A-shares chip stocks diverge: Haiguang Information, Wentai Technology, and others rise, while the storage sector faces pressure

Ask AI · How does AI compute demand drive the differentiated pattern in chip stocks?

On March 26, multiple A-share chip concept stocks rose. As of the time of this report, Jinhaitong (603061.SH) was up more than 4%, Star Semiconductor (603290.SH) was up 3.92%, Hygon Information (688041.SH) was up more than 2%, Wingtech Technology (600745.SH) was up more than 1.5%, Fuman Micro (300671.SZ), Changguang Huaxin (688048.SH), Moore Threads-U (688795.SH) and others were up modestly.

On the evening of March 25, U.S. chip-related shares strengthened. The Philadelphia Semiconductor Index rose 1.21%. Chip design company Arm closed up 16.38%. Earlier, the company released its first in-house chip, the AGI CPU, and said that by 2031, a single product would generate $15 billion in revenue.

In terms of news, there were reports that Intel and AMD have notified customers that CPU prices will be raised in March and April. Whether it is Arm or AMD and Intel, they are all signaling that surging AI compute demand is driving an expansion of supply-demand gaps and growth in market space.

However, it is worth noting that some A-share memory chip stocks fell today. As of the time of this report, Zhaoji Innovation (603986.SH) was down as much as 5%, Hengrui Shares (688416.SH) was down more than 4%, B&W Storage (688525.SH) was down more than 3.5%, Jiangbo Long (301308.SZ), Lankotech (300042.SZ), and others were also down.

In terms of news, Google has released a new AI memory compression technology called TurboQuant. It is claimed that this technology can reduce the cache memory footprint of large language models by at least 6x without losing accuracy, achieve up to 8x acceleration, and address memory bottlenecks in AI inference and vector search, raising market concerns about the outlook for storage demand.

Morgan Stanley noted that the technology only applies to key-value caches in the inference stage, does not affect the high-bandwidth memory (HBM) used by model weights, and is unrelated to training tasks.

Overall, growth in the memory chip sector may be entering a differentiation phase.

On one hand, the market growth momentum brought by AI demand will likely remain in the long term. TrendForce data from its industry survey shows that in the first quarter of 2026, AI and data center demand will continue to intensify global storage supply-demand imbalances, and manufacturers’ bargaining power will only increase. A research report from China Fortune Securities previously analyzed that the current memory shortage may persist until 2027. Although major DRAM manufacturers are expected to increase output by about 26% in 2026 and NAND output by about 24%, substantial expansion on the supply side—including from domestic storage manufacturers—still takes time.

But as competition intensifies due to technology optimization by leading manufacturers, the elasticity of AI demand for storage may return to rational levels. The growth logic for storage chips may shift from broad-based sector gains to structural differentiation, with growth in high-end markets such as HBM more likely becoming the main line.

(This article is from Yicai Finance)