China's CXMT became the most valuable company listed on mainland Chinese markets on its first day of trading, but the valuation is racing ahead of the memory chips it can actually make at scale.
ChangXin Memory Technologies opened at 49.50 yuan on Shanghai's STAR Market on Monday, July 27, against an IPO price of 8.66 yuan. It kept climbing. Reuters reported that the Hefei-based chipmaker raised 57.92 billion yuan, or $8.6 billion, in Asia's biggest IPO this year. South China Morning Post put the early trading gain at 472%, with a market value of 3.31 trillion yuan, about $489 billion. That made CXMT the most valuable company listed on mainland Chinese markets.
That is the hook. Not the listing ceremony, not the chip slogan, not the usual talk about self-sufficiency. A company that is still mostly a conventional DRAM producer was valued like the answer to China's AI memory shortage within hours of trading. You can understand the excitement. You should still be careful with it.
The retail frenzy tells you exactly what kind of moment this is. SCMP reported last week that more than 9.4 million investor accounts applied for shares and that the retail tranche was oversubscribed 212 times. Chinese investors, cut off from a clean domestic Nvidia trade and watching Washington tighten semiconductor controls, finally had one domestically listed DRAM story at scale. So they bought it.
Fewer than one in two hundred got shares.
CXMT is not vapor. According to Omdia figures cited by NAI 500, its global DRAM market share rose from 4.7% in the fourth quarter of 2025 to 7.6% in the first quarter of 2026, ranking it fourth behind Samsung, SK Hynix and Micron. That is a genuine achievement for a firm that barely registered on global memory charts five years ago. But the route matters. As Samsung, SK Hynix and Micron pushed more advanced capacity toward high-bandwidth memory for AI accelerators, CXMT grew in the conventional chips those companies were less hungry to serve: DDR5 for PCs, LPDDR5X for smartphones and standard server DRAM. NAI 500 said more than 98% of CXMT's revenue last year came from conventional DRAM.
That is the business today.
The HBM Gap Is Still The Whole Story #
High-bandwidth memory is a different problem. Nvidia's H100 and newer AI accelerators depend on stacked memory that is harder to package, harder to yield and much more valuable than commodity DRAM. CXMT is working on HBM3, but The Economy reported in April that its mass production timetable had slipped again because of yield problems, equipment constraints and thermal issues. AI Frontiers estimated earlier this year that Chinese domestic HBM production could reach roughly 7 million HBM3 dies in 2026, enough for about 600,000 AI chips if the assumptions hold. The report also called those estimates highly uncertain.
Read that carefully. It isn't nothing, but it isn't SK Hynix either.
The equipment problem sits underneath all of this. CXMT has relied on imported deep ultraviolet lithography equipment and stockpiled tools, while the most advanced EUV systems from ASML have been unavailable to Chinese chipmakers under export controls. Without that equipment, closing the yield and cost gap is slow work. It can be done in parts. It just doesn't happen because investors bid the stock up 472% before lunch.
Here is the tension in the price. Reuters said CXMT was valued at about $85.5 billion during the IPO process. A few hours of trading turned that into nearly $490 billion. That move only makes sense if you believe CXMT will become a serious HBM supplier and that China's domestic AI hardware stack will pay up for its output. It makes much less sense if the company remains heavily exposed to commodity DRAM, a market where prices can turn brutally when supply catches demand.
Micron Should Watch, Not Panic #
Frankly, the near-term picture for Micron is less dramatic than the headline looks. NAI 500 argued that CXMT's rise gives Micron some near-term relief because the Chinese company is filling conventional DRAM demand that the big three have been de-emphasizing, rather than attacking the highest-margin HBM pool. That is the right way to read it. CXMT is a serious fourth player in DRAM. It is not yet a peer in AI memory.
The longer-term risk is real. An $8.6 billion IPO gives CXMT money for production line upgrades and next-generation DRAM development. If it deploys that capital well, if yields improve, if Beijing keeps domestic customers close and if export controls leave enough room to keep buying or replacing critical tools, 2027 and 2028 become more uncomfortable for Micron, Samsung and SK Hynix. That's a lot of ifs.
There is one place where CXMT's progress is already visible to ordinary buyers. TechTimes reported that GIGABYTE, MSI and ASUS have validated CXMT DDR5 on major motherboard platforms, including speeds up to 8,200 MT/s in recent BIOS work. Most US builders still can't easily buy those modules, but the validation matters. It shows CXMT's commodity memory is no longer a curiosity sitting outside the PC ecosystem.
So the company is real, the production is real and the domestic market position is real. The valuation is doing something else. It is pricing in a geopolitical bet: that China can close the memory gap no matter what Washington does next. Maybe it can. But a nearly $490 billion market cap on day one, for a company still weak in volume HBM and locked out of the most advanced lithography tools, is the market running well ahead of the factory floor.
Also read: The AI capex boom is unlike anything America has built before and the funding structure explains why • Samsung is putting stablecoin support in Galaxy Wallet and 241 million phones are the distribution play • SK Hynix just moved $26.5 billion onto Nasdaq and then moved South Korea's currency