Sandisk (SNDK) Q4 FY2026: Cost of Revenue Fell While Revenue Rose 372%
Sandisk closed Wednesday at $1,350.50, down 5.4% on the session, then traded down toward $1,257 after the fiscal fourth quarter release. The stock is up roughly 490% year to date and has been the best performer in the S&P 500 in 2026. Its fifty-two week range runs from $40.53 to $2,354.39, the high set on June 22. Market capitalization sits near $200 billion on 148.1 million shares outstanding, with a beta of 4.13. That context matters, because the quarter was not the problem.
Revenue of $8.965 billion beat consensus. Non-GAAP EPS of $39.25 beat by roughly $5. The number that matters more than either is cost of revenue: $1.383 billion, against $1.403 billion in the year-ago quarter. Revenue rose 372% and the cost of producing it fell 1.4%. Nearly the entire incremental $7.06 billion of revenue landed in gross profit, which is how gross margin traveled from 26.2% to 84.6% in four quarters. That is not operating leverage, which implies fixed costs spread across growing volume. It is a price transfer, executed roughly two-thirds through pricing and one-third through volume on the company’s own attribution, and the structure that permits it sits a few lines down the balance sheet.
Net property, plant and equipment: $674 million. That is the entire owned fixed asset base of a semiconductor company that just booked $20.2 billion of annual revenue and $14.5 billion of annual gross profit. Fourth quarter capital expenditure was $43 million, or 0.48% of revenue. For the full year it was $177 million, 0.9% of revenue. Micron carries something on the order of sixty times the fixed asset base to do a comparable job. The fabs are not Sandisk’s. They belong to Flash Ventures, the three-entity joint venture with Kioxia running eight facilities in Japan, and Sandisk buys wafers out of it. This is why the incremental margin looks the way it does, and it is also the ceiling. Sandisk cannot add supply unilaterally, cannot capture the wafer manufacturing margin, and owes Kioxia roughly $1.2 billion in manufacturing service fees across 2026 through 2029 under an arrangement that runs to the end of 2034. The equity method line completes the picture: Flash Ventures contributed $160 million of losses to Sandisk’s fiscal year while Sandisk booked $14.5 billion of gross profit against it. The asymmetry is contractual rather than structural, and Kioxia is now a listed company with shareholders of its own who can read the same disclosure.
The guide is where the market drew its conclusion. First quarter fiscal 2027 revenue of $10.3 to $10.8 billion came in below a consensus near $11.16 billion, and non-GAAP EPS of $44 to $46 straddled consensus rather than clearing it. More telling than the revenue shortfall is the gross margin range of 83.0% to 85.0%. The midpoint of 84.0% is sixty basis points below the 84.6% just delivered, while revenue at the midpoint climbs 17.7% sequentially. Management is guiding the first sequential margin decline of the cycle into the largest revenue quarter in the company’s history. The peak margin quarter and the peak revenue quarter are not going to be the same quarter, and the second derivative is what memory equities trade on.
The reason for that compression is the same fact the bulls are underwriting. Sandisk has now signed ten New Business Model agreements across eight customers, five announced in April and five since, covering something close to $93.9 billion of minimum revenue at floor pricing with a weighted average term above four years. That is 4.6 times fiscal 2026 revenue contracted at a floor. Bernstein’s argument that these agreements mute earnings downside even under a price collapse worse than 2010 is correct. It is also the reason first quarter revenue guidance disappointed. Bits committed at contracted floors do not reprice with spot. The protection Bernstein is pricing for 2029 is the cap Goldman’s model is running into for 2027. Both statements describe one contract book, and investors who want the downside insurance are buying the upside limitation with it. The cash evidence that customers believe in scarcity is stronger than the pricing evidence: $2.476 billion of NBM prepayments and deposits arrived during the fiscal year, $1.938 billion of it in the fourth quarter alone, alongside $1.242 billion of contract liabilities against $25 million a year ago. Customers are wiring money in advance to hold a place in line.
