CXMT
ChangXin Memory Technologies

Snapshot: Executive Summary

ChangXin Memory Technologies, founded in Hefei in 2016, is China's largest DRAM maker and the only operational at-scale Chinese DRAM manufacturer, often called "China's Micron". It is the linchpin of the country's memory self-sufficiency drive, succeeding where Fujian Jinhua, Wuhan Hongxin, and Tsinghua Unigroup failed. It IPO'd on the Shanghai STAR Market on July 27, 2026 at ¥8.66 and closed its first session at ¥49.00, up ~466%, at ~¥3.28T (~$460-485B), the most valuable company on the mainland (intraday high ¥55.03, peak cap ~¥3.65T)³.

The problem is the price of that leadership. Memory is a deeply cyclical commodity, so a spot P/E is distorted: trailing GAAP P/E is meaningless (2025 net profit of only ¥1.88B implies ~1,700x), and the forward multiple of ~30x sits on annualized 1H26 peak-cycle earnings. The durable anchors are price-to-book and normalized mid-cycle earnings. On book, CXMT trades at a mid-teens-to-20x multiple versus Micron, Samsung, and SK Hynix at ~2-4x: a large scarcity premium for the sole listed at-scale Chinese DRAM pure-play. The debut also re-rated the company ~20x above its ¥158.4B (~$22.7B) June 2025 private round.

The operating story underneath is real: 2025 revenue was ¥61.8B (+156% YoY) with the first-ever annual profit (¥1.88B, ~$276M) and a 41% gross margin, and Q1 2026 revenue jumped to ¥50.8B (+719% YoY). We rate CXMT Hold, a fade-the-pop stance, base-case fair value ¥40 (bull ¥68 / bear ¥20) via a through-cycle EPS x normalized-multiple model cross-checked on price-to-book. Probability-weighted 25% bull / 40% base / 35% bear, the scenarios blend to ¥40, a deliberate downside skew below the ¥49 close. The franchise is a genuine national champion; the debut price, on the durable anchors, is not obviously a bargain here.

Tactical: CXMT trades at ¥49.00, ~23% above our ¥40 base-case fair value. This is a genuine national champion, China's only at-scale DRAM maker, riding a real memory supercycle (2025 revenue +156%, first annual profit). But the debut price embeds a large political-scarcity premium on peak-cycle earnings (price-to-book in the mid-teens versus Micron and SK Hynix at ~2-4x). Hold, a fade-the-pop stance: a framework, not a formal call.

Investment Thesis

Rating: Hold, a fade-the-pop stance. The probability-weighted scenarios (25% bull / 40% base / 35% bear) blend to ¥40 (0.25 × 68 + 0.40 × 40 + 0.35 × 20 = 17 + 16 + 7), a downside skew below the ¥49 close. We set the headline at the ¥40 base. CXMT is a genuine national champion and the supercycle is real, but the debut price prices near-flawless execution on cost, yield, and HBM, and on one day of trading the risk/reward skews to fade rather than chase. Upgrade trigger: a pullback toward the low-¥30s, confirmation the upcycle extends into 2027 with a credible HBM3 ramp, or evidence the cost-per-bit gap to the leaders is closing faster than modeled.

Product & Technology

CXMT makes DRAM, the volatile working memory in every server, PC, phone, and AI accelerator. The bulk of output runs on the G4 (~1z-nm) node, with G5 (~1a-nm) in development without EUV lithography, using multi-patterning on stockpiled DUV tools, the same pre-EUV path Micron once walked. The product line spans DDR4, DDR5 (16/24Gb, up to 8000 MT/s), and LPDDR5/5X (up to ~10667 MT/s); roughly 99% of 2025 revenue came from DDR and LPDDR⁴.

The two structural weaknesses

• Cost and ASP gap. Cost-per-bit is still >30% above Samsung, SK Hynix, and Micron, while the ASP discount is only ~5-10%. The gap may widen as the leaders mix toward HBM and server DRAM. Subsidies, not unit economics, close much of the difference today.
• HBM is the weak spot. High-bandwidth memory, the stacked DRAM bolted beside AI GPUs, is where CXMT lags most: HBM3 8-high yields are ~25% overall and 12-high is harder. HBM capacity of ~5 kwspm at end-2025 is modeled to ~30 by end-2026 and ~55 by 2027, but it is minimal to earnings today and no dedicated HBM line was funded by the IPO.

