Sandisk Corporation SNDK

Key data

The business

SanDisk makes NAND flash memory: the chips and drives that store data without power, from the microSD card in a phone to the enterprise SSDs racked in a data center. Western Digital cut it loose as a standalone company in February 2025. It sells into three end markets: Edge, its biggest line at $12.2B (client and mobile storage); Datacenter, the fastest grower (enterprise SSDs); and Consumer, the cards and thumb drives, $2.9B. No single customer is more than 10% of revenue. It does not own its fabs outright, it makes the wafers through Flash Ventures, a long-running joint venture with Japan's Kioxia, which is why a company this size spends almost nothing on plant, $177M last year. The moat is real but narrow: NAND is a five-player scale oligopoly where node leadership and the shared-fab structure keep it in the game. Pricing power it does not have. NAND is a commodity, and the cycle sets the price.

The numbers

Read the four quarters of fiscal 2026 in sequence and it is a commodity at the top of its cycle.

Gross margin went from 26 cents on the dollar to 85 in four quarters. Almost none of that is cost coming down, nearly all of it is the price of NAND going up, and when a commodity triples, it drops straight to the gross line. The company beat consensus all four quarters, some by more than 60%, which tells you the analysts were as surprised by the pricing as everyone else. The fourth quarter, which a few weeks ago could only be backed out of the annual total, now sits confirmed in the 10-K, and it held: about $9.0B of revenue and $6.9B of net income in a single three-month stretch.

Zoom out and the whiplash is clearer. Three straight years of losses, then a single year that earned $11.4B, more than the prior three lost combined, on revenue that nearly tripled. A 71% full-year gross margin for a memory maker is not a number that recurs; through the last down-cycle this business ran mid-teens margins and bled cash.

So what are you paying for. At $1,740 the stock is 22 times the $73.76 it earned last year, the most it has fetched in its short standalone life, and 16 times book, up from under one times a year ago. The estimates two years out imply earnings triple again, which is why the forward multiple screens at an absurd 6.6 times on eleven analysts; the market plainly does not believe that number, or the stock would sit far higher. The honest read: the market is paying 22 times a cyclical peak and betting earnings step down to something durable rather than collapse. My variant is that the datacenter line, not the overall multiple, settles it. The print that decides it is datacenter revenue holding the first quarter blended prices soften.

Datacenter went from a rounding error to $5.2B in two years; Edge tripled in one. The first is plausibly structural, the second is mostly price.

Management

The record is short, the company has only been on its own since early 2025, but this year's capital moves were decisive. Long-term debt went from $1.85B to zero and cash more than tripled to $4.8B. Management also repurchased $4.5B of stock, and here is what the numbers do not settle: at what price. The fiscal year opened with the stock near $70 and closed near $1,740, so a buyback done early was a masterstroke and one done late was burning cash at the top; the filing does not break out the average paid. Insiders were net sellers, $14.2M across eleven sales and not a single buy, plan status not disclosed in the feed. Nobody bought this run with their own money.

How it fails or surprises you

NAND prices roll over (the main way you lose). The entire 85% fourth-quarter margin is the price of flash, not SanDisk's cost. Add Samsung, SK Hynix, Micron and Kioxia capacity into any pause in AI spending and blended prices fall fast. Gross margin can travel from 71% back toward 30%, and earnings fall harder. The first sequential revenue decline is the tell.

The buyback at the peak. SanDisk spent $4.5B on its own stock in a year that stock rose more than twenty-fold. If most was bought in the back half above $1,000, management turned a cash windfall into value destruction at the classic commodity top. The 10-Q that discloses the average repurchase price proves the read right or wrong.

Enterprise storage stops being a cycle (right tail). Datacenter revenue went from $0.3B to $5.2B in two years as high-capacity SSDs begin replacing nearline hard drives in AI systems. If that displacement is structural, the datacenter line holds through the next price dip, and a stock priced at 6.6 times two-year earnings re-rates from commodity to infrastructure. Datacenter revenue holding while prices soften is the first sign.

Closing thoughts

This plays out over quarters, not one print, so what matters is which tail is fatter and whether you survive the bad one. The fatter tail is down: NAND has cycled violently for forty years, an 85% gross margin is a physical impossibility to hold, and three years of losses sit just two years back in the record. But the left tail is not fatal in the usual way, because the balance sheet is now debt-free with $4.8B of cash, so a price crash hurts the earnings and the stock without threatening the company's existence. What is genuinely at risk is the multiple and the earnings base, not solvency. If the downside linchpin breaks you can lose more than half and the business still stands.

The bet is still that the AI data center boom keeps flash scarce and expensive, so SanDisk's storage keeps selling near today's prices instead of crashing the way memory always has. What breaks it is supply catching up to AI demand, and the one pair of numbers that tells you first is blended selling price against datacenter revenue: if prices soften and datacenter falls with them, this was a cycle and you are late. Falsifiable line: if fiscal 2027 gross margin holds above 60% while competitors add capacity, the step-change was real.