CXMT
What CXMT is up to now
New 08-31 PM: began SMALL-BATCH PRODUCTION OF HBM3E, the high-bandwidth memory Nvidia H200/Blackwell-class accelerators are built around, and is the first Chinese maker to produce it at all (The Information, two insiders). Alibaba's T-Head and Cambricon are testing the silicon for commercial processors as early as 2027. The qualifiers carry the meaning: yields ~25%, and CXMT is assessed three to five years behind Samsung/SK Hynix/Micron, who are already in HBM4 mass production a full generation ahead. Its ~$8.6bn Shanghai IPO prospectus reportedly earmarked no capital for HBM at all. Read with the margin and LPDDR6 items below, the picture is a company that has entered the one memory category export controls were aimed most squarely at denying it — at a yield that makes it a demonstration, not a supply source. New 08-31: stopped being a policy question and became a technology one. Q2 gross margin 87.59%, up 8.44pp from Q1's ~79% and ABOVE both Micron (84.9%) and SK Hynix (83.2%) — it is now earning incumbent margins, not entrant margins. And it has begun MASS-PRODUCING LPDDR6 ahead of Samsung and SK Hynix, reported as the first commercial LPDDR6 anywhere, shipping in Xiaomi's 18 Fold in September; Lei Jun confirmed on Weibo. A company mass-producing LPDDR5 less than a year ago is first to the next generation. Separately and not previously on this record: CXMT SUED THE US DEPARTMENT OF DEFENSE on 08-29 in Washington federal court, naming Secretary Hegseth, to get off the Section 1260H "Chinese military company" list — arguing it makes standard civilian JEDEC-spec parts, and noting DoD published a February notice indicating removal then withdrew it the same day without explanation. That matters to the passed-silent Apple/Senate deadline here: the designation it rests on is now contested in court. New 08-30: the first results as a listed company, and they beat its own prospectus by roughly a third. First-half 2026 revenue RMB 150.31bn (~$22.4bn), up 873.64% year on year, and net profit attributable RMB 77.61bn against a RMB 2.33bn loss a year earlier — against pre-IPO guidance of RMB 110-120bn revenue and RMB 50-57bn profit. DDR5 alone was RMB 69.47bn, 46.3% of main-business revenue. This is the first hard supply-side number from the Chinese memory side this map has held: every prior read here measured the AI memory squeeze from the buyers (Nvidia's $279bn of forward purchase commitments, Micron's margins, Apple's price fight) and treated CXMT as a policy question about whether Apple may buy from it. Its own book says otherwise — the capacity is not waiting on a Washington decision, it is already earning at scale off the same DRAM tightness, which correspondingly weakens the leverage behind the passed-silent Senate deadline. Reported 08-28, caught 08-30; Bloomberg's page 403'd on a direct read, so this rests on SCMP and Benzinga reporting the same filed figures, agreeing to the decimal. New 08-23: its closest comparator moved — YMTC's RMB 33bn (~$4.9bn) Shanghai STAR Market IPO application was formally ACCEPTED, the procedural step after Friday's report of the target. SCMP frames it as China's next mega chip listing, one that could break CXMT's own market records. The China-chip-IPO wave now has a second memory maker in it with a number attached. Separately, the Senate's 08-21 deadline for Apple to reject CXMT/YMTC memory passed in total silence from both sides and is now a settled passed-silent. New 08-21: the pressure point passed without a result. The bipartisan Senate deadline for Apple to commit against sourcing CXMT/YMTC memory came due today and went silent on both sides — no Apple response, no senatorial comment. For CXMT specifically that is neither a win nor a loss: it remains a named target of US supplier pressure with no customer having publicly disavowed it, and no enforcement step taken either. The adjacent movement this week is upstream — Hua Hong, one of the three fabs named as an expected recipient of China's domestic immersion DUV scanners, joined the flagship Hang Seng index effective 09-07, putting the domestic-tool buildout into the benchmark. New 08-16: no further movement — zero CXMT-specific coverage across all four lenses today; the week-close picture below stands as the week's final word, Senate deadline (08-21) still open. week 2026-08-10 (/week): The week's clearest single data point for how completely the AI buildout has repriced Chinese semiconductor names — CXMT overtook Tencent as China's most valuable listed company (08-13, ~$524B vs. Tencent's ~$510B) just 17 days after its 07-27 STAR Market IPO, boosted by MSCI adding the stock to its China All Shares Index (effective 08-10). The supply-lock story that explains why CXMT can hold price held all week: Huawei/Xiaomi's high-level output lock-up, Apple's rebuffed discount push and separately-reported re-testing of CXMT chips, and the bipartisan Senate letter demanding Apple commit to never use CXMT/YMTC memory (reply due 08-21) all stand unresolved. No 08-15-specific development beyond continued no-movement on the Senate deadline. (Board note: board.yaml's `commanded_capital` market-cap figure was already corrected to ~$524B in the 08-15 board pass — this entry's own numbers below are consistent with that, no drift here.) New 08-13: CXMT overtook Tencent to become China's most valuable listed company — a market cap of roughly $524B against Tencent's ~$510B (down 26%+ this year despite rising AI spending) — just 17 days after its 07-27 STAR Market IPO. MSCI's addition of the stock to its China All Shares Index (effective 08-10, via the mega-IPO exemption logged 08-09) added further momentum on top of AI-driven memory-chip demand; read across coverage as the clearest single data point yet for how completely the AI buildout has repriced Chinese semiconductor names. Standing picture from 08-05→08-09, unchanged: CXMT can hold firm on price because Huawei and Xiaomi already locked up its output at high levels — it rejected Apple's push for a discounted DRAM price, while Apple is now reported testing CXMT chips again for China-market devices in tension with its earlier 'gave up' framing. The bipartisan Senate letter (Banks, Schumer, Kim, Shaheen, Crapo, Ricketts) to Apple CEO Tim Cook demanding a commitment to never use CXMT or YMTC memory chips in any product, citing both firms' place on an updated Pentagon China- military-linked list, has a reply due 08-21 — still unresolved. TSMC is reportedly sitting on ~$1B of finished A20 Pro chips it can't complete without matching DRAM. Overhangs unchanged: a possible US DoD ban and the three-year HBM deficit it's racing into.
as of 2026-08-31
China's domestic DRAM push; STAR-Market IPO (CXMT IPO thread).
The metrics
- Posture high
- expanding 1 src →
- Capital · available high
- ~$8.5B fresh IPO proceeds (up to ~$9.8B with 15% overallotment); + state 'Big Fund' backing (Big Fund specifics low-confidence) 3 src →
- Capital · operating low
- no clean operating-cash-flow / FCF figure (newly public, prospectus-only); revenue growing fast but profitability undisclosed here 1 src →
- Capital · deployed low
- IPO proceeds earmarked to domestic DRAM/HBM capacity expansion (specific capex figure not located — low confidence) 1 src →
- Capital · in low
- IPO ~$8.5-9.8B + recent-period revenue ~$7.4B (50.8B yuan, +700% YoY). NB the $7.4B is reported as 'Q1 2026' but that scale implies an ANNUAL figure for a ~7.7% DRAM player — period … 1 src →
- Capital · out low
- capacity capex (state-directed buildout); dividends/buybacks not applicable (growth-stage, state-backed) — figures not located 1 src →
- Optionality med
- locked — state-directed capital (36.29% state pre-IPO); Entity-List/DoD-1260H constrained; strategic, not return-maximizing 2 src →
- Gravity med
- ~7.7% of the global DRAM market GROSS — China's domestic memory-independence supplier (feeds Alibaba / ByteDance / Xiaomi); strategic gravity >> its current commercial scale 2 src →
This week — what changed
- CXMT crossed 10% of global DRAM revenue share — the first double-digit quarter for any Chinese DRAM maker, per Counterpoint Research, and roughly two years ahead of Counterpoint's and UBS's own 2028 projections. ⚠️ Dated 09-01, out of this window, flagged because nothing on the map recorded it. (Counterpoint Research, TechNode, Seoul Economic Daily)
