Business
Know the Business — Sandisk Corp (SNDK)
Sandisk is a NAND-flash pure-play that spun out of Western Digital on February 21, 2025 and, by April 2026, was producing $4.1B of quarterly operating profit on $5.95B of revenue. The stock has compounded from $36 at the separation to roughly $1,980 sixteen months later — about 55x. The question this tab answers: what part of that economic engine is structural, what part is the apex of the most violent cycle in semiconductors, and how should the business actually be underwritten?
Verdict in one line. Sandisk is a commodity-NAND fab-light JV partner whose unit economics are not durably 78% gross margin businesses — but whose multi-year New Business Model (NBM) prepayment contracts with hyperscalers could, for the first time in NAND's history, hold a structural floor under bit pricing. The right way to value SNDK is as a contracted-share-of-an-AI-memory-oligopoly plus a free option on the historical cycle re-asserting itself. Cycle-adjusted earnings power is materially below the trailing run-rate. Underwrite the contracted floor; do not extrapolate the peak.
1. Where the money actually comes from
The economic engine is simple: bits shipped × ASP per gigabyte, minus a per-bit production cost that is shared 50/50 with a Japanese competitor. Everything else — brand, channel, BiCS roadmap, NBM contracts, treasury policy — either pushes ASP up, pushes cost down, or smooths the cycle. Nothing in Sandisk's economics escapes this identity.
Q3 FY26 Revenue ($M)
Q3 FY26 Gross Margin
Q3 FY26 Operating Income ($M)
Q3 FY26 Operating Cash Flow ($M)
Q4 FY26 Revenue Guide ($M, mid)
Q4 FY26 GM Guide (mid)
Cash on hand, Apr 2026 ($M)
NBM agreements signed
Two of those metrics anchor the entire investment debate. 78.4% gross margin has never been printed by any NAND pure-play before — Sandisk's own through-cycle median is closer to 25%, and its FY2023 trough was 7%. $0 of long-term debt is also new: Sandisk fully redeemed the $1.9B of separation-related notes during Q3 FY26 and now runs on operating cash plus $3.7B of cash on hand. Both are products of the AI super-cycle, neither has historical precedent at this magnitude, and both are the levers the market is paying for.
Three of those five levers — bits, ASP, cost — are largely set by the industry. The two Sandisk genuinely controls are mix and contract structure. That is the strategic battleground. Everything else is the cycle.
2. The cycle, the spin, and the chart that explains the stock
Sandisk's quarterly P&L went from a $1.9B operating loss (Q3 FY25 — when the goodwill impairment hit) to a $4.1B operating profit (Q3 FY26) in four quarters. The company also separated from Western Digital in the middle of that swing, on February 21, 2025.
Three reads:
- The trough (3Q23–1Q24) ran gross margin negative — Sandisk sold NAND below cash cost for three consecutive quarters. This is the historical cycle low and it happened only two years before the current peak.
- The first AI lift (3Q24–2Q25) took GM to roughly 36% — a normal good-cycle level. The Q3 FY25 spike in operating loss is the goodwill impairment, not operating performance.
- The second leg (1Q26 onward) is the regime change. Revenue went from $1.9B to $5.95B in three quarters and gross margin tripled. This is not a normal up-cycle — it is a price scarcity event compounded by mix shift to Datacenter, where ASPs per gigabyte are materially higher than retail SD cards or USB drives.
A trough-to-peak swing of this magnitude has never been observed in modern NAND. Micron's record swing — FY2023 trough to FY2025 recovery — moved operating margin from negative 37% to positive 26%, roughly 63 percentage points over two fiscal years. Sandisk's quarterly operating margin moved roughly 110 points in four quarters. The market is right that something structurally different is occurring — but it is critical to separate the price-scarcity event (cyclical) from the contract-restructuring event (structural).
