Changxin Memory Technologies (CXMT), China's HBM3 manufacturer, goes public with $85 billion market debut on Shanghai Stock Exchange.
At an issue price of 8.66 yuan, the IPO raised approximately 57.9 billion yuan in total, exceeding SMIC and making it the largest fundraising IPO on the STAR Market since its launch. Some calculated how much profit a single lot allocation could yield, others predicted the company's market capitalization could reach several trillion yuan, and still others extrapolated which shareholders would be the winners of this IPO based on their holdings.
Changxin Memory was established in 2016. Over the past decade, it has progressed from the production start of 8Gb DDR4 to become the global fourth player. According to Omdia data, based on sales figures for Q4 2025, Changxin's global market share has reached 7.67%, ranking behind Samsung, SK Hynix, and Micron.
In a DRAM market where the top three players have long held over 90% of the share, 7.67% is already substantial enough to establish Changxin as a force that cannot be ignored.
At this moment, AI is driving the concentration of global storage resources toward HBM, while Changxin has just completed an upgrade of its mainstream products. These two forces have converged at precisely the right time.
In the coming years, Changxin will face a race against two speeds: how quickly the three giants replenish conventional DRAM supply, and how fast Changxin improves yield rates, reduces costs, upgrades products, and accumulates customers. This race will determine what legacy today's opportunity ultimately leaves for Changxin.
AI's impact on Changxin goes far beyond "increased computational demand means more memory sales."
From a product structure perspective, Changxin remains primarily focused on mainstream DRAM. The prospectus shows that in 2025, the LPDDR series accounted for 66.43% of company revenue, while the DDR series accounted for 31.87%. The former is widely used in mobile devices such as smartphones and tablets, while the latter primarily enters PCs and servers.
Therefore, defining Changxin simply as an AI company would be inaccurate. While it does not currently depend on HBM for its primary revenue, it genuinely benefits from the industry shifts AI has brought about.
The visible growth stems from demand. Training and inference of large models require more servers and larger-capacity server memory; as AI enters devices like smartphones and computers, it further raises the memory requirements of these devices. Both DDR5 and LPDDR5X benefit from this trend.
HBM is fundamentally DRAM as well, but its manufacturing process is considerably more complex. It requires vertically stacking multiple layers of DRAM chips and achieving high-speed interconnection through silicon-through vias and advanced packaging.
Micron mentioned in regulatory filings submitted in March 2026 that producing equivalent-capacity HBM requires consuming more wafers and cleanroom space. SK Hynix similarly stated in its medium and long-term investment strategy released in July 2026 that due to TSV structures increasing chip area, HBM requires more wafers at equivalent capacity; as HBM's proportion increases, maintaining the same memory output also requires more manufacturing capacity investment.
Over the past two years, HBM has become the most fiercely contested market among the three major manufacturers. With robust demand, concentrated customers, and higher product value, capital, advanced process technology, wafers, and packaging resources naturally tilt toward it.
Manufacturing resources cannot increase proportionately in the short term. As HBM receives higher investment priority, the supply of conventional DRAM products like DDR5 and LPDDR5X is affected. Meanwhile, demand from servers, smartphones, and PCs continues to rise, driving prices higher.
Against this backdrop, Changxin turned profitable in 2025, with Q1 2026 revenue reaching 50.8 billion yuan and net profit attributable to the parent reaching 24.762 billion yuan.
In September 2019, Changxin announced the launch of its self-developed memory chip manufacturing project, with 8Gb DDR4 debuting for the first time and marking a significant breakthrough in large-scale DRAM production in mainland China. Over the following years, its products gradually upgraded from DDR4 and LPDDR4X to DDR5, LPDDR5, and LPDDR5X, with production capacity and customer base expanding in tandem.
Changxin's current DDR5 products achieve maximum speeds of 8000 Mbps with chip capacities up to 24Gb; LPDDR5X reaches maximum speeds of 10667 Mbps with chip capacities of 12Gb and 16Gb. These products have entered the mainstream performance range required by PCs, servers, and flagship mobile devices.
Some analysts have characterized the current market as the three majors competing for premium HBM while leaving mid-to-low-end products to Changxin. This segmentation does not reflect the actual reality of the DRAM market.
