Wall Street has finally realized one thing: even if AI suffers from memory shortages, it doesn’t mean Micron, Samsung, or SK Hynsi can keep selling at premium prices forever. Recently, global memory chip stocks experienced a collective plunge, triggered by China’s CXMT (ChangXin Memory Technologies). On October 10, CXMT announced a key breakthrough in its 4F² DRAM architecture and plans to launch server DDR5 memory using this new design by year-end. Immediately after the news, Micron, SK Hynix, and Samsung all saw sharp declines, and even companies like SanDisk and Western Digital—primarily focused on NAND flash or hard drives—were dragged down. Within six trading days, the Philadelphia Semiconductor Index lost approximately $1.64 trillion in market value. Of course, this downturn was also compounded by broader market concerns over AI capital expenditures, interest rates, and financing costs. I find this particularly interesting. On one side, companies like Micron continue to emphasize tight capacity constraints, with most HBM supply through 2027 already pre-booked by customers, and demand from AI data centers still surging. On the other side, CXMT’s announcement of just one unproven technology path—still lacking large-scale production validation—was enough to make Wall Street reconsider the entire industry’s valuation. Strictly speaking, CXMT’s 4F² architecture primarily targets server DRAM and remains far from directly threatening the high-end HBM market. Production yields, manufacturing costs, and customer certifications have yet to be adequately validated. Over the past year, AI infrastructure development transformed memory chips from a classic cyclical industry into what investors perceived as a scarce asset. Investors began believing that, amid ongoing GPU, TPU, and data center expansion, memory giants could sustain long-term pricing power due to persistent supply-demand imbalances. CXMT’s emergence is now beginning to erode that assumption. Notably, CXMT already holds nearly 10% of global DRAM revenue. Even if 4F² fails to achieve large-scale commercialization in the short term, Chinese manufacturers’ expansion in the mainstream DRAM market is already a tangible competitive force. I’ve always felt that when discussing the AI computing supply chain, far too much attention is paid to NVIDIA and advanced node technologies, while the memory layer is consistently overlooked. Every new data center and every GPU cluster requires massive amounts of memory and storage—and AI’s expansion is reshaping profit distribution across the entire storage supply chain. But growth in demand and growth in corporate profits are two different things. The memory industry has evolved for decades: every cycle of supply shortage brings price hikes and capacity expansion, followed inevitably by renewed competition. AI may extend this cycle—but it won’t eliminate it entirely. What’s most significant about CXMT’s move is that it has prompted Wall Street to begin quantifying the potential price competition from China’s新增产能 (newly added capacity). If Chinese manufacturers ultimately succeed in scaling yields, reducing costs, and ramping up output, the scarcity premium currently enjoyed by global memory giants may prove far less secure than capital markets have assumed.