@dailyanalysts · Tuesday, September 22, 2026 · Single-stock deep dive (Option A)

SanDisk: NAND Is No Longer a Commodity — But $1,870 Is Paying for Perfection

SNDK +5.8% to $1,869.77 on Rosenblatt's $2,400 initiation “Not Your Father's SanDisk” · S&P 100 debut Sep 21 · FY2026 revenue +175% to $20.25B, Q4 gross margin 84.6% · 10 take-or-pay-like contracts back $93.9B of floor revenue
30-second TLDR

The call: SPECULATIVE BUY pullback $1,650–$1,780 toward $2,400, invalidation daily close <$1,420, 1–3 months.

Why now: Rosenblatt reframes NAND as AI-compute-adjacent, and Q4 proves it — datacenter $2.98B (+103% QoQ, +1,298% YoY) with 84.6% margins.

The disagreement: Consensus sees a memory spike to fade; the $93.9B of floor-priced, guaranteed contracts says this cycle has a put underneath it.

The level that changes everything: Micron prints Sep 30 — guide-up confirms the supercycle; any NAND price wobble breaks the $1,420 floor and the thesis.

1. Rosenblatt said the quiet part out loud: inference needs NAND next to compute

Takeaway: the $2,400 target is not a momentum mark — it is a claim that KV-cache and agentic data make flash part of the AI stack.

Rosenblatt analyst Kevin Cassidy initiated SanDisk (SNDK) at Buy with a $2,400 target (~28% above today's $1,869.77 close) on Tuesday, Sep 22, titled “This Is Not Your Father's SanDisk.” The mechanism is specific: larger models, longer context windows and agentic workloads create persistent datasets that must be stored, retrieved and reused — and SanDisk's internal testing found SSD-backed inference used ~75% less energy at ~3x throughput versus DRAM-only setups for KV-cache spillover.

That matters because for 30 years NAND was sold on density and falling cost-per-bit. Cassidy argues AI buyers now pay for density + endurance + performance + supply certainty. The tech hook is BiCS8/BiCS10 (fewer layers for the same die capacity, better areal density) plus High-Bandwidth Flash for inference — built on a 25-year co-development and fab JV with Kioxia. Opinion: this is the first sell-side note that correctly locates NAND in the inference stack rather than hand-waving “AI storage demand.”

Context: SNDK is the S&P 500's top performer in 2026 at +600%+ YTD (52-week low $93.54 on Sep 25, 2025; high $2,354.39 on Jun 22, 2026), spun from Western Digital in Feb 2025, and joined the S&P 100 effective Sep 21, 2026 (replacing Nike). Goldman ($2,200 Buy) and Mizuho ($1,900 Outperform) already sit above consensus; RBC ($1,600 Sector Perform) is the skeptic. Rosenblatt is now the street high.

2. Q4 FY2026 is the receipt: $8.97B revenue, 84.6% margin, datacenter up 437% in a year

Takeaway: this is no longer a PC/smartphone commodity story — datacenter went from $960M to $5.15B in one fiscal year.

Read the primary source — the Q4 FY2026 press release (Aug 5, 2026) — not the headline. Fiscal Q4 (ended Jul 3): revenue $8.965B, +51% QoQ, +372% YoY, roughly two-thirds from pricing, one-third from volume. GAAP diluted EPS $43.97; non-GAAP $39.25. Full FY2026: revenue $20.25B (+175% YoY), GAAP EPS $73.76, non-GAAP $70.88, gross margin 71.5% vs 30.1% a year ago.

End marketQ4 FY26QoQYoYFY26
Datacenter$2,977M+103%+1,298%$5,153M (+437%)
Edge (PC/mobile/auto)$5,432M+48%+392%$12,160M (+195%)
Consumer$556M-32%-5%$2,935M (+29%)

Two details carry the thesis: Q1 FY2027 guide is $10.3–$10.8B revenue at $44–$46 non-GAAP EPS — i.e., growth is still accelerating sequentially — and the balance sheet flipped to zero long-term debt, $4.76B cash, with a $15.5B remaining buyback authorization ($4.5B already repurchased in Q4). Opinion: management is behaving like a company that believes peak earnings are not this quarter. The load-bearing risk: Edge is still 60% of FY26 revenue, and management itself expects mid-teens unit declines in PC/smartphone this calendar year as higher memory prices bite — so datacenter must keep offsetting a soft consumer edge.

3. The contracts are the moat: $93.9B of floor revenue changes what “cyclical” means

Takeaway: 10 New Business Model agreements — with floors, ceilings, volume commits and $16.5B of guarantees — put a put under half of FY27 bits.

Since the April call SanDisk signed 10 NBM deals with 8 datacenter + edge customers, weighted-average duration >4 years, with quarterly volume commitments, fixed/variable pricing, floors/ceilings and financial guarantees. Minimum contracted revenue at floor pricing: $93.9B, plus $91.1B of remaining performance obligations and $16.5B of guarantees. Rosenblatt estimates coverage of >50% of FY27 bits and ~two-thirds of FY28 bits; SanDisk says ~two-thirds of FY28 bits.

Cassidy's phrasing is careful — it “dampens the amplitude without eliminating” the cycle — and that is honest. Floors protect downside; ceilings (like Micron's SCAs, where ~40% of revenue once fully executed sits at or near 2Q26-level ceilings per TrendForce Sep 22) cap upside if spot keeps ripping. Susquehanna's Mehdi Hosseini expects +60% sequential NAND pricing this quarter and +25% next — great near term, but it means FY28 $240 EPS (the base for Rosenblatt's 10x = $2,400) assumes pricing normalizes to an 80.2% gross margin, not today's 84.6%. Opinion: treat NBM as a volatility collar, not a fixed annuity — the floor is real, the ceiling is the reason to demand a PEG discount.

