The next evolution of AI memory architecture is taking shape. 

 SK hynix and SanDisk have jointly announced the first standard specifications for High Bandwidth Flash (HBF), marking a major milestone toward building an open ecosystem for next-generation AI memory solutions. 

 The initial HBF specification supports: Up to 512GB capacity with 8-high and 16-high NAND stack configurations 

 Three bandwidth tiers (Grade 1–3), delivering approximately 0.4TB/s to 3.0TB/s bandwidth Industry-standard UCIe interconnect, enabling integration with a wide range of processors, including GPUs and CPUs Unlike HBM, which stacks DRAM to maximize computing bandwidth, HBF leverages stacked NAND flash to create a new memory layer between high-performance HBM and high-capacity SSD storage. 

 The industry outlook suggests HBF is not a replacement for HBM, but a complementary technology that could help address key AI infrastructure challenges: 

? Reducing memory system costs 

? Expanding AI data capacity 

? Improving scalability for large-scale AI workloads 

 SK hynix also revealed its 375-layer, 10th-generation (V10) 4D NAND, expected to enter mass production in early 2027, targeting improved performance per watt for data center applications. 

 The future of AI will not only depend on faster processors — it will depend on smarter memory architectures.