Sandisk SNDK

Current Stats

Post-Spinoff Performance

Computed from split-adjusted closing prices. Returns are total return — (end price + dividends received − start price) ÷ start price — with no reinvestment assumed.

Post-Spin Analysis — August 25, 2026

Written against the August 24 close of $1,493.12. Primary sources: the FY2026 Form 10-K (filed 2026-08-17, for the year ended July 3, 2026) and the FY2025 Form 10-K. Two full annual reports covering four fiscal years — the deepest record of any name we cover.

The one-line version

Sandisk has returned +2,972% since its February 2025 spin. It is the single best outcome in the 559-name universe, and it is the hardest thing we have had to value.

At $1,493.12 it trades at 20.2x trailing earnings — which sounds unremarkable until you see what produced those earnings. Revenue rose 175% in FY2026 while the company shipped only mid-teens percent more bits. Roughly 91% of the growth was price. Revenue per gigabyte rose ~150% in Datacenter and ~180% in Edge.

The company lost money in three of the four years for which records exist. Cumulative FY2023–25 operating losses were $6.4 billion. FY2026 operating income was $12.4 billion.

A 35% reversion in NAND pricing takes the P/E from 20x to 47x. A 50% reversion takes it to 110x. Against a three-year average operating income the shares trade at roughly 74x. The multiple looks lowest exactly at the top — which is the oldest trap in commodity investing.

Rating HOLD. No price target — see below. Grade B (2.93): a genuinely improved business riding a pricing spike nobody can size.

What the business actually is

The NAND flash memory business separated from Western Digital on February 24, 2025 — Western Digital retained the hard-disk-drive business and Sandisk took flash. It sells into three end markets, and the mix has changed beyond recognition in eighteen months:

Datacenter was 5% of revenue two years ago and is 25% now. That is the AI infrastructure build arriving in a memory company’s accounts.

Manufacturing sits in a joint venture. Sandisk and Kioxia operate Flash Ventures across eight fabs in Japan — six at Yokkaichi, two at Kitakami. The structure matters enormously for the economics: FY2026 capital expenditure was $177 million, or 0.9% of revenue. A semiconductor business that does not carry its own fab capex converts profit to cash at a rate almost nothing else in the industry can match.

That figure understates the real commitment, and Capital allocation below sets out by how much — Sandisk funds fab capacity through loans to the joint venture rather than through its own capital budget, which puts the true FY2026 total nearer 2.2% of revenue.

Management Team

Board members with directly relevant operating histories:

Also on the board: Alexander R. Bradley, Richard B. Cassidy II, Ellyn J. Shook, Miyuki Suzuki. Caulfield and Suzuki joined Western Digital’s board in 2021 and carried across at the separation.

Three observations.

The parent’s chief executive chose this half. Goeckeler ran Western Digital, decided to divide it, and took flash rather than hard drives. That is a stronger signal than any strategic rationale in a Form 10 — the person with the most information about both businesses picked one. Eighteen months later the flash half is worth several times the parent.

The board is unusually well-matched to the problem. A former foundry CEO, a former semiconductor CFO of a decade’s standing, and a two-time semiconductor CEO. For a company whose central question is where it sits in a capacity cycle, that is the right expertise, and it is rarer on spinoff boards than it should be.

Chairman and CEO are combined. Goeckeler holds both roles. That is a governance structure some investors discount, and it is worth stating plainly rather than leaving to a proxy statement.

The insider selling plans, disclosed in the 10-K

Both senior executives adopted Rule 10b5-1 selling plans in May 2026 — weeks before the stock peaked at $2,335 on June 25:

	Plan adopted	Maximum shares	Approximate value at $1,493
David Goeckeler, CEO	May 29, 2026	up to 124,192	~$185M
Luis Visoso, CFO	May 22, 2026	up to 11,000	~$16M

A 10b5-1 plan is the correct and compliant way for an insider to sell, and adopting one is not a signal about the business — the schedule is fixed in advance precisely so it cannot be. Goeckeler’s plan runs to September 2027 and a portion of his shares is withheld for tax at roughly 45%.

