SanDisk and SK Hynix are developing a new memory technology called High-Bandwidth Flash (HBF) that marries NAND Flash—the standard in SSDs—with an architecture modeled after HBM (High Bandwidth Memory), currently used in GPUs and AI accelerators.
This hybrid approach could dramatically reduce the cost and energy demands of AI inference. By combining dense NAND Flash capacity with the speed characteristics of HBM-like architecture, the technology opens the door to running massive AI models with hundreds of trillions of parameters, something that would be prohibitively expensive or impossible with current memory systems.
The practical upside is significant: organizations running AI workloads could lower both operational costs and power consumption, while researchers could experiment with far larger models than existing hardware typically allows. If the partnership achieves commercial viability, HBF could reshape the economics of AI deployment at scale.