3. The end-market mix transformation
Sandisk reports revenue in three buckets that map cleanly to how the business will be underwritten: Datacenter (hyperscaler enterprise SSDs sold under NBM contracts), Edge (OEM client SSDs, embedded mobile, gaming, automotive — the legacy growth bucket), and Consumer (retail memory cards, USB drives, retail SSDs — the legacy cash-cow bucket).
Datacenter is up 645% year-on-year (Q3 FY26 vs Q3 FY25) and accounts for the largest absolute dollar growth, but it is still only 25% of revenue. Edge — predominantly mobile/PC/embedded OEMs — is roughly 62% and Consumer is 14%. The contracted multi-year tail Sandisk is building lives in Datacenter; the legacy SanDisk brand business that has been around since the 1990s lives in Edge and Consumer.
The Datacenter bucket is what the market is paying for and is where the NBM contracts are signed. Five contracts are signed as of the Q3 FY26 call. The chairman's framing on the call is the relevant tell: "customers prefer supply over price." That is the verbatim shift from spot pricing to take-or-pay-style economics.
The Edge bucket is more interesting than it first looks. It tripled year-on-year despite being a "legacy" business. The reason is that mobile and embedded OEMs are increasingly being asked by their own customers to lock supply for AI-on-device features, automotive infotainment, and gaming SSDs — and they pass that NAND scarcity premium through. Edge is the cycle's amplifier. It will fall first when supply normalizes.
The Consumer bucket is the brand asset. SanDisk-branded retail products were the foundation of the company before the 2016 Western Digital acquisition and remain the most recognizable NAND brand in retail. It is also the lowest-margin, most discretionary bucket — the first to roll over in any consumer recession.
4. The Kioxia JV is the most important structural fact
Substantially all of Sandisk's wafers come from Flash Ventures, a joint venture with Japan's Kioxia operating seven fabs in Japan (an eighth ramping in CY2025). Sandisk holds 49.9% / Kioxia 50%, and each side takes half the output at cost-plus-small-markup. Each side also has to fund 49.9–50% of the capex.
Three implications:
1. Sandisk's reported balance sheet understates its real capacity asset base. Most fab PP&E sits on Kioxia's books. "Fixed asset turnover" — a ratio that screens at 10.4x for Sandisk and roughly 0.9x for Micron — is not a real productivity comparison, just an artifact of where the assets live. Investors using Sandisk's reported ROIC or ROA without adjusting for the JV are double-counting capital efficiency.
2. Capacity is co-controlled. Sandisk cannot unilaterally expand bits — both parties must agree and fund. This is a feature during a downturn (no one over-builds) and a constraint during an upturn (Sandisk cannot grab share faster than Kioxia will fund). The fact that Sandisk has supply commitments to sign NBM contracts means Kioxia is funding alongside.
3. The 2029 JV expiry is the single largest structural risk over a five-year horizon. Flash Partners and Flash Alliance expire December 31, 2029 unless extended. They have been extended before. But the terms of any renewal will be set after Kioxia's October 2024 Tokyo IPO, which gave Kioxia its own independent capital structure for the first time. The next renewal is no longer a back-room deal between corporate parents — it is a negotiation between two listed companies.
Bear translation. Strip away the JV and Sandisk is a fabless NAND designer with a brand — closer to AMD's relationship with TSMC than to Micron's relationship with itself. If the 2029 renewal terms shift even modestly in Kioxia's favor — Kioxia takes 55% of the next-node output, or shifts to a higher markup — Sandisk's structural unit economics get noticeably worse without anything visible going wrong operationally.
5. The annual P&L — and the carve-out scarring that hides it
Sandisk's reported annual financials run only as a carve-out from Western Digital before the February 2025 spin. The full year of FY25 still includes goodwill impairment, separation costs, and tax-allocation noise. The chart below shows the GAAP record and then the underlying gross profit, which is the cleanest read on the actual unit economics across the cycle.