It is disclosed in the annual report and an investor should know it exists. After a 2,972% run, the people with the best view of the cycle have arranged to sell into it on a schedule.

The financial record — four years, and it is brutal

Read the gross margin column. 7.1% → 16.1% → 30.1% → 71.5%. That is not a company improving. That is a commodity price.

Sandisk lost money in three of the four years on record, with cumulative operating losses ex-impairment of $6.4 billion across FY2023–25 — including a $1.83 billion goodwill impairment in FY2025 and $671 million in FY2023. Then it earned $12.4 billion in one year.

The decomposition that decides the investment

The 10-K is unusually clear about what happened, and the numbers are the crux:

Total bits shipped rose mid-teens percent. Revenue rose 175%. Roughly 91% of the increase was price, not volume.

The exception is Datacenter, and it is the entire bull case. Volume there rose ~120% — that is real physical demand from AI infrastructure, not a price effect. Datacenter is now $5.2 billion of revenue and growing on units.

Edge is the exposure. At $12.2 billion it is 60% of the company, and it grew on +180% pricing against high-single-digit volume. That is the part that reverts.

Where Sandisk sits in the value chain

Sandisk has almost no durable pricing power and, at this moment, more pricing power than any other business we look at. Those are not contradictory. They are the two halves of what a commodity memory business is, and holding both at once is the only honest way to read the FY2026 accounts.

The price is the industry’s, not the company’s

91% of FY2026 revenue growth was price. Revenue rose 175% on mid-teens bit growth; revenue per gigabyte rose roughly 150% in Datacenter and 180% in Edge. No product decision, customer relationship or technology advantage produced that. The NAND supply-demand balance did.

The structural point, and it is unlike anything else on this site. For most businesses, margin is a function of what the company has built — a data set, an installed base, a patent, a cost position. Sandisk’s margin is a function of what its competitors decide to build. It can execute perfectly and still earn nothing, and it can execute unremarkably and earn $12.4 billion, which is roughly what happened in consecutive years.

That is why this page carries no price target. A target requires a view on mid-cycle earnings; mid-cycle earnings here require a view on four other companies’ capital budgets. We do not have one, and neither does anyone quoting a number.

The joint venture cuts both ways, and the second edge is under-weighted

The favorable half is well understood. Flash Ventures means Sandisk does not carry solo fab capital expenditure, which is why headline capex was $177M on $20.2B of revenue and why free cash flow converted at 57% of sales in FY2026. Almost nothing in semiconductors does that.

The unfavorable half is that it also removes control. Capacity decisions are joint. When the cycle turns — and the entire question on this page is when — a solo operator can idle its own fabs. A joint venture negotiates.

The one development that could change the structural answer

Datacenter went from 5% of revenue two years ago to 25% today. That matters beyond the growth rate, because enterprise SSD is a more qualified product than consumer flash. Hyperscalers specify endurance, latency consistency and failure characteristics, and they validate a supplier’s parts into a platform.

If a meaningful share of Sandisk’s output becomes qualification-gated rather than spot-priced, the business becomes structurally different — closer to a components supplier with design-in protection than to a commodity producer. It is far too early to claim that has happened. But it is the only mechanism by which this company escapes being a price-taker, and it is worth watching in the mix disclosure rather than in the margin.

Why the stock moved the way it did

The forced-selling flush was severe and it worked. Day-1 volume ran 15.4x baseline, the stock fell 6.5% on its first session, and it kept falling for 57 sessions to $29.62 — nearly 40% below the day-1 close. Anyone who bought the flush low has made roughly 50 times their money.

Then nothing happened for four months, and the stock traded between $32 and $52 through August 2025. The inflection was September 2025, when it went from $52.47 to $112.20 in a single month as NAND pricing turned.

The largest single sessions are all pricing and earnings events: +27.6% on 2026-01-06, +22.0% on 2026-06-25 (the peak), −20.3% on 2025-11-20, and +26.0% on 2026-07-30 — the day after the post-peak trough, on the FY2026 results.

