Shares in ChangXin Memory Technologies rocketed as much as 531% within hours of opening on Shanghai’s STAR Market this morning, turning a decade-old DRAM manufacturer into the most valuable company listed on mainland China’s stock exchanges. Traders piled into a stock that raised RMB 57.92 billion ($8.6bn) in what is now Asia’s largest listing of the year, and the numbers alone would justify a headline. However the most interesting element to the story sits underneath the trading screen.
CXMT builds DRAM, the workhorse memory that sits inside every server, laptop and phone on the planet. It has done so without access to the extreme ultraviolet lithography machines that Samsung, SK Hynix and Micron rely on, because Dutch export rules have blocked ASML from shipping that equipment to China since 2019. Instead, CXMT has leaned on older deep-ultraviolet tools, running each chip layer through multiple exposure passes to approximate what EUV does in one. It works, but it costs more, and analysts still put CXMT’s cost-per-bit more than 30% behind its rivals.
None of that stopped the market piling in on Monday. CXMT’s post-listing valuation landed somewhere between RMB 3.3 trillion and RMB 3.66 trillion, comfortably ahead of Intel and every Chinese bank on the exchange. The scale of the reaction says as much about Beijing’s semiconductor ambitions as it does about the balance sheet.
Why data centre buyers can’t ignore a Shanghai memory listing
The commercial logic behind CXMT’s rise is straightforward once you follow the supply chain. Samsung, SK Hynix and Micron have all redirected advanced production capacity towards high-bandwidth memory for AI accelerators over the past two years, chasing the fatter margins that Nvidia’s GPU customers are willing to pay. That left a gap in ordinary DDR5 and LPDDR5X, the memory that fills conventional servers, PCs and phones. CXMT stepped into it. Gartner projected in February that combined DRAM and SSD prices would climb roughly 130% by the end of 2026, a squeeze that is already showing up in server procurement budgets across the sector.
The sovereignty layer that doesn’t disappear at listing
This is where the story stops being a straightforward capital markets tale and becomes one about infrastructure sovereignty, CXMT’s newly listed public status changes nothing about the legal obligations it carries under Chinese law. China’s National Intelligence Law requires organisations under its jurisdiction to support state intelligence work when asked, and the Data Security and Cybersecurity Laws layer on further localisation and access requirements. None of that is unique to CXMT, but it is unavoidable for any enterprise weighing CXMT memory for sensitive workloads, regardless of where the servers physically sit.
The US side of the ledger is just as concrete. The Pentagon’s Section 1260H designation, restored in June, already bars the Department of Defense from buying CXMT-linked components, and Section 5949 of the FY2023 NDAA extends that ban to all federal agencies from December 2027. A bill working through the House Foreign Affairs Committee, the MATCH Act, would go further still and try to close CXMT’s DUV lithography workaround entirely. For any enterprise that supplies, or hopes to supply, US government contracts, that’s a direct constraint on component sourcing that has nothing to do with performance.
There is also a genuine technology gap that data centre engineers will care about more than politicians do. CXMT remains roughly three years behind Samsung and SK Hynix on high-bandwidth memory, the format Nvidia’s accelerators actually need, with fewer than 2% of its current wafer starts allocated to HBM production.
It is, however, moving faster than expected on an alternative called bonded DRAM, which fuses separately patterned wafers together to claw back some of the density EUV would otherwise deliver. Whether that closes the gap or simply narrows it is the technical question worth watching over the next 18 months, not the share price.
What this means besides the rest of the memory race
CXMT’s debut lands in the middle of a much wider scramble for memory sovereignty. Samsung has opened a dedicated next-generation semiconductor R&D lab built around High-NA EUV, South Korea has committed $576bn to AI chips and data centres to defend its lead, and even Intel and SoftBank are chasing an alternative low-power memory architecture designed to sidestep HBM’s power costs altogether. On the export control side, Beijing has been weighing its own restrictions on AI chips and models, while Huawei’s chip business has ridden a wave of domestic demand as US restrictions push Chinese buyers towards homegrown silicon.
RELATED STORIES
Trump renews push for AI giants to foot data centre power bill
Trump’s UK visit: A pledge to support data centre investment and industrial growth






