ChangXin Memory Technologies officially listed on Shanghai’s STAR Market on July 27, 2026. Data from third-party market cap tracker 8Marketcap showed the company’s intraday valuation at about RMB 3.35 trillion, placing it No. 31 among listed companies worldwide at the time. That marks a steep rise from its IPO valuation of RMB 579.2 billion.
The contrast is hard to miss. As of the end of 2025, CXMT still had accumulated uncovered losses of RMB 36.65 billion, even as investors pushed the company to one of the biggest market debuts in China’s semiconductor sector.
The “506” project and Zhu Yiming’s return to DRAM
CXMT’s starting point goes back to May 6, 2016. On that date, Zhu Yiming and the Hefei municipal government discussed a memory-chip strategy, giving birth to the domestic DRAM push known internally as the “506” project. One month later, CXMT was registered in Hefei.
Zhu was not new to semiconductors. He earned bachelor’s and master’s degrees in physics from Tsinghua University and later studied at Stony Brook University in New York. In 2005, he founded GigaDevice in Beijing’s Zhongguancun area. At a time when foreign suppliers dominated the memory market, GigaDevice entered through NOR Flash, a smaller segment than DRAM but one with real technical barriers and steady demand in embedded devices. Through product development and continued R&D, GigaDevice grew into a major player in the global NOR Flash market and went public in China in 2016.
That track record gave Zhu credibility. But DRAM was a different business entirely. Often described as the crown jewel of semiconductors, the segment had been controlled for decades by Samsung Electronics, SK hynix, and Micron. It is both capital-intensive and technology-intensive, with investment requirements so large that a purely market-driven startup would struggle to survive. Around 2016, the global DRAM market was entering an upswing, driven by smartphone adoption and rising mobile memory demand, while China’s massive electronics manufacturing base still relied on imported chips. That mismatch set the stage for CXMT’s creation.
Hefei’s role was central from the outset. Before backing CXMT, the city had already built a reputation for using state capital to attract heavy-asset industrial projects, most notably through BOE. Officials saw strategic value in DRAM and chose to provide the financial base needed to start the project.
In 2018, Zhu stepped down as general manager of GigaDevice and formally became CEO of CXMT. He pledged not to take salary or bonus before the company turned profitable. He also allocated 768 million shares to employees. Based on the IPO price of RMB 8.66 per share, that equity incentive package was worth more than RMB 20 billion. Zhu also promised not to transfer his own shares during the first 10 years after listing.
A decade of losses and the long climb from zero
The barriers in DRAM extend far beyond capital spending. Manufacturing involves extremely small capacitor and transistor structures and demands high precision in lithography, etching, and thin-film deposition. On top of that, established players built extensive patent walls over decades, making it difficult for later entrants to avoid infringement or move quickly.
CXMT faced full-spectrum challenges in its early years, from process design to yield improvement. The company built its intellectual property system through a combination of in-house development and selective technology introduction. Yield ramp-up took time. The report said the process moved from the low tens in percentage terms to the high 80% to 90% range only after engineers repeatedly adjusted thousands of process parameters.
The company’s first-phase project involved total investment of roughly RMB 18 billion, with Hefei Industry Investment contributing RMB 14.4 billion, or 80%. In 2018, CXMT’s first production line began wafer starts. In September 2019, the company launched and mass-produced its first 10nm-class 8Gb DDR4 chip, a milestone described in the report as mainland China’s breakthrough from zero to one in DRAM.
CXMT continued to expand its product line after that. In November 2023, it released China’s first LPDDR5 product. By 2025, the company had covered DDR5 and LPDDR5/5X products.
As of the end of 2025, CXMT had 19,298 employees, with R&D staff making up 32.43% of the total. It operated three 12-inch wafer fabs, two in Hefei and one in Beijing.
The financial cost of catching up was steep. By the end of 2025, accumulated uncovered losses had reached RMB 36.65 billion. In 2024 alone, revenue was RMB 24.178 billion and net loss came in at RMB 7.145 billion. The report framed those losses as the price of process catch-up, lower yields, and higher bit costs during a period when the company relied on price competition and domestic replacement demand to win orders.
Patient capital and the jump in paper returns
CXMT’s financing history reflects unusually long-duration capital. Investors kept adding money through years of losses rather than pulling back. The company moved from a first-phase investment of RMB 18 billion to a pre-money valuation of RMB 140 billion in its eighth capital increase in 2024, and then to RMB 158.4 billion in the final round before listing.