Realised volatility is 105.5% annualised, and volume still runs 7.4x the post-spin baseline. This is not a stock that has settled.

Capital allocation

The balance sheet was transformed in twelve months, and the buyback is now the central capital decision in front of this board.

Read the last two rows together. Sandisk entered FY2026 with $1.9B of debt and $1.5B of cash. It ended with no debt at all and $4.76B of cash, having returned $4.5B to shareholders in between. That is a complete balance-sheet reconstruction inside one fiscal year, and it was funded entirely by the pricing cycle.

The capital-expenditure figure understates the real commitment

This page has said capex was $177M, or 0.9% of revenue. That is the headline number and it is correct as far as it goes — but it is not the whole capital commitment, because Sandisk funds fab capacity through the joint venture rather than on its own balance sheet:

Total FY2026 capital commitment was roughly $452M — capex plus net JV funding — or 2.2% of revenue. Still extraordinarily light for a semiconductor business, but two and a half times the headline figure, and the more useful number.

And there is a signal buried in that table. In FY2024 and FY2025 the joint venture was a net repayer of cash to Sandisk. In FY2026 it became a net borrower, for the first time, by $275M.

That is what the beginning of a capacity build looks like in a cash flow statement. Capacity is the mechanism that ends every NAND pricing cycle, and it takes two to three years to arrive. The single most important variable on this page has just turned, quietly, in a line item nobody reads — and it turned in the direction that eventually ends the earnings this valuation rests on.

The $20 billion authorization

Against FY2026 free cash flow of $11.5B, $20B is under two years of cash. Against the three-year average of roughly $3.6B, it is five and a half years. Which of those is the right denominator is the entire question on this page, restated as a capital-allocation decision.

The uncomfortable arithmetic. The board authorized $6B in April and tripled the program to $20B in August — after the shares had risen roughly thirty-fold from the post-spin low. Every dollar spent at today’s price is a bet that current earnings are closer to permanent than to peak.

That is not automatically wrong. With no debt, $4.8B of cash and negligible capital needs of its own, Sandisk has few alternative uses; and a company that hoards cash through a cycle peak has its own history of destroying value. But it is the same bet the equity holder is making, made with the equity holder’s money, at the equity holder’s price — and it is being sized off a single year of results.

The tell to watch is pace, not authorization. $4.5B in a year against a $20B authorization is measured. A sharp acceleration from here would say management believes the cycle has further to run — which is information, and the only forward-looking statement about NAND pricing this management team is likely to make.

Valuation — and why there is no price target

The balance sheet is pristine. Debt went from $1.9 billion to zero and cash stands at $4.8 billion. On any measure of financial risk this company is now unassailable — which is precisely what a peak-cycle balance sheet looks like.

What price reversion does

Holding volume and cost of revenue flat and varying only realised pricing:

For scale: FY2025 revenue was $7,355M. Pricing back at FY2025 levels on FY2026 volume would put revenue near $8.4 billion — the bottom two rows of that table.

Mid-cycle earnings

This is the whole valuation problem in one table. On peak earnings the stock is unremarkable. On any normalized measure it is extraordinarily expensive. Which is right depends entirely on whether NAND pricing has structurally re-rated or is at a cyclical top — and nobody knows.

Why no price target

Every other company we cover gets one. Sandisk does not, and the reason is methodological rather than evasive.

A price target requires a defensible normalized earnings figure. For Sandisk the plausible range runs from an operating loss to $12.4 billion, and the correct answer depends on a commodity price that has moved 180% in a year and could move as far back. A target would imply a precision the data cannot support, and publishing one would be false confidence.

What can be said instead:

• At unchanged pricing, the shares are worth more than they trade for — a 20x business growing datacenter volume at triple digits is not expensive.
• At FY2025 pricing with FY2026 volume, the equity is worth a fraction of today’s price.
• The market at $1,493 is discounting roughly a 20–25% price reversion, which is neither obviously optimistic nor obviously conservative.

