AI Storage Demand Explodes: NVIDIA's CMX to Consume Over 100 Million TB of NAND Next Year, Samsung Ramps Up Supply

As artificial intelligence (AI) technology transitions from training to large-scale inference, demand for NAND flash memory in AI servers is entering an explosive growth phase. According to the Seoul Economic Daily, Samsung has begun supplying NVIDIA (NVDA-US) with its 10th-generation V-NAND and has allocated approximately 60% of its production capacity to the 9th-generation product to meet the demand for the new CMX (Context Memory eXpansion) devices.

CMX Driving Explosive Demand

NVIDIA's CMX devices adopt a standalone external storage architecture, with each unit capable of housing up to 576 solid-state drives (SSDs), delivering a total storage capacity of up to 9,600 TB. Industry analysts estimate that NAND demand from CMX will surge from 35 million TB this year to over 100 million TB next year. This level of demand is equivalent to introducing a new customer the size of Apple (AAPL-US) into the NAND market, posing significant implications for the global supply chain.

Samsung's Three-Generation Technology Strategy

In response to this massive business opportunity, memory giant Samsung Electronics has swiftly adjusted its strategy. The company currently produces over 100,000 V-NAND wafers per month and has actively repositioned its operations to meet NVIDIA's requirements:

V9 (9th Gen) Stable Production: Samsung has allocated about 60% of its V-NAND capacity to V9 production lines, with yields stabilizing above 80%, making it the core product for CMX supply.

V10 (10th Gen) Mass Production: Samsung began mass production of V10 in the first half of this year and has started shipping to NVIDIA. V10 features a 400-layer stacking architecture, increasing storage density by over 50% compared to V9. It also commercially introduces molybdenum (Mo) wiring, replacing traditional tungsten (W), reducing wiring thickness by 30% to 40%.

V11 (11th Gen) Pilot Production: Pilot production has commenced, targeting 400 to 500 layers of stacking. Samsung plans to expand pilot production in the second half of the year to accumulate yield data.

Record Profits and Consumer Supply Constraints

Benefiting from the structural growth in AI storage demand, Samsung is expected to report strong profitability in 2026. However, NVIDIA's large-scale reservation of high-end NAND capacity means supply for other enterprises and consumer markets (such as SSDs for PCs and laptops) will be significantly constrained. As supply tightens, market prices face substantial upward pressure, potentially following the same path as DRAM, which previously experienced shortages and price hikes due to AI demand.

Lee Jae-yong and Jensen Huang Meeting

To solidify long-term collaboration, Samsung Electronics Chairman Lee Jae-yong is expected to meet NVIDIA CEO Jensen Huang in San Francisco soon. In addition to discussing supply arrangements for HBM (High Bandwidth Memory) and next-generation NAND, the two leaders are anticipated to explore comprehensive cooperation, including potential data center construction in South Korea, further strengthening the alliance between the two tech giants.