Analyst Jukan relays Korean industry reporting that Samsung is widening its Nvidia partnership, already spanning DRAM, high-bandwidth memory and foundry, into NAND flash, the chips that store data. Using its scale, Samsung is boosting supply of its ninth-generation V-NAND (V9) to Nvidia while starting mass production of tenth-generation V10, and pushing a 500-layer-class V11. V9 now takes about 60% of its NAND capacity, up sharply from a year ago when older V8 dominated, and monthly output tops 100,000 wafers. Why the rush? Nvidia's next-gen storage system, Context Memory Storage (CMX), lands in the second half. AI inference generates a fast-growing pile of "KV cache", the running context a model keeps in memory. Holding all of it in expensive DRAM is costly, so Nvidia wants to offload much of it to high-speed NAND sitting a tier below. That turns NAND from a commodity into an inference-critical memory layer. The read-through: NAND may be catching the same AI demand wave that lifted HBM, and Samsung's raw capacity makes it the default supplier as the CMX era begins.
Why it matters · If CMX turns NAND into an inference-tier memory, Samsung's sheer capacity makes it the supplier best placed to ride the next leg of the AI memory boom.
Worth asking · Is NAND the next HBM, a structural AI winner, or is CMX-driven demand narrower than the excitement suggests?