That is an honest description of a bet, not a valuation.

Tail scenarios

Upside tails

1. NAND pricing is structurally re-rated, not cyclically peaked. If AI infrastructure demand keeps supply tight for several years — an argument the industry is making seriously — then FY2026 is a new base rather than a peak. At unchanged pricing the shares compound from a 20x multiple with no debt. This is the bull case and it is not absurd.

2. Datacenter keeps compounding on volume. It grew ~120% in exabytes, not just price. At $5.2 billion and 25% of revenue, another two years of that mix shift would make Sandisk a fundamentally different, less cyclical business.

3. The $20 billion buyback. Sized and dated in Capital allocation above. With no debt and a capital budget near 2% of revenue, the cash is genuinely there. Retiring shares at a cyclical peak is the risk; retiring them into a reversion would be exceptional capital allocation.

4. Capital-light economics persist. The Flash Ventures structure keeps Sandisk’s own capital budget under 1% of revenue, and even including net funding to the joint venture the FY2026 total was about 2.2%. Whatever the cycle does to margins, this company keeps more of its profit than an integrated memory maker. “Permanent” would be too strong — the JV became a net borrower from Sandisk in FY2026 for the first time, which is what a capacity build looks like early.

Downside tails

1. The reversion is faster and deeper than the table above. Memory pricing does not glide down; it falls. In prior NAND downcycles contract prices halved within four quarters. The −50% row is not a tail — it is a normal memory downcycle, and it puts the shares on 110x.

2. Supply responds. Every NAND producer is seeing the same prices. Samsung, SK Hynix, Micron, Kioxia and YMTC all have capacity decisions in front of them, and the industry’s history is that they all add at once. The 10-K’s own risk factors name declining average selling prices first.

3. The buyback destroys value. $20 billion authorised at a share price up 2,972%, funded by peak-cycle cash flow. If pricing reverts, the company will have retired stock near the top with money it needed at the bottom — the mistake memory companies have made in every previous cycle.

4. Edge normalizes hard. Sixty percent of revenue grew on 180% pricing against high-single-digit volume. There is no demand story underneath that mix — it is consumer and client devices absorbing higher prices, which they do until they do not.

What the tails say about the position

The distribution here is wider than for anything else we cover, in both directions, and it is not symmetric in time. The upside requires several years of sustained tightness; the downside can arrive in two quarters.

That argues for a small position or none, and against any position sized on the 20x headline multiple.

What this most resembles

Nothing in the universe, which is itself the finding.

Sandisk’s annualised excess return is roughly sixty times the next best name in the universe. It is not an outlier within the distribution; it is outside it.

The closest structural analogue is not a spinoff at all — it is any commodity producer at a cycle peak: enormous trailing earnings, a low trailing multiple, a pristine balance sheet, and an earnings base that can halve in a year. Every element of the Sandisk setup has been seen before in memory, oil and shipping. What is unusual is that it happened to a company eighteen months out of a spinoff.

In depth: competitive dynamics

1. The industry structure

NAND flash is a five-player oligopoly with no meaningful barriers on the demand side and enormous ones on the supply side:

2. The JV is the single most important structural fact

Sandisk does not own fabs outright. Flash Ventures with Kioxia operates eight plants in Japan, and Sandisk takes its share of output. The consequences run both ways:

✅ Own capital budget of 0.9% of revenue. Sandisk captures memory-cycle upside without carrying memory-cycle capital intensity. In FY2026 it converted $12.4 billion of operating income on $177 million of capex — though net funding to the joint venture added a further $275 million, so the honest figure is 2.2% of revenue rather than 0.9%.

✅ Capacity discipline is shared. Neither partner can unilaterally flood the market, which is a genuine restraint on the industry’s worst instinct.

⚠️ No control over the technology roadmap or capacity timing. Sandisk cannot independently decide to add supply when prices are high — or refuse to when its partner wants to.