Two observations. First, gross profit doubled from FY24 to FY25 ($1.07B to $2.21B) yet operating income got worse (loss of $468M to loss of $1.38B). That is the goodwill impairment ($1.83B) and standalone-company costs hitting all at once. On a clean operating basis, FY25 was genuinely better than FY24. Second, nine months of FY26 already produced $5.35B of operating income — more than the prior four years combined at the gross-profit line. The Q4 FY26 guide ($7.75–$8.25B revenue, ~80% gross margin) implies a full FY26 operating profit of roughly $12B on $19B+ of revenue. Whether that rate can sit through FY27 is the entire equity case.
6. Cash, capital structure, and capital allocation
Three numbers carry the whole capital-structure story:
- Long-term debt: $1.83B → $0. Sandisk fully redeemed its separation-related notes during Q3 FY26 with a $46M extinguishment loss. The company now runs with net cash of roughly $3.7B. Interest expense, which was $71M for nine-month FY26, will be near-zero in FY27.
- Contract liabilities: $25M → $511M. This is the NBM prepayments showing up on the balance sheet — a tangible measure of the contract book.
- Receivables: $1.07B → $2.73B. Revenue grew faster than the sales cycle could collect. DSO has improved from roughly 54 days to 41 days at the new revenue scale, but the absolute exposure to any single hyperscaler customer not paying is now meaningful.
Capital allocation policy is just being formed. Q3 FY26 announced a newly authorized share repurchase program and the CEO referenced the zero-debt balance sheet and strong cash generation. No dividend has been declared. With a market cap near $290B and $3.7B of net cash, buybacks at current prices would retire roughly 1.3% of shares per $3.7B deployed — not a needle-mover at this valuation. The more interesting use of cash will be opportunistic re-investment in Flash Ventures capex if Kioxia signals a Phase 2 of NBM-backed expansion, or M&A in adjacent storage layers (controller IP, HBF assembly).
7. The moat question — honestly
Is this a moat business?
Honest verdict on moat: Sandisk has a narrow-to-no moat at the unit-economic level. It is a commodity producer in a five-supplier oligopoly. The thing that could become a moat over the next three to five years is the NBM contract book — long-duration, prepaid, take-or-pay supply contracts that effectively convert NAND from a spot commodity to a contracted utility. This is not yet proven and it is the single most important thing to watch. If NBM holds through the next downturn, the through-cycle ROIC math changes permanently. If it breaks at first contact with surplus supply, Sandisk goes back to being a normal NAND maker.
8. Peer scoreboard — who Sandisk is actually compared against
Two reads. Uncomfortable: Sandisk's current-quarter gross margin (79%) is the highest in the peer set — higher than Pure Storage, which is a software-wrapped subscription business. That is a structurally improbable equilibrium; either Pure Storage will re-rate up or Sandisk will re-rate down, but they will not co-exist at this gap. Sandisk's EV/run-rate revenue (11.6x) is below Micron, Western Digital, and Seagate — the market already partially discounts the spike. Sandisk does not trade as if 79% gross margin is durable.
Informative: Compare Sandisk to Micron — DRAM and NAND, owns its fabs, twice the revenue at run-rate, trades at 24x EV/run-rate revenue versus Sandisk's 11.6x. The market is paying for Micron's HBM exposure and its capital intensity moat. Sandisk has neither — and that price differential is the verdict on JV-light economics. Pure Storage's 70% gross margin holds because it sells subscription software wrapped around someone else's NAND; PSTG's EV is just $24B. That is the cap on what "high-margin NAND" can be worth without scale; Sandisk is being valued primarily on scale plus the NBM thesis.
9. Cash generation versus reported profit — the honest free cash flow path
Through-cycle, the gap between reported net income and actual cash to shareholders matters most for a JV partner. Sandisk reports both Free Cash Flow (OCF − capex) and Adjusted Free Cash Flow (FCF less Flash Ventures activity, net). The chart below shows both for the historical periods plus the FY26 run-rate.
