Micron explores near-gpu nand flash for enhanced AI model scalability

On September 2, 2026, DigiTimes reported that Micron is researching high-endurance NAND flash modules designed to be placed closer to GPUs rather than distant storage pools. This new approach, referred to as 'near-GPU NAND,' aims to balance storage density and durability, offering a solution for non-low-latency/high-bandwidth workloads. The technology could allow GPUs to directly access hundreds of gigabytes of NAND storage, either within the GPU package or on the PCB, similar to HBM or GDDR7. Compared to traditional TLC or QLC NAND, this variant prioritizes I/O speed and bandwidth over storage density. The storage layer sits between GPU memory and standard storage, acting as a high-speed buffer, making it ideal for large-scale AI language models. Once implemented, LLM inference could no longer be constrained by GPU memory limits.