⚠️ Concentration in Japan. Eight fabs in two locations, in a seismically active country.

3. Moats and weak spots

⚠️ There is no product moat. NAND is a commodity. Bits are fungible, priced on spot and contract markets, and qualified across suppliers. The 10-K’s first listed risk is “average selling prices, volatile demand, technological change, industry consolidation, lengthy product qualifications, and supply constraints.”

✅ The brand has real value in Consumer and Edge. SanDisk-branded retail products command a premium that pure bit suppliers do not.

✅ Customer concentration is unusually low for semiconductors. The top ten customers are 44% of revenue and no single customer exceeds 10% — better diversification than most names we cover.

❓ Datacenter qualification may become a moat. Enterprise SSD qualification at hyperscalers is slow and sticky. If Sandisk’s ~120% datacenter volume growth reflects design wins rather than allocation, that is a durable position. The filings do not distinguish the two, and it is the most important undisclosed fact about this company.

In depth: valuation against peers

Memory companies are valued on where you think you are in the cycle, and the convention is deliberately inverted: a low trailing P/E at peak earnings is a warning, not an attraction; a high one at trough earnings is often the entry.

The like-for-like comparison that matters is Sandisk against itself. At the April 2025 low the company was loss-making and the market capitalization was roughly $4.3 billion. Today it earns $11.4 billion a year and is valued at $218.6 billion. The business improved enormously; the question is how much of that improvement is permanent.

We do not publish peer trading multiples we cannot compute from primary filings. Samsung, SK Hynix, Micron and Kioxia figures were not independently verified and are therefore described rather than quoted.

In depth: is Sandisk an acquisition target?

No, on three separate grounds.

Size. At a $218.6 billion enterprise value the list of possible acquirers is empty. This is now one of the largest semiconductor companies in the world by market capitalization.

§355(e). The two-year window from the Western Digital separation runs to February 24, 2027. An acquisition of 50% or more inside it would make the spin taxable to Western Digital.

Antitrust. The only industrially logical buyers are the other four NAND producers, and any combination among the top five would be blocked in every major jurisdiction.

The interesting corporate-action question is the reverse: does Sandisk buy something, or restructure Flash Ventures?

With $4.8 billion of cash, no debt and $12 billion of annual operating income at current pricing, Sandisk has more capital than at any point in its history — and a joint venture structure that leaves it without control of its own manufacturing. Consolidating Flash Ventures, or a transaction with Kioxia itself, is the transaction that would change the company most.

Memory companies making large acquisitions at cycle peaks has a poor history, and the $20 billion buyback suggests the board has chosen a different route.

What would change this view

Investment Scorecard

Scored on the five weighted dimensions set out in our methodology.

What the capital-allocation and value-chain work changed: nothing, and that is the finding

Both sections were added after this grade was set, and neither moved a score — the scorecard had already priced the commodity position, the pristine balance sheet and the awkwardness of a $20B authorization struck after a thirty-fold move.

One figure was corrected. Capital intensity is 2.2% of revenue, not 0.9%, once loans to Flash Ventures are counted alongside the capital budget. That is still remarkably light and does not change the conclusion.

And one new indicator was found, which is recorded in What would change this view rather than in a score: the joint venture became a net borrower from Sandisk in FY2026 for the first time. Capacity is what ends every NAND pricing cycle, and this is what its beginning looks like in a cash flow statement — two to three years before it reaches the price.

First grade for this company, and the lowest we have assigned

B (2.93) on a stock that has returned +2,972%. That is not a contradiction — the grade measures business quality, and NAND flash is a commodity business with no pricing power at the trough. The four-year record shows exactly that: three years of losses, one extraordinary year.

The rating is HOLD rather than SELL because the datacenter volume growth is real, the balance sheet is pristine, and a structurally tighter NAND market is a genuine possibility rather than a hope.

What would take it to A−: evidence that datacenter demand is design-win driven and durable, which would make Sandisk a specialty supplier rather than a commodity producer. What would take it to C: a pricing reversion with the buyback already spent.