DDR5 is becoming the mainstream memory standard for PCs and servers, while LPDDR5X is widely used in flagship smartphones, tablets, and thin-and-light computers. As edge AI continues to develop, these devices will have increasingly demanding requirements for capacity, bandwidth, and power efficiency. This shows that Changxin is pursuing a continuously upgrading mainstream market, not an obsolete market the giants are ready to abandon.
Published specifications do not represent the full picture of capability. Large-scale commercial viability of DRAM depends on process ramp-up, yield improvement, platform compatibility, and customer qualification. Even with identical speeds and capacities, differences may persist in cost, stability, power efficiency, and sustained supply capacity.
According to media reports citing supply chain sources, Changxin's LPDDR products have achieved adoption rates exceeding 30% in domestic Android-brand smartphones. More than a list of customer names, this figure demonstrates that Changxin has progressed from entering the supply chain to scaling adoption.
Similar progress has emerged in the server market. Reuters previously reported, citing three informed sources, that Changxin has signed a long-term DRAM supply agreement with Tencent valued at over 20 billion yuan, covering server memory supply over several years ahead. Neither Changxin nor Tencent has publicly commented, and the exact amount and agreement terms remain to be confirmed. However, this report aligns with the trend of Changxin entering the supply chains of leading cloud computing customers.
When supply is tight, customers are more willing to invest resources in bringing in new DRAM suppliers. Once the market loosens again and alternative suppliers become available, that urgency diminishes. Changxin must capitalize on this period to convert temporary orders into stable partnerships.
Smartphone, PC, and server manufacturers will not skip testing and qualification due to shortages. New memory suppliers must prove their products can run stably long-term and maintain consistency across production batches. Server and automotive customers have even higher requirements, with the journey from testing to volume procurement often taking considerable time.
The critical questions are whether a qualified product can enter more device models, whether the next generation will continue using it, and whether short-term orders can extend into multi-generation partnerships. These questions matter far more than momentary shipment growth.
DRAM production involves numerous intricate and complex process steps, with yield rates improving incrementally only through sustained production. Larger shipment volumes mean more process data and engineering experience. With each wafer batch calibrating the production line, greater numbers of qualified products make it easier to amortize depreciation and R&D costs.
Orders keep the production line running at full capacity, mass production drives yield and cost improvements, and more stable products create opportunities for the next round of orders. The faster this cycle turns, the more customers Changxin retains once industry supply normalizes.
Fabs can be built in a matter of years, but customer trust can only accumulate one batch of products at a time. What makes this window most valuable is that customers are willing to give Changxin more opportunities for testing and supply.
Samsung, SK Hynix, and Micron continue advancing conventional DRAM upgrades and capacity expansions, with advanced processes also increasing output per wafer. Should HBM supply and demand eventually balance, some manufacturing resources could shift back toward conventional DRAM. At that point, the supply environment Changxin faces will shift again.
Before then, product and process innovation must continue advancing. The IPO's three funded projects total 34.5 billion yuan in investment, with 29.5 billion yuan of raised capital to be deployed toward memory wafer manufacturing line upgrades, DRAM technology upgrades, and advanced technology R&D. The significance of this capital is that production line upgrades and multi-generation product development can proceed in parallel.
In 2025, Changxin's overseas sales excluding Hong Kong accounted for only 2.79% of total sales, with the market remaining highly concentrated domestically. China's vast smartphone, PC, server, and electronics manufacturing industries can support Changxin's expansion. Entering more global customers would further test its product quality, delivery reliability, and service capabilities.
Advanced server memory and HBM also require continued development. While they may not become major revenue sources in the near term, they are crucial to whether Changxin can build a more complete product portfolio and have greater flexibility when industry cycles shift.
Memory manufacturing is a business requiring sustained investment while grappling with price cycles. Zhu Yiming has pledged not to transfer his pre-IPO shares within ten years of Changxin's listing. This unusually long lock-up period reflects the extended cycles of DRAM R&D, capacity expansion, and customer validation.
Today, Changxin holds nearly 8% global market share, has achieved mass production of mainstream products, and continues to expand its customer base, while securing substantial capital during a favorable industry cycle. A decade of accumulation has perfectly aligned with the supply and demand shifts AI has triggered.
Today, Changxin is converting this near-term supply-demand window into lasting competitive advantage. When DRAM prices return to normal, what it will have built is a stable customer base, a more mature manufacturing system, and the capability to continue expanding its market share.