ChangXin Memory Technologies (CXMT) priced its Shanghai STAR Market IPO at 8.66 yuan per share on July 25, raising roughly $9.8 billion at an $85.2 billion valuation. Trading begins July 27. The fourth-largest DRAM manufacturer globally, CXMT now has nearly double the capital it stated for expansion, giving China's AI labs a domestic memory supplier outside US export control reach. The company still trails Samsung and SK Hynix by roughly two process nodes and lacks HBM3/HBM4 capability, but the IPO buys it the runway to close the gap.
China's DRAM champion raised $9.8B. The real story is what it can now build. #
ChangXin Memory Technologies (CXMT) priced its Shanghai STAR Market IPO at 8.66 yuan per share on July 25, raising roughly 66.6 billion yuan ($9.8 billion) at an implied valuation of $85.2 billion. Trading begins July 27. It is the largest A-share listing by a Chinese chip company, surpassing SMIC's 2020 record, and the STAR Market's second-largest IPO ever.
The numbers are large. The implications for the AI infrastructure race are larger.
What CXMT Is and Why It Matters for AI #
CXMT is the world's fourth-largest DRAM manufacturer at approximately 7.7% global market share, trailing Samsung, SK Hynix, and Micron. DRAM — high-bandwidth memory — is the bottleneck in AI training and inference. Every Nvidia H100, B200, and Vera Rubin GPU depends on DRAM. Every AI data center runs on it.
Until CXMT's rise, the AI memory supply chain was a three-company oligopoly. Two of those companies are in South Korea (Samsung and SK Hynix), one is in the United States (Micron). All three operate under export control frameworks aligned with US policy. China's AI labs — DeepSeek, Alibaba, ByteDance, Moonshot — import their memory from companies in countries that restrict what they can sell.
CXMT changes the geometry. A domestic DRAM supplier at scale gives Chinese AI labs a memory source outside the US export control perimeter. The $9.8 billion raised in the IPO accelerates CXMT's ability to build fabrication capacity, invest in advanced node development, and close the technology gap with Samsung and SK Hynix.
The IPO Details #
The 8.66 yuan per share price represents 10% of CXMT's enlarged capital, implying the $85.2 billion valuation. The offering was heavily oversubscribed — Chinese retail and institutional investors bid for significantly more shares than were available, reflecting both nationalist sentiment around semiconductor self-sufficiency and genuine belief in CXMT's growth trajectory.
The $9.8 billion raised nearly doubles CXMT's stated investment needs. The company originally projected requiring roughly $5 billion for its expansion plans. The oversubscription means CXMT now has nearly twice the capital it said it needed. Some of that will go to capacity expansion. Some will go to R&D. Some gives CXMT a war chest for acquisitions, talent poaching, or aggressive pricing against incumbent DRAM suppliers.
CNBC flagged a concern worth noting: the IPO's size may pull liquidity from China's already-strained equity markets. When a single listing absorbs this much capital, smaller companies find it harder to raise funds. The counterargument is that CXMT is a strategic national asset — the capital allocation reflects government priorities, not market efficiency.
The Geopolitical Dimension #
CXMT's IPO lands at a moment when the US is tightening chip export controls and China is accelerating domestic semiconductor development. The Chengdu Statement — signed at WAIC 2026 — frames open AI models as China's strategic alternative to US-controlled AI infrastructure. CXMT's DRAM gives that alternative a hardware foundation.
The US has thus far focused export controls on logic chips — GPUs, AI accelerators, advanced processors. Memory has been a secondary target because the assumption was that China could not produce competitive DRAM at scale without Western equipment. CXMT's $85 billion valuation — and the capital to buy equipment from non-US suppliers — tests that assumption.
Samsung and SK Hynix are watching closely. Both have fabs in China. Both sell significant volumes to Chinese customers. Both operate under US export licenses. If CXMT can produce competitive DRAM at scale, those Chinese customers have a domestic alternative — one that does not require US government approval to ship.
What CXMT Still Cannot Do #
The valuation should not obscure the technology gap. CXMT's DRAM process node trails Samsung and SK Hynix by roughly two generations. HBM3 and HBM4 — the high-bandwidth memory variants used in AI accelerators — are not in CXMT's current production capability. The company makes standard DDR4 and LPDDR4 DRAM, which is used in consumer devices and lower-end servers. It does not yet compete in the premium segments that matter most for frontier AI.
The IPO gives CXMT the capital to close that gap. It does not give it the technology, the equipment, or the talent automatically. Advanced DRAM fabrication requires extreme ultraviolet (EUV) lithography equipment manufactured exclusively by ASML in the Netherlands. ASML does not sell EUV machines to Chinese companies under current export controls. CXMT must either find workarounds — using older deep ultraviolet (DUV) equipment with multi-patterning — or wait for China's domestic lithography program to deliver competitive alternatives.
Both paths take years. The IPO buys CXMT the years it needs.
The AI Memory Supply Chain After CXMT #
The DRAM industry has been a three-player game for decades. Samsung dominates at roughly 40% market share. SK Hynix and Micron split most of the rest, with smaller players like Nanya and Winbond occupying niche segments.
CXMT at 7.7% share and $85 billion valuation is no longer a niche player. With $9.8 billion in fresh capital, it can invest at a scale that forces the incumbents to respond. Samsung has already announced expanded HBM4 production. SK Hynix locked in long-term deals with Nvidia. Micron is building fabs in the US and India. The DRAM investment cycle is accelerating — and CXMT is now a participant with a balance sheet to match its ambitions.
For AI labs, the long-term implication is clear: memory supply diversity is increasing. If CXMT reaches HBM3 capability within three years, AI training infrastructure has a fourth credible memory supplier. That means more supply, more price competition, and reduced single-supplier risk for GPU manufacturers and cloud providers. The IPO is not the end of CXMT's story. It is the beginning of a chapter where China's memory champion has the capital to compete globally. The rest depends on execution.
Get AI news in your inbox
Daily digest of what matters in AI.