AI Inference Demand Drives NAND Market Revenue to Record Highs
Nashnova编辑部
Global NAND revenue hit a record in Q2 2026, surging roughly fivefold year-on-year as AI workloads shifted from training to inference — the resulting hunger for fast, high-capacity storage has tightened enterprise SSD supply and is reshaping pricing and competition across the memory chain.
Why did revenue jump fivefold?
AI workloads are structurally shifting from training to inference. This means → models are no longer just learning; they are being used at massive scale, and every inference call reads and writes large volumes of data in real time.
Inference requires storage systems that can handle key-value caches — intermediate results an AI model needs to retrieve instantly — and full datasets at high speed and low power. Enterprise SSDs (solid-state drives built for servers and data centers) are now in short supply.
In plain terms = AI training mainly consumed GPU compute; now AI inference is consuming storage too — NAND chips have become the new bottleneck.
How has the supplier pecking order changed?
Samsung Electronics led with a 25% shipment share but faces a capacity squeeze: the company is prioritizing higher-margin DRAM, compressing NAND output.
SK Hynix followed at 22%, powered by subsidiary Solidigm, whose bit shipments grew 40% quarter-on-quarter.
YMTC (长江存储) rose to third at 14%, up 22% year-on-year and 5% quarter-on-quarter. This reflects two forces stacking: a global supply shortage opened the window, while YMTC expanded shipments to domestic Chinese OEMs.
What does YMTC's technology push signal?
YMTC began volume production of 267-layer 3D NAND in the same quarter and is advancing 300-plus-layer products based on its Xtacking architecture — a process that fabricates memory cells and logic circuits separately, then bonds them together.
This means → YMTC is not merely riding a supply shortage; it is locking in share through technology iteration. More layers = more data per unit area = a steeper cost advantage.
A caveat: Micron ranked fifth by shipments but still led YMTC in revenue. In plain terms = shipping more does not mean earning more — unit price and product mix matter just as much.
Enterprise SSD prices are surging — who wins, who loses?
Enterprise SSDs accounted for 48% of quarterly shipments; the supply-demand imbalance pushed consumer NAND average prices to record highs as well.
Kioxia steered over 30% of shipments toward servers, yet rising enterprise SSD prices dampened buyer willingness, capping its growth. This means → the price surge is a double-edged sword — margins widen, but customers start to hesitate.
This reflects the market's central tension: red-hot demand pulling against the limits of price tolerance.
What decides who wins before 2027?
Counterpoint projects that by end-2026, servers will absorb more than 50% of global NAND bit shipments.
The firm states plainly: raw shipment volume is no longer the core competitiveness metric. In plain terms = the race is no longer about who ships the most — it is about who has the most valuable product mix.
This means → suppliers that can build and sell high-value enterprise SSDs effectively will lead through 2027 — the era of winning on volume alone is ending.
Content is for reference only, not financial advice.