Morgan Stanley: AI Inference Demand Is Structurally Reshaping the NAND Market
N.R. Finch
Morgan Stanley analyst Joseph Moore says AI inference is pushing NAND up the memory hierarchy into a functional data-center tier, with cloud set to become NAND's largest end market this year.
How is NAND's role changing?
Large language models need massive context caching (KV cache) at runtime — DRAM alone can't hold it all, so the overflow lands on NAND.
This means → NAND is no longer just a "cheaper cost-per-bit alternative." It is moving up the memory hierarchy into a functional complement role.
In plain terms = NAND used to be a cheap warehouse; AI is turning it into a fast staging area — closer to compute, worth more.
Sandisk has long argued NAND and HDD are complements, not substitutes. AI inference demand reinforces that view.
How fast is cloud demand growing?
Morgan Stanley notes cloud will become NAND's largest end market later this year.
Sandisk's cloud business grew 233% quarter-on-quarter in Q1, after a 64% QoQ rise in Q4 — a clear acceleration.
This reflects cloud buyers' far lower price sensitivity versus PC and mobile. This means → the industry's supply-demand dynamics are being redefined by higher-value customers.
TLC, SLC, QLC — how do the three chip types divide the work?
Steady-state demand expectations for TLC — a storage chip balancing speed and capacity — and SLC — the fastest but smallest-capacity chip, used in solutions like HBF — have been revised upward.
The reason: AI inference imposes hard constraints on endurance and latency that QLC — the highest-capacity but slowest chip — cannot yet meet.
In plain terms = AI's demand for "fast" and "durable" has made high-performance chips more entrenched than they were six to nine months ago.
QLC "Stargate" drives begin shipping this quarter. QLC's share will rise over time, but it won't displace TLC and SLC quickly.
What is Sandisk's strategic bet?
Three priorities: ① gross margins that reflect the "fair value" of its technology; ② long-term customer contracts to lock in those margins; ③ sustained investment targeting mid-to-high-teens percentage growth.
This means → Sandisk is not competing on price. It is betting the structural premium from AI inference demand can endure.
Moore maintains an Overweight rating with a $1,750 price target.
Where is the key proof point for this story?
The central question: after QLC volumes ramp, will cloud customers' low price sensitivity to NAND hold?
If cloud buyers start bargain-hunting like PC and mobile buyers, Sandisk's high-margin logic weakens.
In plain terms = Sandisk is telling an "AI makes NAND more valuable" story — the market will score it by how NAND prices behave after QLC ships at scale.
Content is for reference only, not financial advice.