Sandisk Introduces High-Bandwidth Flash for AI Inference
HBF marks a shift in memory technology positioning, targeting AI efficiency amidst emerging models.
What Changed
Sandisk, together with SK Hynix and Synopsys, has announced the development of high-bandwidth flash (HBF) technology. This innovation positions itself uniquely in memory hierarchy, between DRAM and NAND. Compared to high-bandwidth memory (HBM), HBF is designed for AI inference applications and plans to roll out samples by late 2026, marking the first concerted effort to use flash in this manner since Sandisk's foundational work in storage devices.
Strategic Implications
The introduction of HBF could alter dynamics in the AI hardware landscape. By moving storage closer to processing units, this approach decreases latency and increases energy efficiency, giving adopters like Synopsys a competitive edge. However, this shift might lessen the reliance on traditional DRAM suppliers, while SK Hynix could strengthen its market position by supplying co-designed technologies.
What Happens Next
Standardization through the Open Compute Project (OCP) reflects a potential push for broader industry adoption and collaboration. If successful, first inference devices utilizing HBF are expected by early 2027, likely prompting competitors to accelerate similar offerings. We anticipate industry discussions shaping around integration complexities and further development of nonvolatile memory (NVM) standards.
Second-Order Effects
HBF technology could influence supply chains by increasing demand for specific materials used in stacked die architectures. Moreover, it may spur regulatory discussions on international collaboration in tech standards, given the strategic importance of AI capabilities and the involvement of multinational entities like Sandisk and SK Hynix.
Free Daily Briefing
Top AI intelligence stories delivered each morning.