AI storage demand surges again: Nvidia CMX may consume over 100 million TB NAND next year, Samsung has started supplying

The demand for NAND flash memory in Nvidia AI servers is entering an explosive growth phase, and Samsung Electronics is rapidly transforming this trend into a strategic opportunity by leveraging its production capacity advantage.

According to a report by Seoul Economic Daily on July 20, Samsung Electronics has begun mass production of the tenth-generation V-NAND (V10) and is officially supplying it to Nvidia, while allocating about 60% of its total V-NAND capacity to the ninth-generation product (V9) production line to meet Nvidia's demand for its upcoming generation of storage devices, "Context Memory Storage" (CMX). Industry forecasts predict that the NAND capacity required for CMX will surge from 35 million TB this year to more than 100 million TB next year.

This surge in demand will have a profound impact on the entire NAND market supply landscape. Nvidia’s large-scale lock-in of NAND means that the supply of storage chips available to other enterprises and consumers will shrink significantly, putting upward pressure on market prices.

Meanwhile, Samsung Electronics Chairman Lee Jae-yong is expected to meet with Nvidia CEO Jensen Huang in San Francisco soon, where both parties will hold in-depth discussions on cooperation related to HBM and next-generation NAND product supply, as well as the construction of data centers in South Korea.

CMX Drives Demand Surge for NAND, Nvidia Locks in Samsung Production Capacity

The birth of Nvidia’s CMX stems from the sharp expansion of “Key-Value Cache” (KV Cache) data during AI inference operations. Previously, GPU servers could handle KV cache data internally, but as data volumes continue to grow, standalone external storage devices equipped with thousands of TB-class SSDs have become an essential configuration for AI servers.

According to wccftech, each Nvidia CMX device is equipped with 576 SSDs, with a total storage capacity of up to 9,600TB. Industry estimates put the NAND capacity required for CMX this year around 35 million TB, which will exceed 100 million TB next year. Independent semiconductor industry analyst @jukan05 pointed out on X that a scale of 100 million TB is roughly equivalent to adding a demand source the size of Apple to the NAND market, and its impact on supply should not be underestimated.

Samsung Electronics currently has a V-NAND wafer production capacity of more than 100,000 units per month. According to reports, prior to CMX delivery, Nvidia requested Samsung to expand NAND production, and Samsung has responded actively. Samsung's total NAND output is expected to reach around 250 million TB this year, and the company plans to leverage its global leading capacity to establish its position as the core supplier for Nvidia's CMX.

V9 in Stable Production, V10 in Mass Production, V11 in Pilot Production—Samsung Advances Three Generations in Parallel

Samsung is simultaneously advancing the production processes for three generations of V-NAND products in its product line layout.

For V9, Samsung has increased its production share to about 60% of total V-NAND capacity, with yields stabilizing above 80%, reaching a mature mass production stage. A year ago, V8 still accounted for a considerable portion of V-NAND output, but its proportion has now fallen below 40%, demonstrating Samsung’s rapid response to AI server demand through fast production line switching. In addition, the cleanroom space dedicated to NAND at Samsung’s Pyeongtaek Plant 4 (P4) still has more than half remaining available for future production line expansion.

For V10, Samsung officially began mass production in the first half of this year, although output still accounts for less than 5% of total V-NAND production as it ramps up. V10 adopts an approximately 400-layer stacked architecture, which increases storage density by more than 50% compared to V9’s 286 layers, allowing higher-capacity SSDs to be assembled into the same CMX modules. Notably, V10 will be the first to use molybdenum wiring on a large scale instead of traditional tungsten, reducing wiring thickness by 30%–40%, whereas Samsung had only applied this on a small scale in V9.

For V11, Samsung began pilot production earlier this year, targeting a layer count of 400 to 500, about 100 layers more than V10. Samsung plans to double the size of the pilot production in the second half of the year to collect enough yield data and accelerate the construction of a mass production system.

Supply Concentration Effects Emerge, Consumer Market Under Pressure

The deepening NAND alliance between Samsung and Nvidia is boosting Samsung's performance expectations while further polarizing the supply within the entire memory industry.

Reportedly, Samsung's profit in 2026 is expected to exceed the total cumulative earnings from the past 40 years, largely due to the structural boom in AI server storage demand. However, the other side of this windfall is that when Samsung locks in a large proportion of production capacity for major tech clients like Nvidia, the NAND supply available to other enterprises and consumers will be significantly squeezed, and the risk of rising storage prices will increase accordingly.

Previously, the AI boom first triggered a DRAM shortage and significant price increases, and the NAND market now faces similar supply tightening dynamics. As Nvidia’s CMX shipments continue to expand, this pressure is expected to intensify further.

In the long run, the accelerated development of V11 NAND could provide some relief for the consumer market—theoretically, 500-layer stacking technology could enable large-scale, lower-cost production of high-capacity consumer-grade SSDs, provided Samsung is willing to allocate capacity for the purpose. However, based on current capacity configuration direction, this prospect remains uncertain.

Lee Jae-yong and Jensen Huang Set to Meet, Cooperation Map Continues Expanding

The partnership between Samsung and Nvidia has extended from DRAM and HBM to foundry services and now includes NAND flash memory within its alliance framework, continually expanding both in depth and breadth.

According to reports, Lee Jae-yong is expected to meet with Nvidia CEO Jensen Huang in San Francisco soon. Topics for the talks are expected to include supply arrangements for HBM and next-generation NAND products, as well as cooperation plans for building data centers in areas such as Gwangju, South Jeolla, Korea.

Industry sources said, "The alliance between Samsung and Nvidia spans the entire spectrum of memory and foundry business, and cooperation discussions between the two leaders will go even deeper." As the shipment of Nvidia CMX approaches, the strategic significance of this meeting will be further highlighted.