SK Hynix and Sandisk Begin Global Standardization of Next-Gen Memory ‘HBF’

Key Takeaways

• HBF sits between HBM and SSD.
• Standardization under OCP workstream launched.
• Targets AI inference scalability and power efficiency.
• Reduces total cost of ownership for AI systems.
• Demand projected to surge by 2030.

Summary

SK hynix and SanDisk announced a joint initiative to standardize High Bandwidth Flash (HBF), a new memory tier positioned between high‑bandwidth memory (HBM) and solid‑state drives (SSD). The partners will create a dedicated workstream under the Open Compute Project to develop industry‑wide specifications, leveraging their expertise in HBM and NAND manufacturing. HBF aims to bridge performance and capacity gaps, delivering higher scalability and power efficiency for AI inference workloads. Analysts expect demand for such hybrid memory solutions to accelerate sharply toward 2030.

Pulse Analysis

The AI industry is rapidly moving from model training to inference, where latency, power draw, and data throughput become critical. Traditional memory hierarchies—HBM for speed and SSDs for capacity—leave a performance gap that hampers large‑scale inference services. High Bandwidth Flash (HBF) is engineered to occupy this middle ground, offering bandwidth closer to HBM while maintaining the density of SSDs, thereby enabling more efficient data pipelines for real‑time AI applications.

By anchoring HBF standardization within the Open Compute Project, SK hynix and SanDisk are fostering an open, vendor‑agnostic framework that can be adopted across data‑center ecosystems. This collaborative approach leverages both companies' deep experience in HBM stack integration and NAND flash packaging, accelerating the path from prototype to mass production. A unified specification reduces fragmentation, simplifies system‑level design, and encourages broader ecosystem participation, from server manufacturers to cloud providers.

Market analysts project a steep rise in demand for hybrid memory solutions as AI inference workloads dominate by the early 2030s. Standardized HBF promises lower total cost of ownership by cutting power consumption and reducing the need for multiple discrete memory components. For enterprises, this translates into more scalable, cost‑effective AI services, while chipmakers gain a new reference architecture to differentiate their offerings. The move positions SK hynix and SanDisk as pivotal players in the next generation of AI‑optimized hardware infrastructure.