Tesla CEO Elon Musk believes that once China masters the manufacturing technology for a specific product, its capacity for mass production often surpasses that of all other countries combined; lithium-ion batteries and solar panels are prime examples. He warned that the memory (RAM) industry could be the next to face competition from China’s massive scale of production, though significant technical barriers remain for High Bandwidth Memory (HBM) and advanced 3D stacking technologies.
According to a report by Wallstreetcn, Musk posted on the social media platform X on the 12th, citing SpaceX as a rare exception; he noted that SpaceX’s orbital delivery capability is roughly an order of magnitude superior to that of all other launch vehicles in the world combined. However, he immediately added that even so, China has already captured the majority of the market share for orbital launches outside of SpaceX, describing SpaceX as an “outlier in the matrix.”
Musk’s assessment regarding the memory sector is more direct. He noted that until about three years ago, memory chips were low-margin, standardized commodities; with LPDDR5 and LPDDR6 utilizing shared specifications, China appears to be rapidly mastering the relevant technologies. At the same time, he pointed out that newer-generation HBM architectures, SRAM (Static RAM) layers positioned beneath logic layers, and 3D-stacked RAM placed above logic layers—while simultaneously addressing heat dissipation—represent far more difficult technical challenges.
Musk’s comments did not come out of nowhere. On the 10th, Cao Kanyu, President of ChangXin Memory Technologies (CXMT), announced at an industry event that the company had achieved a key breakthrough in next-generation 4F² memory cell architecture. The company plans to launch DDR5 server memory products based on this architecture by the end of 2026.
The core advantage of the 4F² architecture lies in its smaller memory cell area. In the 6F² architecture commonly found in traditional DRAM—where “F” represents the minimum feature size of the manufacturing process—the 4F² design can theoretically accommodate approximately one-third more bits per unit area. This reduces unit costs and makes the technology better suited for the high-capacity memory modules required by AI servers. Cao Kanyu stated that the design utilizes proprietary vertical Gate-All-Around (GAA) transistors to suppress leakage current, directly targeting the high-margin AI server DRAM market.
This news sharply intensified market concerns regarding Chinese manufacturers accelerating their penetration of the high-end memory market, acting as the immediate trigger for a broad sell-off in chip stocks on the 12th. Technology stocks were the worst-performing sector in the U.S. market that day, with memory chip-related stocks bearing the brunt of the decline. Micron fell below the $1,000 mark, ultimately closing down approximately 4%; SanDisk dropped nearly 6%; SK Hynix ADR and Super Micro Computer both fell more than 4%; Seagate Technology declined over 3%; Quantum plunged more than 9%; and Corning also suffered a significant drop.