Sandisk's HBF claims 16x HBM capacity

Sandisk's announcement targets the growing pressure on memory systems from AI inference, where conventional DRAM-based HBM may be limited by capacity and cost. By proposing a NAND-based alternative, Sandisk highlights a trade-off: bandwidth parity with HBM, but with a capacity advantage that could better suit workloads requiring large memory footprints. The claim of similar cost is significant because it addresses the primary barrier to deploying high-capacity memory in AI systems. However, these figures are Sandisk's own assertions, not independently verified, and actual performance in real-world inference scenarios remains to be seen. The positioning of flash closer to processors suggests a potential shift in memory architecture, but whether it will displace or complement HBM in practice is yet to be determined.