Business
Know the Business
SanDisk is a pure-play NAND flash storage maker — the most cyclical, capital-intensive sub-industry in technology, currently in the most violent up-leg of a cycle ever recorded. The stock has compounded ~40x since its February 2025 spin-price close (~$50) and ~67x off the April 2025 post-spin low ($29.62), and now trades on peak-cycle margins. The central question every investor must answer: is this the start of a structurally higher-margin AI-storage business locked in by multi-year hyperscaler contracts, or the classic NAND-cycle top where consensus pays a peak multiple for peak earnings.
At $1,980/share and 148M shares outstanding (per Q3 FY26 10-Q), SanDisk's market cap is ~$293B — larger than Western Digital, Seagate, NetApp, and Pure Storage combined. The market is no longer underwriting a cycle; it is underwriting a regime change.
1. How This Business Actually Works
SanDisk turns NAND wafers made by a Japanese joint venture into branded storage products. It owns no fabs, but is contractually entitled to half of the JV's bit output — and contractually obliged to half of the JV's fixed costs whether or not it takes those bits. That single asymmetry explains nearly every line of the income statement.
The economic engine is ASP × bits sold − fixed JV cost share − operating cost. Bits sold is roughly half of whatever Flash Ventures produces; SanDisk's cost-per-bit falls with each node generation; ASP is set by the industry's supply/demand balance. When ASP rises, almost every incremental dollar flows through gross profit because the cost base barely moves. The Q3 FY2026 print is what that looks like in the extreme.
Revenue went up 3.5x quarter-over-quarter from Q3 FY25 to Q3 FY26, while cost of goods sold actually fell. The reason is mechanical: SanDisk's cost base — wafer cost at the JV, plus operating expenses — is largely fixed quarter-to-quarter; what changed was the price per bit. In NAND, the incremental margin on price-driven revenue is effectively 100%. That works in both directions: in FY2023, the same operating leverage took gross margin from 33% to 7% and net income from +$1.1B to −$2.1B.
The Kioxia JV is the most important non-obvious feature of SanDisk. It makes the company "fab-light" on the income statement (capex is only 3–4% of revenue) but fully exposed to fab fixed-cost absorption through the half-share of JV fixed costs. SanDisk cannot meaningfully diversify wafer supply — both partners are contractually restricted from third-party manufacturing while the JVs operate, currently through 2029 and 2034.
2. The Playing Field
NAND is a five-supplier oligopoly. Of the five, Samsung is too-conglomerate to compare cleanly and YMTC is private and constrained by US export controls. That leaves four direct competitors plus two storage substitutes and two downstream system buyers — the peer set below is the playing field that actually matters.
Two things jump off this table. First, SanDisk's LTM operating margin (40.7%) is now higher than Micron's, higher than Seagate's, and approaching SK hynix's — a level reached only because Q3 FY26's 69% operating margin pulled the average up. Second, the EV/LTM-sales multiple of ~22x is higher than every disk-storage peer and only modestly below Micron — and is being paid against revenue running at four times the FY2023 trough rate. The market is paying a peak multiple on already-peak earnings.
The only true economic substitute for SanDisk is Kioxia — the JV partner with the same wafer supply, similar cost structure, similar product mix, and similar lack of DRAM. Kioxia's FY ending March 2026 operating margin (37%) and revenue trajectory closely mirror SanDisk's. Micron and SK hynix earn high margins right now partly because HBM (DRAM stacked for AI accelerators) is in extreme tightness — SanDisk has no HBM, so its margin must come entirely from NAND ASPs.
"Good" in this industry is not a margin level — it is a position in the cycle. Through-cycle operating margin for SanDisk has historically been roughly 5–10%. Through-cycle for Micron is mid-teens because of DRAM diversification. The bull thesis for SanDisk requires that the NEW through-cycle margin is structurally higher than the old one because (a) datacenter mix is durably larger and (b) NBM agreements lock in pricing visibility.
3. Is This Business Cyclical?
Deeply. NAND is a commodity cyclical — pricing, margin, and free cash flow swing harder than the demand line, because fab supply is added in $10–20B lumps with 12–18 month lead times while demand from PCs, phones, autos, and data centers moves on its own clocks. The four years FY2022–FY2026 contain a full cycle in miniature: peak → trough → recovery → AI super-cycle.
