Sandisk Targets AI Inference Market With HBF Push
Sandisk is seeking to expand beyond conventional NAND flash into higher-value AI infrastructure with its internally developed HBF technology. The product is designed to ease the memory bottleneck in inference workloads by offering high bandwidth and capacity at a lower cost. Successful data-center adoption could improve Sandisk’s product mix and reduce its earnings exposure to the pronounced supply-and-demand cycles that have historically shaped the memory industry.
At its latest investor day, Sandisk outlined its HBF roadmap and set an 80% floor for its long-term non-GAAP gross margin, underscoring management’s expectations for premium AI products. Mizuho Securities subsequently reiterated its Outperform rating and assigned a $1,900-a-share price target. As of August 2026, investors are focused on customer qualification, the production timetable and whether HBF can deliver the margin contribution implied by the company’s long-range target.
All Coverage
1 original reportsThe Backstory
The history behind this eventSK Hynix, SanDisk Unveil First HBF Specification for AI Inference
High Bandwidth Flash, or HBF, is designed to bridge the gap between high-performance but capacity-constrained HBM and larger, slower conventional NAND storage. Using a stacked architecture similar to HBM, the technology targets the growing memory demands of AI inference, including model weights and cache data. SK Hynix and SanDisk say the approach could improve GPU utilization and reduce the cost of generating each token.
SK Hynix and SanDisk unveiled the first HBF standard specification at the 2026 FMS conference, outlining devices with capacity of up to 512GB and bandwidth of as much as 3 TB per second. The companies plan to use UCIe interconnect technology to link HBF more closely with computing chips. The project remains at an early standardization and development stage, with initial samples expected no earlier than 2028.
SK Hynix and Sandisk Launch HBF Standardization Push for AI Inference Storage
AI inference requires repeated access to massive models and datasets. Conventional SSDs offer lower bandwidth, while HBM is fast but constrained by cost and capacity. The high-bandwidth flash, or HBF, technology being advanced by SK Hynix and Sandisk is designed to sit between the two, balancing capacity, bandwidth and energy efficiency to reduce the total cost of ownership of AI infrastructure and improve scalability.
SK Hynix and Sandisk have formally launched a global HBF standardization initiative and will jointly develop specifications and an industry ecosystem focused on AI inference storage demand. As of July 20, 2026, the companies had not disclosed the value of the partnership, a date for finalizing the standard or a mass-production schedule. Their immediate priority is to encourage adoption across industries and establish HBF as a new storage tier between HBM and SSDs.
Subscribe to Mark Radar Weekly
Every Friday, the week's strongest signals in your inbox. Unsubscribe anytime.
If you search news on Google, you can set Mark Radar as a preferred source—our coverage will show up more often in your results. Set as preferred source on Google →