Hefei’s state-owned capital system stands out as the biggest early winner. In the first-phase project, Hefei Industry Investment put in RMB 14.4 billion. Before the IPO, the broader Hefei state capital system held about 36.79% of CXMT. At a market value of RMB 3.35 trillion, that stake would be worth more than RMB 1.1 trillion. The report cited a Hefei Industry Investment executive as saying the bet was on Zhu Yiming himself. During the hardest loss-making years, Hefei state capital chose to keep funding the company.
Brokerages and market-oriented investors also entered early. China Merchants Securities invested RMB 324 million in the angel round and now holds about 0.84%. Cornerstone Capital led a RMB 1.2 billion investment in September 2021, while Walden International invested nearly RMB 900 million. Cornerstone Capital chairman Zhang Wei said the project required large-scale resource coordination, balancing multiple interests, and the ability to handle intense pressure. Dr. Peng Guie of Walden International said that when an industry becomes too difficult for financial models to quantify, judging the founder becomes the most important model.
In June 2024, during a downturn in the sector, the AIC units of China’s five largest banks entered together at RMB 2.61 per share. Compared with the IPO price of RMB 8.66, their paper gain had already exceeded 230%.
In the final financing round before the listing, Alibaba Cloud invested RMB 6.1 billion for a 3.85% stake, while Tencent held 1.50%. The report linked their participation to efforts to secure AI computing supply chains. By then, CXMT’s shareholder base had expanded from local state capital to market investors and industrial capital.
Q1 2026 earnings changed the narrative
The immediate catalyst behind CXMT’s RMB 3.35 trillion market value was its first-quarter 2026 financial performance.
Revenue for Q1 2026 reached RMB 50.8 billion, up 719% year over year. Attributable net profit came in at RMB 24.762 billion, while consolidated net profit was RMB 33.012 billion. On the report’s calculation, that equals roughly RMB 367 million in net profit per day during the quarter.
The report tied the surge to a combination of DRAM pricing and AI demand. At the start of 2026, DRAM contract prices rose 58% to 63% from the previous period. At the same time, the boom in AI servers lifted demand for high-bandwidth memory.
It also described the mechanics behind that demand. Training large AI models requires handling massive numbers of parameters, which puts heavy pressure on memory bandwidth. Traditional DDR products are not enough for GPU throughput in those workloads. HBM, or high-bandwidth memory, stacks multiple DRAM chips through advanced packaging and has become a crucial component in AI infrastructure. AI server growth also supports demand for conventional DDR5 and LPDDR5, because inference systems and edge devices need larger memory capacity as well. In the report’s reading, that shift in demand structure pushed up the pricing center for the broader DRAM market.
CXMT’s global DRAM market share reached 7.67% in the fourth quarter of 2025, ranking it fourth worldwide and first in China, according to the article. In HBM, the company has delivered 16nm HBM3 samples and plans mass production of HBM3E in 2027. The report said CXMT remains about two to three years behind SK hynix in HBM, but the roadmap gives it a path into data-center memory.
For full-year 2025, CXMT posted revenue of RMB 61.799 billion and its first annual profit of RMB 1.875 billion. It also expects net profit for the first half of 2026 to reach RMB 50 billion to RMB 57 billion, pointing to a clear earnings inflection.
What supports the valuation, and what still weighs on it
The article argues that CXMT’s RMB 3.35 trillion valuation reflects both a domestic substitution premium and an AI premium.
On valuation metrics, the company is trading at more than 300 times earnings based on 2025 profit, far above the 5 to 8 times price-to-earnings range cited for international peers. The report also said the combined market capitalization of Samsung Electronics, SK hynix, and Micron is about $4.1 trillion. Even with CXMT’s market value now approaching parts of the global memory group, there is still a clear gap in industrial fundamentals.
Process technology remains one of the key constraints. CXMT’s main process node is around 17nm to 18nm, roughly 1.5 to 2 years behind the leading global level, according to the report. In DDR5, its bit cost is more than 30% higher than that of the three market leaders. That matters because any downcycle in memory tends to magnify cost disadvantages and test earnings resilience.
Product mix is another issue. CXMT is still centered on DDR4 and LPDDR5, while global leaders have shifted more capacity toward DDR5 and HBM. Customer concentration is also high. The company’s top five customers account for more than 68% of sales, and it also has related-party transactions with GigaDevice. Global leaders, by contrast, have broader customer diversification across markets. In addition, the RMB 36.65 billion in accumulated uncovered losses still needs to be absorbed over future profitable years.
The report closes on a simple point: even with a clear profit inflection, CXMT remains a highly cyclical company. Process gaps, cost pressure, and valuation risk have not disappeared just because the stock market has repriced the company so aggressively.