Two other lines deserve more attention than they will get. Consumer revenue fell 32% sequentially and 5% year over year in a quarter when NAND pricing roughly doubled, which means consumer bit volume was gutted. The segment is now 6.2% of quarterly revenue against 31% of fiscal 2025 revenue. Sandisk is starving the retail brand to feed the datacenter, which is the correct allocation decision on this quarter’s arithmetic and a slower decision to reverse than the spreadsheet implies. Shelf space, channel relationships and category presence do not sit idle waiting. Separately, refund liabilities climbed to $1.5 billion from $126 million, an accrual worth 17% of a quarter’s revenue set aside against future channel price protection and rebates. The company is reserving against the exact scenario the multiple says will not happen.
Capital allocation is doing something more ambiguous than the headline suggests. Sandisk repurchased $4.524 billion of stock in the fourth quarter, roughly 89% of adjusted free cash flow, and the board added $14 billion to bring remaining authorization to $15.5 billion. Shares issued and outstanding still rose over the year, from 146 million to 149 million, because equity compensation vesting outran the buyback. Taxes paid on vested awards ran $630 million for the year. Diluted share count was 157 million in the fourth quarter and is guided to about 155 million for the first quarter, so $4.5 billion of repurchase bought roughly a 1.3% reduction in the guided diluted count. The remaining $15.5 billion is 7.7% of the market capitalization at Wednesday’s close and would be genuinely accretive at these prices. It was deployed during the highest-priced quarter in the company’s history. Note also that GAAP EPS of $43.97 exceeded non-GAAP EPS of $39.25, an inversion driven by an $804 million mark-to-market gain on $970 million of marketable equity securities purchased during the quarter and now carried at $1.777 billion. That position is unidentified in the release and is worth a question on August 13.
On the moat, Morningstar’s no-moat position is right about the chip and wrong about the position. The wafer is a commodity and Sandisk does not make it. What Sandisk holds that a NAND wafer does not is the contract book, the Stargate QLC platform now shipping for revenue, and the High Bandwidth Flash specification released through the Open Compute Project with SK hynix. HBF is the only serious attempt to move NAND from a storage tier into the inference memory hierarchy. If it lands with a named hyperscaler, the durable advantage is real and the terminal multiple is not a memory multiple. If it does not, the advantage is a wafer supply agreement that expires in 2034 and a brand being deliberately withdrawn from the market where it was built.
Base case assumes undersupply persists through fiscal 2027 with contracted mix limiting realized ASP upside. Revenue of $44 to $48 billion and non-GAAP EPS of $175 to $195 at eight to ten times puts the stock at $1,400 to $1,950. Catalyst is the August 13 Investor Day. Bull case requires tightness through fiscal 2028, NBM count past ten with the contracted floor stepping higher on renewal, and an HBF design win at a named customer. Twelve times $200 gets to $2,400, which is where the average street target sits and just above the June high. Goldman’s James Schneider carries $2,200 after doubling his target on July 5, the high estimate on the tape is $3,169, the low is $1,000, and the ratings distribution is 25 buys, five holds and no sells, which is itself a risk. Bear case is not a demand collapse. It is a cohort derating. At roughly 7.5 times annualized guided earnings and about 4.6 times annualized revenue on an enterprise value near $193.5 billion net of $6.5 billion in cash and securities against zero debt, the stock is already priced as a cyclical at peak. Multiple compression from 7.5 times to four or five times on entirely flat earnings delivers $700 to $900 without a single downward revision. That is how memory equities have always broken. The 47% drawdown Sandisk took during July, on no fundamental news, at a beta above four, was the rehearsal. Micron, SK hynix, Samsung, Kioxia and Western Digital derate together, and Sandisk derates hardest, because it has the most gross margin to give back and the least owned capacity to defend it with.
The number to take into August 13 is not revenue, margin or the buyback. It is the share of fiscal 2027 bit supply already committed at NBM floor pricing. Everything above 60% makes this a contracted cash flow business with a memory chart attached. Everything below 40% means the first quarter guide was conservatism rather than structure, and the miss was noise.