How I value it (checkable): Rosenblatt FY28 EPS $240 × 10x = $2,400, framed as a 30% discount to 1.0x PEG. Management's 2028–30 framework (mid-to-high-teens revenue growth, ~80% non-GAAP GM, ~75% OM, ~50% adj. FCF margin) implies FY30 EPS ~$300. At $1,870 today that is ~6.2x FY30 EPS — cheap if the framework holds, expensive if NAND mean-reverts to 40% margins. My fair-value range: $2,100–$2,500 on FY28 contracted earnings; $900–$1,100 if pricing reverts. The spread is the trade.

4. The trade: don't chase the S&P 100 pop — buy the Micron-week wobble

Takeaway: the setup is SPECULATIVE long into strength confirmation, sized for a 55%-drawdown stock.

SNDK fell 55% from its June $2,354 high into summer, then ripped 26% in a day around Q4 and another 40%+ in the past month into index inclusion. Beta is 3.8. CEO David Goeckeler sold $53.3M (33,841 shares, Sep 17) and $51.7M (Sep 14) under a 10b5-1 adopted May 29 — pre-planned, but it tells you insiders are harvesting volatility too. And Micron reports Sep 30 (Q4 FY26, consensus ~$50.4B revenue, ~$31.14 EPS, 86–87% GM) — the whole memory complex will reprice on its guide and HBM4/SCA commentary.

Flagship: EQ-SNDK-1Level
Entry zone$1,650–$1,780 (pullback to pre-initiation / index gap; no chase above $1,910 intraday high)
Target$2,400 Rosenblatt / tech-roadmap validation; stretch $2,550 on Q1 FY27 beat + Micron guide-up
InvalidationDaily close below $1,420 (~24% below here; breaks Aug breakout + 50-day structure)
Timeframe / conviction1–3 months into Q1 FY27 print + Micron Sep 30; SPECULATIVE (one signal: contracted floor; pricing durability unconfirmed)
Audience / sizeTrading sleeve only; half-starter given beta 3.8 and CEO selling; add only on daily hold above $1,910

Bull (30%): Micron guides up, NAND +25% lands, SNDK datacenter hits Rosenblatt's $21.7B FY27 path — $2,400–$2,550. Base (45%): in-line Micron, SNDK consolidates $1,650–$1,950 into Q1 print — hold starter. Bear (25%): Micron flags ceiling drag or Edge unit collapse, NAND rolls — $1,420 breaks, opens $1,100–$1,200 (prior breakout shelf). No contradiction with open EQ-MU-adjacent work: both are long memory-supercycle; SNDK is the higher-beta NAND leg, so size it smaller than MU.

What would prove me wrong: (1) daily close <$1,420; (2) Micron Sep 30 gross-margin guide <84% or explicit NAND price-peak language; (3) SanDisk Q1 FY27 datacenter revenue <$3.2B or any NBM cancellation/guarantee release; (4) Kioxia JV disruption. Any one triggers exit — no second chances on a 600% runner.

Second-order effects most miss: if NAND stays system-critical, laptop/smartphone OEMs (Dell, HP, Apple via Kioxia) eat margin compression — Morgan Stanley already demoted HPE/HP/Dell on memory cost; enterprise SSD (STX, WDC) gets pricing umbrella; and hyperscaler capex ($1.17T combined 2028 per NICE via TrendForce) increasingly funds memory, not just GPUs — which re-rates Kioxia supply-chain leverage over SanDisk's JV terms.

Appendix — check the work

Contents: A. Data snapshot · B. Model & load-bearing assumption · C. Sources

A. Data snapshot (closing/intraday Sep 22, 2026; post-20:00 UTC rule not triggered — morning ET pull)

ItemLevelTimestamp/source
SNDK$1,869.77 (+5.84%, O $1,759.23 H $1,909.48 L $1,756.82, prev $1,766.64)financial-data handler 16:01 UTC Sep 22
MU / WDC / STX$1,077.12 (+3.18%) / $452.27 (+0.91%) / $889.46 (+1.38%)same pull
SNDK fundamentalsmcap ~$258.7B; TTM P/E ~22.6x; rev growth +175.3% YoY; GM 71.47%; 52w $93.54–$2,354.39handler fundamentals Sep 22
MarketSPY $773.09, QQQ $744.59, BTC $86,438, F&G 78 Extreme Greedhandler Sep 22
Q4 FY26 / FY26Q4 rev $8.965B; FY26 $20.248B; Q4 GM 84.6%; Q1 FY27 guide $10.3–$10.8B, EPS $44–$46company IR Aug 5
NBM10 deals, $93.9B floor, $91.1B RPO, $16.5B guarantees, >50% FY27 bitsRosenblatt via Benzinga/Invezz Sep 22

B. Model — FY28 bridge and where it breaks

Rosenblatt: datacenter $21.7B FY27 → $28.6B FY28 (~48% of total) → FY28 EPS $240 → 10x = $2,400 (30% discount to 1.0x PEG for cyclicality/execution). My check: $240 × 155–157M diluted shares ≈ $37–38B net income → needs ~$55–60B revenue at ~65–70% net margin — plausible only if 80%+ GM holds. Load-bearing assumption: NAND pricing stays ≥2Q26 ceiling levels through FY27. If wrong (mean reversion to 45–50% GM), EPS is ~$90–$110 and 10x = $900–$1,100. Watch Micron Sep 30 GM guide as the proxy.

C. Sources (linked, working)