GF SEC: The advantages of HBF in the main application are significant, and the commercialization process is accelerating.

GF SEC released a research report stating that HBF has significant advantages in read-oriented applications, and the commercialization process is accelerating. This technology effectively fills the gap between HBM and traditional solid-state drives, providing an ideal solution for capacity and cost-sensitive read-intensive applications. The continuous expansion of AI memory models enhances the capabilities of AI-related applications such as AIAgent. The upstream infrastructure value and importance of AI memory will continue to rise. It is recommended to pay attention to core beneficiaries in the industry chain. Key points from GF SEC: -HBF is a high-bandwidth stacked storage medium based on 3DNAND -HBF is a stacked flash memory technology based on NAND flash memory, stacking multiple NAND flash memory chips together using a packaging/interconnection method similar to HBM, creating a storage layer with high bandwidth and large capacity. -HBF has significant advantages in read-oriented applications -HBF is positioned between HBM and SSD, targeting AI inference scenarios, aiming to provide higher capacity expansion, better energy efficiency, and lower overall ownership costs. (1) Cost and capacity advantages: In the same physical space, a single HBF stack can provide up to 512GB of capacity, an order of magnitude higher than HBM, significantly reducing system unit capacity costs; (2) High read bandwidth, high energy efficiency: According to Sandisk 2025 Investor Day data, the target parameters of the first generation of HBF include: 16-die stacking, single stack 512GB capacity, 1.6TB/s read bandwidth, the read bandwidth can approach HBM levels, and due to no need for continuous DRAM refresh, its static power consumption is significantly lower; (3) Limited write durability: HBF has weaker constraints on read-side lifespan, with write/erase lifespan being the main limitation, making it more suitable for handling data layers with more reads than writes, such as some historical blocks in shared KVCache and some weight/parameter shards, while data that is most sensitive to latency and updates should still be prioritized by HBM. Overall, HBF effectively fills the gap between HBM and traditional solid-state drives, providing an ideal solution for read-intensive applications sensitive to capacity and cost. -Commercialization process of HBF is accelerating According to trendforce data, in August 2025, Sandisk announced a collaboration with SK hynix to promote the standardization of the HBF ecosystem. According to SK hynix's public channel, in February 2026, the two parties announced the start of the next generation of HBF global standardization process under the OCP (Open Compute Project) system. According to trendforce data, SanDisk plans to provide HBF module samples in the second half of 2026 and launch the first batch of AI inference servers integrated with HBF in early 2027; SK hynix will incorporate HBF into its AIN (AI-NAND) product line in the AINB (Bandwidth) direction at OCP 2025; According to trendforce data, Samsung Electronics began early concept design work for their HBF products in 2025, showing that mainstream storage manufacturers are continuing to focus on this technology path. Risk warning AI industry development and demand are below expectations; AI server shipment volume is below expectations, and the progress of domestic manufacturers in technology and product development is below expectations.