CXMT, China’s marquee IPO, is set to debut in what can only be considered a roaring year for memory chip stocks. The IPO stands at 6.7 billion shares, which represents 10% free float post issuance. The offering price has been set as CNY 8.66 per share, which puts CXMT at a valuation at CNY 579 billion.
Ahead of the IPO, we break down our views on CXMT’s business, its prospects and our view of the broader industry.
Company Overview
CXMT is the 4th largest DRAM manufacturer globally, with a 9% bit market share as of 2025. DRAM chips, or Dynamic Random Access Memory, is a type of memory semiconductor that stores data temporarily during data processing. DRAM is considered a volatile memory product and can only operate when the device’s power systems are online.
Founded in 2016 from the acquisition of intellectual property of the now-defunct memory maker Qimonda, the company’s products mainly consist of conventional memory chips. CXMT is headquartered in Anhui, China, with 97% of its sales as of 2025 coming from the greater China region.
Business Outlook and Prospects
CXMT’s DRAM bit shipment market share rose from 1% in 2021 to 9% in 2025, driven by rapid expansion and improvement in its manufacturing capabilities while supported by an ongoing government push for the adoption of domestic semiconductor technologies. The deepening undersupply in memory chips as a result of the tremendous surge in AI investments has also led to growing adoption of its products by global consumer electronics leaders, as procurement needs become increasingly unfulfilled with memory makers prioritizing AI memory supply over consumer products.
We expect these factors to continue acting as catalysts for CXMT’s growth moving forward. AI investment intensity is starting to pick up, and we foresee strong private and public sector investment into AI infrastructure as AI progress increasingly becomes both a matter of business competitiveness and national security. Global adoption in consumer electronics is also expected to continue, as electronics players who are currently facing margin headwinds as a result of deep memory undersupply will likely aim to diversify their memory supply chain in order to weaken the pricing power of existing memory leaders. We foresee CXMT’s market share to reach 10% in 2026.
That said, while we expect domestic AI investments to be a major demand driver for CXMT’s DRAM products, we do not expect the company to capture a material share of the global AI memory market due to technological gaps versus global leaders which stem from restricted access to chipmaking equipment and geopolitical tensions which limit international adoption.
Current Technological Progress versus Peers, and CXMT’s Roadmap from Our Lens
CXMT’s DRAM manufacturing expertise has shown tremendous improvement, driven by aggressive R&D and capacity investments. R&D and capital expenditures averaged 17% and 188% of revenue respectively over the past 2 years, well above the industry average. These outsized investments seem to have borne fruit, as the company’s rapid supply growth has allowed it to accommodate accelerating demand for domestic DRAM chips both in China and internationally. CXMT’s high R&D intensity has also enabled it to close the technological gap with its peers from 6-9 years to just 3-4 years, a tremendous pace of improvement in our view.
CXMT’s IPO proceeds will also be mainly spent on growth investment, with roughly 69% used for equipment upgrades and the rest for R&D, which we view as necessary to remain competitive in its offerings moving forward.
That said, we see growing challenges for CXMT to further close the gap against global memory leaders. We believe physical limitations are approaching CXMT in terms of node advancement. Memory leaders are expected to use EUV lithography and even high NA EUV lithography from the latest 1c DRAM node. CXMT’s latest G5 technology is equivalent to memory leaders’ 1a DRAM node 2 generations back, and without access to EUV lithography, we believe that conventional DRAM advancement will become increasingly difficult for CXMT beyond the next generation. We foresee CXMT circumventing this by pushing the limits of extreme multi-patterning as well as innovating on back-end packaging. We believe these innovations could possibly prevent the gap in process knowledge from expanding but are insufficient in further narrowing the technological rift against global players.
High Bandwidth Memory, or HBM remains a small part of CXMT’s shipments. We believe this is driven by low economic yields due to HBM’s significantly higher production complexity versus conventional DRAM. That said, we still expect CXMT to invest heavily into HBM expertise as customer interest accelerates alongside China’s AI investments and believe HBM will become an increasingly important part of CXMT’s business moving forward.
