On October 24, stocks related to storage chips experienced a significant rally. By the market's close, $普冉股份 (688766) and $香農芯創 (300475) soared by 20%, while $恆爍股份 (688416) rose over 18%. 江波龍 (301308.SZ) gained 16.73%, and other notable increases were seen in 華虹公司 (688347.SH), 佰維存儲 (688525.SH), and 德明利 (001309.SZ). In the Hong Kong market, 華虹半導體 (01347.HK) rose 13.73%, and 中芯國際 (00981.HK) surged by over 8%.
The pullback was driven by news that Samsung Electronics and SK Hynix have increased their DRAM and NAND flash prices by up to 30% in the fourth quarter and passed this new pricing structure onto customers. This indicates a tense imbalance of supply and demand in the current storage market. Other storage manufacturers are joining in as well. According to predictions by Citigroup and Morgan Stanley, the average price of DRAM is expected to rise by 25-26% in the fourth quarter, potentially intensifying the price increase trend.
In recent years, the rapid advancement of AI technology and accelerated data center infrastructure have driven a surge in demand for upstream storage chips. In the AI era, storage chips have transitioned from being supporting components to becoming core bottlenecks and breakthroughs. The AI wave is igniting a "storage power" revolution. It is worth noting that many major clients are actively engaging in discussions with leading storage manufacturers for long-term supply contracts. During a visit to South Korea, OpenAI CEO Sam Altman established a strategic partnership with Samsung Electronics and SK Hynix to procure 900,000 wafers per month for the "Gateway" project. Although some skeptics labeled this deal as mere "wishful thinking," it has significantly boosted market sentiment and heightened price expectations in the storage sector.
Specifically, this round of price rises in storage chips is primarily represented by HBM chips. Micron's Chief Business Officer, Sukhi S C Sadhana, mentioned that part of the increase in DRAM prices is due to supply tightness, heavily driven by soaring demand for HBM. HBM consumes more than three times the wafer capacity of standard DRAM.
Due to the explosive growth in model parameters and training data, traditional memory technologies have become "memory walls" that limit computational power. Meanwhile, HBM, with its ultra-high bandwidth, low power consumption, and compact size, is becoming the mainstream choice for AI chips, resulting in a rapid increase in demand, making it a highly sought-after product in the storage chip arena.
In addition to strong demand for high-end HBM chips, the traditional DRAM and NAND chips are also in short supply. As AI investments shift from large-capacity data training to inference, the demand for traditional storage chips is expected to accelerate. According to Jeff Kim, head of research at KB Securities in South Korea, if the current momentum continues, profitability for non-HBM memory chips could even surpass that of HBM next year.
Looking ahead, many institutions remain optimistic about the sustainability of this storage cycle. Morgan Stanley predicts a potential super cycle in the memory chip industry next year, as imbalances in supply and demand could emerge in the storage sector. Amidst high market demand, the storage chip industry is also beginning to see a series of chain reactions, including shortages, price increases, and changes in ordering patterns. Analysts also indicate that the price surge for AI server storage products may continue until 2026. Domestic storage companies are expected to benefit from the dual drivers of "price recovery + domestic substitution." Coupled with increased demand for consumer electronics, domestic storage chip companies are experiencing stable improvements in production. This may also explain the recent strong performance of domestic storage chip concept stocks.