Growth in the memory chip sector may be entering a phase of divergence. On March 26, several A-share semiconductor concept stocks rose. As of the time of writing, Jht Design Co.,Ltd. (603061.SH) gained over 4%, Starpower Semiconductor Ltd. (603290.SH) advanced 3.92%, Hygon Information Technology Co.,Ltd. (688041.SH) increased more than 2%, and Wingtech Technology Co.,Ltd. (600745.SH) rose over 1.5%. Firms like Fullhan Microelectronics Co.,Ltd. (300671.SZ), Suzhou Everbright Photonics Co.,Ltd. (688048.SH), and Moore Threads Intelligent Technology Co.,Ltd. (688795.SH) also saw modest gains.
On the evening of March 25, U.S. semiconductor stocks strengthened, with the Philadelphia Semiconductor Index rising 1.21%. Chip design company Arm saw its shares close up 16.38% after announcing its first self-developed AGI CPU and projecting that a single product could generate $15 billion in revenue by 2031. In related news, Intel and AMD reportedly informed customers of CPU price increases scheduled for March and April. Both Arm and companies like AMD and Intel are signaling that surging demand for AI computing power is widening supply-demand gaps and expanding market opportunities.
However, it is noteworthy that some A-share memory chip stocks declined today. As of writing, Gigadevice Semiconductor Inc. (603986.SH) fell up to 5%, Zbit Semiconductor,Inc. (688416.SH) dropped over 4%, and Biwin Storage Technology Co.,Ltd. (688525.SH) declined more than 3.5%. Other stocks such as Shenzhen Longsys Electronics Co.,Ltd. (301308.SZ) and Netac Technology Co.,Ltd. (300042.SZ) also experienced losses.
Market concerns were partly triggered by Google's release of TurboQuant, a new AI memory compression technology. The company claims the technology can reduce the cache memory footprint of large language models by at least sixfold without sacrificing accuracy, while achieving up to eightfold acceleration, addressing memory bottlenecks in AI inference and vector search. This has raised worries about future storage demand prospects.
Morgan Stanley noted that the technology only affects key-value caching during the inference phase, does not impact the high-bandwidth memory (HBM) occupied by model weights, and is unrelated to training tasks.
Overall, growth in the memory chip sector may be entering a phase of differentiation. On one hand, the market growth momentum driven by AI demand is expected to persist long-term. According to TrendForce research data, AI and data center demand will continue to exacerbate global memory supply-demand imbalances in the first quarter of 2026, strengthening manufacturers' bargaining power. Huafu Securities previously analyzed that the current memory shortage could last until 2027. Although major DRAM producers anticipate increasing output by about 26% in 2026, with NAND output rising around 24%, substantive expansion from suppliers, including domestic memory manufacturers, will still take time.
However, as competition intensifies due to technological optimization by leading companies, the elasticity of AI-driven storage demand may return to more rational levels. The growth logic of the memory chip sector may be shifting from broad-based gains to structural differentiation, with expansion in high-end markets like HBM more likely to become the main trend.
Comments