On June 4, as the A-share market experienced a short-term correction, storage chip concept stocks surged against the trend, with several stocks like Yingfangwei and Dawei Co., Ltd. hitting their daily price limits.
Capital markets on the east coast of the Pacific also confirmed this trend: on June 3, the U.S. stock chip storage sector rose against the market. SanDisk surged over 6%, closing above $1,800 for the first time; Western Digital rose over 5%; Micron Technology gained over 1%, while the computing power benchmark Nvidia fell over 3%.
Between the rise and fall, the market’s pricing of the AI industry chain is undergoing changes.
Multiple A-share storage chip sector stocks hit their daily price limits.
Storage chips are shifting from a “supporting role” in computing power to a “focus” in the AI competition
GPUs “idling” is becoming a new reality in the AI industry.
Elon Musk’s AI company xAI owns 550,000 of Nvidia’s top-tier GPUs, but according to an internal memo obtained by The Information, its actual utilization rate is only 11%, with a large number of GPUs facing “idling.”
The reason for this phenomenon is not simply insufficient computing power, but rather the difficulty in matching multiple aspects such as data supply, network communication, and system scheduling to meet the computing demands of large-scale AI clusters. Storage capacity is becoming one of the important factors affecting overall operational efficiency.
2026 is seen by the industry as the “Year of Agents,” and the contradiction between computing power and storage capacity is further amplified. Compared to traditional chatbots, agents require multiple self-dialogues to complete actions, with each action leading to exponential increases in Token consumption and computing power demands, thereby placing higher requirements on “storage capacity.”
The market has also sensed the shift. Goldman Sachs’ latest report indicates that the global storage chip market is facing its most severe supply shortage in 15 years, with supply-demand gaps for DRAM, NAND, and HBM hitting record highs in 2026, and predicts the shortage may persist until 2027.
“The era of simply piling up computing power is over. The global AI competition is increasingly becoming storage-centric,” said Alper Ilkbahar, Chief Technology Officer of SanDisk, in an interview.
The global storage landscape is beginning to feature “China”
The explosion in storage demand is also bringing the global storage industry back into focus.
From a technical route perspective, storage chips are mainly divided into two categories: DRAM and NAND flash memory. The former is responsible for high-speed data reading and writing, serving as the core of running memory; the latter is responsible for long-term data storage, determining the device’s storage capacity. High Bandwidth Memory (HBM), which supports AI data centers, is an important branch of the DRAM family.
For a long time, the global storage landscape has been almost monopolized by South Korea, the United States, and Japan. According to industry data, in the first quarter of 2026, Samsung, SK Hynix, and Micron together accounted for about 90% of the global DRAM market share; in the NAND flash memory field, Samsung, SK Hynix, Kioxia, SanDisk, and Micron form the main supply structure; the HBM market is mainly dominated by SK Hynix, Samsung, and Micron.
But now, this seemingly rigid industrial landscape is introducing new variables.
YMTC X2-6070 128L QLC 1.33Tb 3D NAND.
According to CFM flash memory market data, in the first quarter of 2026, Hefei Changxin Technology’s global DRAM market share rose to 7.7%, ranking fourth globally; according to Counterpoint data, YMTC’s NAND Flash market share rose to 13%, competing with Kioxia, Micron, and SanDisk for the third global position.
In May this year, these two Chinese storage companies initiated their A-share listing processes.
The “domestic storage duo” is accelerating the engraving of “China” on the global storage map.
Both “domestic storage duo” but taking different “chip” paths
CXMT and YMTC, one making DRAM and the other 3D NAND, deeply cultivate different technical paths.
Distinguishing the two is not complicated. Taking buying a mobile phone as an example, in the common “
