SK Hynix (SKHY.US) and SanDisk (SNDK.US) have released the world's first HBF standard specification: Alphabet Inc. Class C enters the ecosystem, aiming at the "memory wall" in the AI inference era.

As the competition for artificial intelligence computing power enters a critical phase, storage industry giants SK Hynix (SKHY.US) and SanDisk (SNDK.US) jointly announced the first standard specification for High Bandwidth Flash (HBF) at the "2026 Future Storage and Memory Summit" (FMS 2026) held in Santa Clara, California. This milestone achievement, which took approximately six months and was driven by industry alliances, marks the official entry of a "heavyweight player" into the AI storage arena. HBF is positioned as a new storage tier between High Bandwidth Memory (HBM) and solid-state drives (SSD), aimed at providing solutions that balance high bandwidth and large capacity to meet the growing demands of AI inference. Technical Breakthrough: Addressing the "Memory Wall" Challenge in the AI Era As AI expands from model training to large-scale inference applications, traditional storage architectures are facing severe challenges: while HBM offers extremely high bandwidth, it is costly and has limited capacity, and SSDs, although high in capacity, lack sufficient bandwidth. HBF was born to overcome this "memory wall" bottleneck. According to the announced specifications, HBF achieves a balance between bandwidth and capacity in its technical parameters: Capacity and Stacking: Supports 8-layer and 16-layer NAND chip stacking configurations, with a maximum capacity of up to 512GB. Bandwidth Levels: Categorized into three performance grades (Grade 1~3), with data transfer capabilities covering a scalable range from approximately 0.4TB/s to 3.0TB/s. Open Interconnect Standards: Adopts the industry-standard universal chip interconnect technology UCIe as the connection interface, laying a foundation for the flexible integration of HBF with various processors such as GPUs and CPUs. Notably, this specification is not a proprietary standard but is publicly released to the entire industry through the Open Compute Project (OCP), the world's largest open data center technology organization. This move indicates that HBF will become a universally accepted open standard in the industry, potentially accelerating the maturity of the entire AI storage ecosystem. Ecosystem Assembly: Alphabet Inc. Class C and Tenstorrent Join In Standardization is merely the first step; building an industrial ecosystem is key. Currently, the HBF alliance has successfully attracted Alphabet Inc. Class C (Google) and AI chip startup Tenstorrent to join. SK Hynix will showcase its HBF strategy during multiple events at the FMS 2026 conference: Keynote Speech: SK Hynixs Vice President for solution development, Kim Chun-seong, and Vice President for next-generation product planning, Kang Uk-seong, will jointly present a speech titled "Building Efficient AI Infrastructure Based on Tiered Memories in the Era of Agent-Based Artificial Intelligence." Round Table Discussion: On August 6, SK Hynix, SanDisk, and Alphabet Inc. Class C DeepMind will participate in a special discussion on "Breaking Through the Memory Wall with HBF." Strategic Depth: The Tenth Generation 375-layer 4D NAND Makes Its Debut As a supporting "heavyweight" of this announcement, SK Hynix also publicly showcased the development of its tenth generation (V10) 375-layer 4D NAND flash wafer and products for the first time at the FMS 2026 exhibition. According to SK Hynix, compared to the previous generation, the performance-to-power ratio of this product has improved by 2.5 times, making it particularly suitable for AI data center environments that demand high performance and energy efficiency. The company plans to begin mass production of high-performance, large-capacity enterprise-level solid-state drives (eSSD) based on this NAND flash in early 2027. From the release of the HBF standard to the debut of next-generation NAND technology, SK Hynix is constructing a complete AI storage solution matrix that spans from storage media to new storage tiers. "The Rule Maker" in the AI Storage Arena For investors, the release of the HBF standard carries multiple profound implications. Firstly, SK Hynix is evolving from an "HBM leader" to an "AI storage architecture definitor." Following its dominant position in the HBM market, the introduction of HBF further expands its technological footprint in the AI storage field. Secondly, the strategy of open standards helps accelerate market penetration. By releasing open standards through OCP and attracting ecosystem partners like Alphabet Inc. Class C, SK Hynix is promoting HBF to become the mainstream storage solution in AI inference scenarios. Thirdly, this layout is expected to open up new incremental markets. As AI transitions from training to inference, the demand for storage solutions that balance bandwidth and capacity will grow exponentially. HBF is aimed precisely at this emerging market. Kim Chun-seong, Executive Vice President of SK Hynix, stated: "With the rapid proliferation of AI applications, we are at a point where we must redesign the overall data processing architecture. Through HBF technology, SK Hynix will expand the boundaries of memory and storage, contributing to the construction of a new architecture that enhances overall system efficiency."