Put together, CXMT is a fast-scaling volume DRAM maker closing the gap on the mainstream nodes, not yet a leading-edge or HBM competitor. That distinction is the difference between the bull's bit-share ramp and the bear's cost-and-yield ceiling.

Fabs & Capacity

The volume story is the clearest part of the bull case. Wafer capacity scales from ~265 kwspm at end-2025 toward ~500 kwspm by 2028 (SemiAnalysis), with a long-term target of up to 600k wafers/month. For reference, Micron sits around ~385 kwspm in 2026E, so CXMT is closing the volume gap with the leaders even as it lags on cost and node⁴.

Funding the ramp

The July 2026 IPO sold 6.688B shares (~10% float) for ¥57.92B (~$8.6B), the largest mainland IPO since Agricultural Bank of China in 2010². Of the ¥29.5B of proceeds earmarked to projects, roughly 69.5% funds new wafer lines and technology upgrades and 30.5% funds DRAM R&D. Notably, no dedicated HBM line was funded, which is why HBM stays a longer-dated option rather than a near-term earnings driver. Extensive, unquantified government subsidies sit alongside the IPO cash.

China Policy & Export Controls

CXMT is roughly half state-owned. Hefei government entities hold the largest stake at ~36.8%, alongside the National IC Fund ("Big Fund"), the State Structural Adjustment Fund, China Ordnance, and China Mobile. The strategic shareholders that matter commercially are Alibaba Cloud (~3.85%), the anchor customer, and GigaDevice (~1.8%); the chairman is Zhu Yiming⁵. The company is core to China's memory self-sufficiency drive, and it succeeded where three prior national attempts failed, taking China's DRAM share from ~0 to ~5%+⁶.

The export-control nuance

The December 2, 2024 BIS rules added ~140 entities plus 24 tool types and restricted HBM and advanced-DRAM equipment. The differentiator: CXMT is NOT on the Entity List (reportedly after Japanese-industry pushback), unlike many Chinese chip peers⁸. That is a genuine advantage, but it is indirect: CXMT is still boxed by the DUV-tool and HBM-technology controls, and its G5 node is being developed without EUV. The right frame is "off the list, boxed by an equipment ceiling": unlimited state capital cannot buy the tools that would let it match the leaders on the leading edge, which is the durable limit on how fast the cost-per-bit gap can close.

Competitive Landscape

Global DRAM is a tight oligopoly of Samsung, SK Hynix, and Micron, which together hold the overwhelming majority of bits and virtually all of HBM. Into that structure steps CXMT as the #4 maker worldwide and #1 in China at ~7.67% global share, and, critically, the only operational at-scale Chinese DRAM pure-play. That scarcity is the entire basis of the debut premium.

Why there is effectively no domestic rival

• YMTC is NAND, not DRAM. China's other memory champion makes NAND flash, a different product; it is a sister national champion, not a DRAM competitor.
• Fujian Jinhua is inactive. The one prior at-scale DRAM attempt was effectively halted by US action, leaving CXMT alone in the segment.
• Adjacencies, not rivals. Cambricon (AI logic) and Huawei (a key customer, including for HBM) are part of the same domestic ecosystem, not competitors for DRAM share.

The result is a company with a near-guaranteed domestic demand pull (localization mandates plus Alibaba and Huawei) and no meaningful Chinese competitor, but which still has to close a real cost, yield, and technology gap against three entrenched global leaders. Scarcity supports the multiple; the competitive gap is what the bear case presses on.

Financial Health

CXMT's financials swung from red to a supercycle windfall in a single year. 2024 revenue was ¥24.18B with a loss and a negative gross margin. 2025 revenue reached ¥61.8B (+156% YoY) with the first-ever annual profit of ¥1.88B (~$276M) and a 41.02% gross margin⁷. Q1 2026 revenue then jumped to ¥50.8B (+719% YoY) with net profit of ¥25B, versus a ¥1.6B loss in Q1 2025. Management guides first-half 2026 revenue to ¥110-120B (~$16-18B) and net profit to ¥66-75B (attributable ¥50-57B). The driver is pricing: DRAM contract prices rose +75-98% YoY in Q4 2025 and Q1 2026, and TrendForce models 2026 DRAM prices up +150-200%⁹.

