Micron Studies Near-GPU NAND as a Capacity Tier for Large AI Models

Micron is reportedly exploring high-endurance NAND modules positioned near GPUs for AI workloads that need more capacity than expensive HBM can provide. The proposed tier would not match DRAM latency or bandwidth, but closer placement could reduce the protocol and distance penalties of accessing conventional storage. If adopted, the architecture could create a new premium NAND use case for large-model inference and training systems.