JPMorgan Upgrades Sandisk to "Overweight" with $2,250 Price Target
Nashnova编辑部
JPMorgan upgraded SanDisk to Overweight with a $2,250 target — roughly 47% above the latest close; the real thesis is not the price but whether long-term agreements can shift NAND valuation from peak-earnings multiples to predictable free cash flow.
Why is JPMorgan turning bullish now?
JPMorgan had rated SanDisk Neutral at initiation; the upgrade to Overweight carries a $2,250 target.
The two ratings were set at different points, share prices, and earnings baselines — a mechanical "Neutral-to-Overweight" comparison oversimplifies.
This means → the sell-side view has shifted directionally: long-term supply agreements (LTAs) are now treated as a structural variable, not a sign of customers front-loading orders at the cycle peak.
In plain terms = if the LTA logic holds, the market's valuation anchor for SanDisk moves from "how high can peak EPS go" to "how stable is future cash flow."
What do ~$94 billion in LTAs actually protect?
SanDisk has signed LTAs with 8 customers worth roughly $94 billion at floor prices, with a weighted-average tenor exceeding four years and about $16.5 billion in financial guarantees.
JPMorgan breaks the contract stack into three layers: Layer 1 is total contract value — it answers "how much to produce." Layer 2 is remaining performance obligations (quarter-end $59.8 billion, or $91.1 billion including post-quarter deals) — it tracks whether contract drawdowns match new signings. Layer 3 is the financial guarantee cushion when customers under-fulfill.
This means → the three layers offer meaningful protection, but not risk-free revenue — the "~80% gross margin at floor prices" is an economic estimate, not an unconditional guarantee.
In plain terms = if customers shift to lower-priced products, production costs rise, or large buyers concentrate volume at lower pricing tiers, reported gross margins can still fall short of model assumptions.
How is AI inference reshaping NAND market size?
JPMorgan estimates the NAND market growing from roughly $70 billion in 2025 to over $300 billion in 2026; SanDisk management projects a potential ~$500 billion by 2027.
The core assumption: data centers in the AI inference phase adopt enterprise SSDs as persistent key-value cache and low-cost storage for massive model data, lifting data centers' share of NAND revenue from about 30% in 2025 to roughly 50% in 2026.
This reflects a different pull from AI training — training consumes compute; inference consumes storage capacity and read/write throughput.
Can supply discipline actually hold?
Current NAND capacity is still about 30% below the prior peak, yet utilization is near 100%, and some post-2027 supply has already been allocated.
LTAs can constrain SanDisk's own capacity releases, but cannot stop peers from expanding when margins are high — supply discipline is a variable that must be re-verified every quarter.
Traditional PCs, smartphones, and consumer electronics still account for 60%–70% of industry-wide NAND bits. This means → if smartphone capacity upgrades, PC refresh cycles, and consumer SSD sales all weaken at once, inventory builds first in legacy products, then drags on blended gross margins through mix and pricing.
Where do BiCS10 and high-bandwidth flash stand?
The latest BiCS10 generation uses a 332-layer, 2 TB QLC design; versus BiCS8 it delivers roughly 60% higher storage density, ~100% more read/write bandwidth, ~75% better energy efficiency, and about 65% more bits per wafer.
In plain terms = bit growth comes mainly from technology migration rather than new fabs, keeping capex relatively contained — the company's disclosed generational compound bit-growth rate is 27%.
High-bandwidth flash (HBF) — a new architecture that places flash memory close to the GPU, similar to HBM — has larger upside but thinner evidence: company simulations show capital efficiency up to 8× current solutions at minimum config, and GPU efficiency up 2× at maximum output.
The first HBF die has taped out, with samples planned for delivery to inference-device customers in 2027; until then, HBF sits outside the company's financial model as a longer-term technology reserve.
What does the market need to verify next?
Two critical checkpoints: whether LTAs can genuinely smooth NAND's cyclical swings, and whether legacy end-market demand can digest inventory without dragging margins.
This means → the viability of JPMorgan's "predictable cash flow" pricing framework depends not on the LTAs themselves, but on whether industry supply-demand outside the LTAs cooperates.
In plain terms = whether LTAs are a durable moat or merely a peak-cycle volume lock will only be answered when the next downturn arrives.
Content is for reference only, not financial advice.