Sandisk produced negative free cash flow in three of the last four fiscal years. Only the AI-quarter inflection (Q1–Q3 FY26) produced $4.4B of FCF, more than the prior decade's cumulative cash flow under Western Digital ownership. This is the central reason not to extrapolate. Reported capex (~$200M/year) is misleadingly small because most fab capex is funded through Flash Ventures notes receivable, which sit in investing cash flow, not capex. When the JV's next capacity expansion is approved (likely in tandem with NBM commitments), the cash demand on Sandisk's balance sheet will grow proportionately.
The Q3 FY26 cash flow statement itself: operating cash flow of $3.04B was 84% of net income — a clean conversion. But $486M of that came from the increase in contract liabilities (NBM prepayments), $631M from income taxes payable, and $1.49B of working capital from receivables ate cash. The right run-rate normalized operating cash flow at this revenue scale is probably $1.5–$2B per quarter, not the headline $3B.
10. Customers, geography, and concentration
Sandisk has not disclosed customer-level concentration in the press release. The qualitative read from the 10-K and transcripts: hyperscaler concentration in Datacenter is high and increasing — the five NBM contracts almost certainly cover the top two-to-three hyperscalers each on multi-year commitments. Edge is more diversified across mobile/PC/embedded OEMs; Consumer is broadly distributed through retail channels.
Geographically, roughly 80% of revenue ships outside the U.S. This reflects (1) Asia-anchored manufacturing (Japan fabs + Malaysia assembly), (2) Asia-anchored customer geography (Chinese mobile OEMs, Korean/Taiwanese OEM assemblers), and (3) hyperscaler regional spend allocation. The U.S. is a smaller direct revenue book than the headline U.S. listing suggests.
This matters for valuation because a U.S.-listed pure-play with 61% Asia revenue exposure carries FX, tariff, and export-control risk that the U.S. peer comp set (Micron, WDC, STX) carries less of. The risk is bilateral: a U.S.-China deterioration would hit Sandisk's Chinese OEM revenue first; a Taiwan-strait disruption would hit Sandisk's Taiwanese assembler customers. None of this is immediate, but all of it is structural.
11. How to value this business — the lens that fits
Sandisk does not fit a single valuation lens cleanly. The right approach blends three.
The lens that actually works is Contracted-revenue NPV plus a cycle option. Treat the NBM-contracted Datacenter book as quasi-utility cash flow with a known floor margin; treat Edge and Consumer as a normal NAND-cycle business that will deliver something like a 20-25% through-cycle gross margin; and price both separately.
Worked example. If NBM contracts cover 25% of Sandisk's bits at a 50% locked gross margin for five years, that locks roughly $2.5B/year of gross profit (≈$1.8B operating profit) on the contracted layer. Apply a 15x multiple — appropriate for a contracted memory business with a 5-year tail — and the contracted layer is worth roughly $27B. The remaining 75% of the business is cyclical NAND; apply Micron's blended 4x mid-cycle EV/EBITDA on roughly $4B of through-cycle EBITDA = $16B. Sum of parts: ~$43B. That is meaningfully below the $260B current market cap. The implied bull case has NBM covering closer to 60-70% of bits at progressively higher locked margins, plus a cycle that does not regress. Decide whether that is a base case or an aggressive scenario.
The math above is a teaching example, not a price target. A NAND business that you cannot defend on cycle-neutral economics is a momentum trade, not an investment.
12. What to watch, in order of weight
13. The one-page mental model
Think of Sandisk as three businesses stapled together by a common bit-stream.
That table is the entire investment thesis in one frame. The bull case is that bucket #1 grows from 25% to 50%+ of revenue at a locked margin while the other two buckets ride the AI cycle. The bear case is that bucket #1 stalls at the current contracts, bucket #2 reverts to mean as supply re-enters, and the consolidated business proves to be a normal NAND cycle dressed up briefly by a scarcity event.
The discipline a serious investor needs to hold: separate what's contracted from what's cyclical, every quarter, on every line.