Haitong: AI-driven storage demand growth helps industry recovery by improving supply and demand.

Haitong released a research report stating that the storage industry is experiencing significant growth in demand driven by the rapid development of AI technology. At the same time, the supply side is gradually improving the supply-demand structure through the reduction measures of NAND Flash original factories, and the industry is expected to see a recovery. The rapid development of AI technology in areas such as AI smartphones, AI PCs, AIoT, and smart cars has further increased the single-device storage capacity. Biwin Storage Technology (688525.SH) is entering high-value-added areas such as AI glasses through the layout of storage + advanced packaging and testing integrated, with huge potential for future development. Haitong's main points are as follows: Optimistic about the benefits of the storage industry: (1) Demand side: DeepSeek accelerates the comprehensive flowering of AI edge applications, driving the demand for storage in AI+ semiconductors; (2) Supply side: Various NAND Flash original factories have restarted production cuts, laying the foundation for a rebound in storage prices in the second half of the year. The firm believes that storage module companies will significantly benefit from the improvement in supply-demand structure and expectations of price rebound, especially high-quality companies that can synergize advanced packaging and testing with AI edge areas. Biwin Storage Technology's layout of storage + advanced packaging and testing integration, fundraising for wafer-level advanced packaging and testing manufacturing, and covering well-known overseas AIoT customers, has entered the high ASP field of AI glasses. The profound impact of DeepSeek mainly lies in two dimensions: 1) Under the same training effect, the low-cost solution provided by DeepSeek will provide the industry with new training ideas. Major companies will be inspired by its depth when training large models, while domestic cloud providers will also shift their attention to the potential development of domestic computing cards; 2) The open-source nature of DeepSeek will bring global AI equality, and the power of local deployment of open-source models by both B and C ends will be greatly enhanced. In other words, the demand for computing cards for cloud-based inference and local storage will continue to increase to support large models, becoming a medium- to long-term industry trend. The firm believes that application ends are expected to benefit deeply, while storage demand will see a comprehensive increase along with the implementation of AI+ applications: 1) AI smartphones: DeepSeek's cost-effective open-source solution is expected to boost the edge computing configuration. While cloud-side computing dominates, the edge-side collaboration is expected to be significantly strengthened; in addition, AI smartphone brand manufacturers are expected to gradually strengthen their voice under the background of continued integration of data with third-party app developers; 2) AI PCs: The penetration of AI PCs in B and C terminals was constrained by data security and high consumption thresholds in the early stage. DeepSeek can help B-end private cloud deploy local large models for comprehensive privacy protection, reducing the need for data transmission by processing data on local devices, thereby reducing the risk of data leakage. C-end users will also strengthen their deployment intentions locally, and the growth logic of AI PCs is expected to be gradually reshaped; 3) AIoT: Edge applications are expected to blossom, and the penetration of AR glasses is expected to further enhance; 4) Smart car intelligence: The firm believes that smart cars are expected to introduce open-source large models to become larger AIoT terminals, and AI may deeply empower the entire lifecycle of cars. By 2024, the average single-device storage capacity for various AI extended applications such as smartphones, servers, and laptops is expected to grow. The firm believes that by 2025, with further penetration of AI applications, the single-device storage capacity will continue to grow, and with the improvement in storage production cuts and demand, storage prices are expected to rebound. In early 2025, all NAND Flash original factories will take more resolute production reduction measures, reduce the annual production scale, and hope to effectively reduce the growth rate of supply bits, mainly through lowering the capacity utilization rate in 2025 and postponing process upgrades to achieve the goal of production reduction, which is conducive to quickly alleviating the pressure of market supply-demand imbalance and laying the foundation for price rebound. In Q1 2025, the NAND Flash market continues to face the challenge of oversupply, leading to continuous price declines and suppliers facing losses. The firm believes that the supply-demand structure of the NAND Flash market is expected to significantly improve in the second half of the year, including original factory production cuts, clearance of smartphone inventories, AI, and DeepSeek effects, which will drive NAND Flash demand and thus ease the oversupply situation, with price rebound expected in the second half of the year. As for DRAM, the firm believes that with the improvement in network demand, DRAM is also expected to see a price rebound in the second half of the year. Biwin Storage Technology's layout of storage + advanced packaging and testing integration, fundraising for wafer-level advanced packaging and testing manufacturing: Biwin Storage Technology has built a business model focusing on the semiconductor storage industry chain, establishing an "integrated research and testing" model. The company will issue stocks to specific objects to raise funds totaling no more than 1.9 billion yuan, which will be used for the expansion and construction project of Huizhou Biwin's advanced packaging and testing and storage manufacturing base and wafer-level advanced packaging and testing manufacturing project. The company uses its subsidiary Huizhou Biwin as its advanced packaging and testing and storage manufacturing base. Huizhou Biwin has leading domestic packaging technology, currently possessing the production capacity for advanced processes such as 16-layer stacked Die, 30-40m ultra-thin Die, and multi-chip heterogeneous integration, reaching the international first-class level. Huizhou Biwin has provided OEM services to wafer manufacturers, IC design companies, and storage manufacturers, creating new business growth points. Biwin Storage Technology covers overseas well-known AIoT customers, entering the high ASP field of AI glasses: Biwin Storage Technology's ePOP series products are currently being used by well-known companies such as Google, Meta, and Xiaotiancai in their smart watches, VR glasses, and other smart wearable devices; the company's eMCP and uMCP series products have been widely recognized by smartphone and tablet clients. According to the Weishen Wellsenn XR WeChat public account, in the BOM composition of Ray-Ban Meta smart glasses, the hardware cost of ROM+RAM is $11, accounting for 12.3% of the mainboard chip cost ($89.1),..EMCP storage module accounts for a high proportion of semiconductor value in AI glasses, at 12.35%.Risk Warning: End demand recovery is lower than expected; AI application penetration is lower than expected; storage terminal prices are rising less than expected; domestic semiconductor substitution process is lower than expected.