2025 print and 2026 guidance

Note: 2024 and 2025 figures are reported (prospectus and press); 2026 figures are company guidance or annualized author estimates. The 2024 negative gross margin and net loss are shown in parentheses. Government subsidies are extensive but unquantified. All values in CNY (¥).

Valuation

Memory cannot be valued on a spot P/E, because a deep cycle distorts it both ways. CXMT's trailing P/E is ~1,700x (on only ¥1.88B of 2025 profit) and its forward P/E is ~30x on annualized first-half 2026 earnings, but that forward number is a peak-cycle figure. Through the cycle, the durable anchors are price-to-book and normalized mid-cycle earnings. On book, CXMT sits in the mid-teens-to-20x range versus Micron, Samsung, and SK Hynix at ~2-4x: the debut prices a large political-scarcity premium, and it re-rated the company roughly 20x above its ¥158.4B June 2025 private round¹⁰.

Confidence note. The exact share count and closing market cap are MEDIUM confidence (sources put the cap between ~$460B and ~$540B intraday), and the price-to-book and normalized-P/E figures are computed illustratively from an illustrative post-IPO book value per share of ~¥2.40. Because trailing earnings are meaningless and forward earnings are peak-cycle, the scenario model in the next section anchors on a normalized mid-cycle EPS at a normalized multiple, cross-checked on book, rather than on any single spot number. TrendForce's +150-200% 2026 DRAM-price view is the swing input for whether peak earnings persist⁹.

Scenario Model: EPS × Normalized Multiple

The model is the heart of the report. Each scenario values memory through the cycle: normalized EPS × a normalized P/E = implied fair value, cross-checked against price-to-book. Toggle Base / Bull / Bear to load the anchor inputs; drag the EPS and P/E sliders to see the implied value update. The base (normalized EPS ¥2.00 at 20.0x) reconciles to ¥40, cross-checked at ~16.7x an illustrative ¥2.40 book value per share (the ¥49 close is ~20x that book). Bull ¥68 (¥2.72 × 25.0x) and bear ¥20 (¥1.60 × 12.5x, still ~2x the ¥8.66 offer) bracket it.

The two live levers are normalized mid-cycle EPS (¥) and the through-cycle P/E multiple. The P/B cross-check reads an illustrative book value per share of ¥2.40. Memory is a book-value and mid-cycle-earnings story, so both the EPS and the multiple are the swing variables; the sensitivity grid brackets the base.

Sensitivity grid: EPS × P/E (fair value in ¥)

Fair-value calculator (EPS × multiple)

Risk / Reward calculator

Note: The assistant reasons from the dashboard's data snapshot and thesis sections: it does not browse the web or access real-time fundamentals beyond what's in data.js. Treat its responses as scenario-modeling support, not primary research. Author judgments on rating, fair value, and probabilities remain with the analyst.

Risk Factors

• Cyclicality (the pre-eminent risk). The guided first-half 2026 margins are a cyclical peak. Memory always mean-reverts, and any DRAM price rollover, which TrendForce and others see as a 2027 risk once capacity floods in, reverses peak-cycle margins fast, because operating leverage cuts both ways. The whole bear case is this peak turning.
• Policy and geopolitical. CXMT is off the US Entity List today, but that could change: a future listing, or tighter DUV-tool and HBM-technology controls, would hit the equipment pipeline the ramp depends on. The company is a strategic target in the US-China chip contest.
• Technology. HBM3 yields are ~25% and 12-high is harder; the G5 node is being developed without EUV; and cost-per-bit is still >30% above the leaders. If the cost-and-yield gap fails to close, the bit-share ramp stalls and the premium is unjustified.
• Valuation and technical. The debut is a day-one +466% pop on a thin ~10% float with lock-ups still to expire and a heavily retail tape; the CSRC has flagged capital-reallocation risk. Price-to-book in the mid-teens-to-20x versus peers at ~2-4x leaves no room for disappointment, and a froth unwind can move the stock independent of fundamentals.
• Governance and disclosure. The company is roughly half state-owned, with government objectives that may not always align with minority shareholders, and subsidies that are extensive but unquantified. Disclosure is thinner than for the global peers.
• Customer and demand concentration. Demand leans on domestic localization plus anchor relationships with Alibaba and Huawei. That guaranteed pull is a strength today, but a China-skewed customer base concentrates the risk if domestic AI or handset demand softens.