiaomi, Transsion, Honor, OPPO and vivo as major customers. That list is not matched to the 53.32% figure anywhere in the filing. Which companies clear the 10% threshold is not disclosed, so the two pieces should not be joined.
High concentration is ordinary for a company at this stage. The question is how it behaves in a downturn. Large customers concentrate volume in a shortage and exercise pricing leverage when the shortage ends.
What I actually watch
| Line item | Where | How to read it |
|---|---|---|
| Regional revenue split | Quarterly and annual reports | Whether 63.8% holds or reverts tells you if the export shift is durable |
| Overseas non-current assets | Balance sheet notes | Any movement off 0.097B yuan is the first sign of geographic diversification |
| Share from 10%+ customers | Same filing | Rising 53.32% means deepening dependence; falling means a broader base |
| Global OEM adoption stage | Company disclosure and press | Evaluation, discussion, qualification, adoption and volume supply are five different events |
Risks to this view
▶ Booking geography. Whether "outside China" revenue reflects final demand cannot be established from the filing. Every reading here sits inside that limit.
▶ Reporting stage. The HP, Acer and Apple items are press reports, not company disclosure. The Apple item in particular was described as early-stage discussion, not adoption.
▶ Regulatory direction. Listing is not sanction, but further action under other authorities remains open — and so does relaxation.
The short version
Across three posts: CXMT earned a record half from plant it had already built (Part 1), the margin that produced it owes more to price than to cost (Part 2), and its market has globalized while its production has not (Part 3). The expansion money is already in the account.
How those three lines read in the next set of filings is the answer to whether Chinese DRAM can get through a full cycle intact.
Sources: ChangXin Technology Group H1 2026 interim report (unaudited, 688825); TrendForce, August 4, 2026; The Wall Street Journal, August 9, 2026; reporting on the June 8, 2026 Section 1260H list update. Everything here is from public sources.
Disclaimer: This post is for informational and educational purposes only. It does not constitute investment advice or a recommendation to buy or sell any security. All investment decisions are your own responsibility.
Comments
Post a Comment