We think CXMT is well positioned to capitalize on rising domestic AI demand. CXMT is expected to commercialize HBM3 by 2026, and a rapid emergence of domestic AI chips that use HBM provides strong prospective demand for CXMT to fulfil.
The National Angle, Geopolitical Tensions and Implications
AI is increasingly becoming an issue of national security for China, and we believe CXMT is a key beneficiary. We view China’s continued support of domestic chip technology and capacity, from raw silicon to completed, packaged chips and broader data center infrastructure as a long-term goal towards becoming self-sufficient in semiconductor manufacturing.
To this end, China is encouraging home-grown chip adoption by domestic internet giants who are spearheading AI model advancement. While CXMT’s technologies still lag global memory leaders, we expect robust adoption of its chips, as the government encourages AI architectures to be built with domestic technology in consideration.
That said, we do not foresee global memory players being shut out from fulfilling China’s memory needs. CXMT’s wafer capacity remains small compared to its peers, which we estimate at 325,000 wafer starts per month by the end of 2026, roughly one-sixth the projected 2026 capacity of the 3 memory leaders combined.
Furthermore, while AI chip access increasingly becomes a matter of national self-sufficiency, so is AI model progression. We see a delicate balance to be maintained between getting access to leading chips and ensuring that domestic chipmakers are not excluded from AI investment buildouts. We expect a scenario where domestic internet companies are required to purchase a certain allocation of CXMT’s DRAM chips but are not denied access from leading global memory chips that can be accessed under existing trade conditions, allowing advancement of both domestic AI software and hardware.
Does the Memory Industry, and CXMT in particular have an Economic Moat?
We maintain that the memory industry does not possess an economic moat, and even more so for CXMT, whose technological expertise remains meaningfully behind peers.
DRAM pricing is primarily a reflection of demand and supply conditions. Overall, DRAM chips are relatively commoditized products with little differentiation and are interchangeable between brands, possessing low switching costs. Furthermore, competitive dynamics change from one technology generation to the next, and resultingly, no incumbent enjoys a technological edge that can be maintained over a decade. As such, we do not think any player is able to preserve excess returns structurally.
CXMT’s technological progress has been nothing short of impressive, but its inability to access EUV lithography technology is becoming a significant technological barrier to the progression of commercial and economically sound technologies, as the hardware limitations will make continued node advancement increasingly costly and complex. In turn, this is likely to prevent the company from materially closing the technological gap versus global memory leaders.
We do not think that overall product innovation will stagnate for CXMT and believe the company will find different ways to prevent the technological gap from re-expanding, such as introducing innovations in back-end packaging to support higher data transfer rates and bandwidth even with lower bit density. That said, we expect these workarounds will further weigh on final yields and keep profitability well under its peers, further underpinning our no-moat view of the company itself.
The Bottom Line; Our Fair Value Estimate and What Really Matters to Investors
Our fair value estimate for CXMT is CNY 14.90, and we view shares as undervalued versus its IPO offering price of CNY 8.66.
Our FVE implies a fiscal 2027 price to book ratio of 1.8 times, lower than our pure-play memory coverage which we value at 2.1–2.3 times, but significantly higher than its IPO pricing implied at 2027 book value. We expect a valuation uplift from the ongoing memory upcycle.
Conventional DRAM Price (USD/GB)
Multiples remain strictly under pure-play memory leaders in our coverage given CXMT’s lagging technological capabilities, translating into lower DRAM prices versus its peers. We do not expect this pricing gap—and the corresponding valuation discount—to narrow unless CXMT can overcome the EUV constraint while maintaining economic profitability.
We forecast CXMT’s revenues to grow 405% and 64% in 2026 and 2027 respectively, with operating margins of 75% and 78% across this period significantly higher than 14% in 2025. Robust AI demand has led to deep DRAM undersupply, and limited capacity growth through 2027 implies strong near-term pricing dynamics.
That said, we view the currently strong DRAM prices as temporary. Beyond CXMT’s expansion, we expect sharp capacity growth from memory leaders in the latter half of 2027 and into 2028, which should ease the tight demand-supply dynamics and drive price erosion from 2029 onwards.