The cycle hits in three places at once: ASP (gross margin from 33% to 7% to 78%), utilization (the JV's fixed costs become "underutilization charges" when SanDisk slows wafer take — $252M of charges in FY2024), and working capital (accounts receivable went from $605M at the FY23 trough to $2.73B at Q3 FY26, a $2.1B headwind funded out of incremental gross profit). Capital markets compound it: SanDisk issued $2.5B of debt in FY2025 to fund the spinoff at a moment when operating cash flow was barely positive, then repaid all of it within three quarters once cash poured in.
Four years of cumulative free cash flow before FY2026 were negative. Nine months of FY2026 produced $4.4B of free cash flow. This is what extreme operating leverage looks like in a fab-dependent business — and why thinking about SanDisk on a single-year basis is dangerous in either direction.
4. The Metrics That Actually Matter
Forget P/E, ROE, and quarterly EPS through this kind of cycle. The five metrics below are what separate signal from noise in a NAND maker.
Three of these five (datacenter mix, NBM count, contract liabilities) are the new metrics that did not exist in the WDC-era NAND segment. They are the bull case for why the cycle peak might hold longer than past peaks. The other two (gross margin, JV underutilization) are the old metrics that warned of every prior downturn — both currently flashing green, but both will turn first if the cycle does.
The metric NOT to use: trailing P/E. The denominator just went from negative $11.32 EPS (FY25) to $29.42 EPS (9M FY26 diluted) to a guided ~$30+ EPS in Q4 FY26 alone. P/E is mathematically meaningless when the earnings line is moving by hundreds of percent per quarter. Use EV/Sales, EV/EBITDA on through-cycle assumptions, or FCF yield on a normalized cash flow.
5. What Is This Business Worth?
The right lens for SanDisk is EV/normalized-cycle earnings, not P/E on peak or trough. The value of a NAND maker is determined by three things — bit volume growth times normalized ASP-cost spread, durability of the datacenter/NBM mix shift, and the optionality embedded in the Kioxia JV. Each can be assessed; none can be priced precisely.
A useful sanity check. Q4 FY2026 guidance midpoint is $8B revenue and $31.50 non-GAAP diluted EPS. Annualizing the Q4 quarterly run-rate gives ~$32B revenue and ~$126 EPS on roughly 157M diluted shares. At $1,980, that is an annualized forward P/E of ~16x. The catch: this is annualized peak-quarter earnings. The single most important valuation exercise is asking what normalized earnings power looks like — and that depends entirely on whether the bull case (datacenter mix structurally higher; NBM contracts make pricing more like a contracted utility) holds, or whether margins mean-revert to the 25–40% gross / 5–15% operating range that has characterized SanDisk historically.
A reasonable framework: bull case — through-cycle gross margin holds at 45–55% (datacenter premium + NBM stickiness) and op margin holds at 20–30%, supporting normalized EPS of $40–60. At a 15x normal-cycle P/E that implies $600–900/share. Bear case — margins mean-revert to historical norms (gross margin 20–25%, op margin 5–10%) and normalized EPS settles at $10–20. At a 12x trough-cycle P/E that implies $120–240/share. The current $1,980 price requires the bull case AND a multiple premium.
6. What I'd Tell a Young Analyst
Track three things every quarter, not ten. (1) Datacenter revenue dollars and growth rate. (2) Number of NBM contracts signed and contract-liabilities balance on the balance sheet — that is the visibility ledger. (3) Flash Ventures underutilization charges in the COGS reconciliation — the first cycle-turn signal you will get, and it shows up in COGS before it shows up in revenue.
Do not extrapolate Q3 FY26. A 78% gross margin in a commodity-NAND business is not a steady state. It is the upper extreme of a cycle that has historically taken gross margin from 33% to 7% in twelve months. The bull case is that the new floor is higher; nobody knows the new floor.
The Kioxia question is the single most important corporate-action question. Both companies are public, JV terms run through 2029/2034, and a combination would create the #2 NAND maker globally. Track Japanese press, Kioxia conference-call language, and any disclosure about JV agreement amendments.
Read the contract liabilities line, not just the revenue line. Each new NBM agreement should show up as deferred revenue. $323M current + $188M non-current at April 2026 is a start; if that line keeps building, customer prepayments are real and the "more durable earnings power" language has substance. If it stalls while revenue keeps rising, the contracts are looser than management implies.
The market may be misjudging the symmetry, not the direction. Consensus seems to be underwriting "AI demand → high margins forever" with a peak multiple. The right question is not whether margins are high now (they are) — it is how steep the down-leg looks when supply discipline eventually breaks. SanDisk's 2023 trough produced a −33% operating margin; the next trough will be tested against the new $13B revenue base and a market cap that priced in permanence.