NVIDIA CMX to Drive Over 100 Million TB of NAND Demand Next Year, Samsung Locks In Supply
Claire Weston
Nvidia's external storage device CMX is set to push NAND demand from 35 million TB this year to over 100 million TB next year. Samsung has begun shipping first and committed roughly 60% of its V-NAND capacity to ninth-generation production, signaling a structural supply squeeze across the NAND market.
What is CMX, and why does it need so much flash memory?
CMX — Context Memory Storage — is an external storage device built for AI inference. A single unit holds 576 SSDs with a combined capacity of 9,600 TB.
It exists because of the explosive growth of key-value cache (KV cache) data — the intermediate data AI models generate during inference. Larger models and longer conversations mean far more data to store.
This means → CMX is essentially a bolt-on memory warehouse for large AI models. Its demand scales directly with AI deployment volume.
How big is 100 million TB?
The industry expects CMX's NAND requirement to jump from roughly 35 million TB this year to over 100 million TB next year — nearly a threefold increase.
Independent analyst @jukan05 noted that 100 million TB is roughly equivalent to adding an Apple-sized buyer to the NAND market from scratch.
In plain terms = the NAND market is gaining a new customer the size of Apple overnight. The supply-side pressure is hard to overstate.
How is Samsung handling the order?
Samsung's monthly V-NAND wafer output exceeds 100,000 wafers. Total NAND production this year is projected at roughly 250 million TB.
Samsung has allocated about 60% of its V-NAND capacity to its ninth-generation line (V9). V8's share has dropped below 40%.
Nvidia asked Samsung to expand capacity ahead of CMX shipments. Samsung is actively responding, aiming to secure its position as CMX's core supplier.
V9, V10, V11 — what does advancing three generations at once mean?
V9: capacity share at 60%, yield stable above 80% — the current shipping workhorse.
V10: mass production began in H1 this year. Current share is under 5%, still ramping. It uses roughly 400 layers, boosting density over 50% versus V9's 286 layers. V10 marks the first large-scale commercial use of molybdenum — a metal replacing traditional tungsten wiring — cutting wire thickness by 30% to 40%.
V11: pilot production started early this year, targeting 400 to 500 layers. Pilot scale will double in H2 to accumulate yield data.
This means → Samsung is running a "mass-produce one, ramp one, pilot one" cadence, but V10 and V11 cannot deliver volume in the near term.
What does this mean for ordinary consumers?
Samsung is locking a large share of capacity for Nvidia and other major tech clients. Consumer-facing NAND supply will be visibly squeezed, pushing storage prices higher.
The AI boom already triggered a DRAM shortage and sharp price increases. NAND is now following the same supply-tightening logic.
Accelerated V11 development could theoretically ease the consumer market, but given current capacity allocation, that outlook remains difficult to be optimistic about.
What comes next?
Samsung chairman Jay Y. Lee is expected to meet Nvidia CEO Jensen Huang in San Francisco soon. The agenda is expected to cover HBM (high-bandwidth memory) and next-generation NAND supply, as well as data-center partnerships in Gwangju, South Korea.
Samsung's 2026 profit is projected to exceed the company's cumulative earnings over the past 40 years, driven largely by the structural expansion of AI server storage demand.
This reflects a clear trend: NAND supply is concentrating toward AI servers. Whether this eases before V11 reaches mass production is the key checkpoint for the broader storage-market price trajectory.
Content is for reference only, not financial advice.