Financial Gurkha | New York City | August 6, 2026

Disclaimer: This article is for educational purposes only and is not investment advice. Trading and investing are subject to volatility and market risk, including partial or entire loss of capital. Please consult your financial advisor before making investment decisions. Financial Gurkha and its writers are not liable for your losses, nor do we take credit for your gains. Figures are drawn from Sandisk Corporation's fiscal fourth quarter and full year 2026 earnings release (Exhibit 99.1, filed August 5, 2026).

Sandisk's Gross Margin Went From 26% to 85% in One Year#

The 60-Second Version#↑ Contents

Sandisk Corporation (NASDAQ: SNDK) — the company that commercialised flash memory, spent nine years inside Western Digital, and was spun back out in February 2025 — just reported one of the most extreme single-year financial reversals in semiconductor history.

The nine things that matter:

1. Fiscal Q4 gross margin hit 84.6%, up from 26.2% a year earlier — a 58.4 point expansion in four quarters. Software companies do not routinely earn 85% gross margins. A memory manufacturer just did.
2. Cost of revenue fell in absolute dollars in Q4 — $1.383 billion versus $1.403 billion a year earlier — while revenue rose 372%. That single comparison is the entire story.
3. Datacenter revenue grew 437% for the year, from $960 million to $5.153 billion. It went from a rounding error to a quarter of the business.
4. Consumer is now just 14.5% of revenue, down from 30.8% — not because it shrank (it grew 29%) but because everything else grew far faster. Q4 consumer revenue fell 32% sequentially, and prior-year data suggests that is not simply seasonal.
5. Ten "New Business Model" agreements are now signed — long-term supply deals with customer prepayments. Contract liabilities jumped from $25 million to $1.242 billion.
6. Capital expenditure was $177 million on $20.2 billion of revenue — 0.9%. Manufacturing sits inside the Flash Ventures joint venture with Kioxia, which is why free cash flow margin reached 57%.
7. Long-term debt is now zero, down from $1.829 billion. They paid it off and still bought back $4.5 billion of stock in a single quarter.
8. A $14 billion buyback expansion brings remaining authorisation to $15.5 billion — roughly equal to the company's entire shareholders' equity.
9. Q1 FY27 guidance is $10.30–10.80 billion of revenue with non-GAAP EPS of $44.00–46.00. The midpoint alone would exceed all of fiscal 2025.

The one-line thesis: Sandisk is not benefiting from the AI boom indirectly — it is a supply-constrained monopolistic-feeling seller in a market where buyers are prepaying to secure allocation. The risk is that memory is the most brutally cyclical industry in technology, and 85% gross margins have never been a stable state in its history.

What's in this report#

• Sandisk's Gross Margin Went From 26% to 85% in One Year
  • The 60-Second Version
    • What's in this report
  • Why Sandisk Matters
  • The Arc: Written Down, Then Rewritten
  • The Numbers
  • The Line That Explains Everything
  • Datacenter: From $960 Million to $5.2 Billion
  • What Actually Sits in Each Segment
    • What the company states
    • The reasonable mapping
    • The open question that actually matters
    • The brand-versus-business gap
  • Consumer: Repriced, Not Abandoned
  • The New Business Model Agreements
  • Why the Cash Flow Is So Extreme
  • The Balance Sheet Transformation
  • Guidance: The Next Quarter Is Bigger Than Last Year
  • The State of the Global NAND Industry
    • The market: $46 billion in a single quarter
    • Why prices exploded
    • China: YMTC is the story nobody is pricing
    • The capacity split: who is building and who is not
    • The technology race: layers versus density
    • Other regions
    • When does this end?
  • What Could Break This
  • Peers and the Memory Complex
  • Financial Gurkha Takeaways
  • Frequently Asked Questions
  • About This Analysis

Why Sandisk Matters#↑ Contents

Before the numbers, the context — because most coverage will treat this as a storage company beating estimates, and that undersells what Sandisk actually is.

Flash memory is the substrate the entire digital era is written on. Sandisk was founded in 1988 by Eli Harari, Sanjay Mehrotra and Jack Yuan, and commercialised the technology that became the CompactFlash card, the SD card, the USB drive, and eventually the solid-state drive. Every phone photograph, every drone flight log, every camera roll, every game console save file has passed through a lineage of chips this company helped invent. Mehrotra later left to run Micron.

That history matters right now for a specific reason. Training an AI model is a compute problem. Running one at scale is a data problem — checkpointing model weights, feeding training clusters, serving inference from vector databases, and keeping petabytes of retrieval data close enough to the GPU to be useful. All of it lands on NAND flash. The industry has spent three years talking about who makes the accelerators. Rather less attention has gone to the fact that the accelerators are useless without somewhere to put the data.

Sandisk is one of a handful of firms on earth manufacturing NAND at scale. This quarter is what happens when demand for that capacity outruns supply.

The Arc: Written Down, Then Rewritten#↑ Contents

The narrative arc here is remarkable enough to be worth stating plainly.

Sandisk was acquired by Western Digital in 2016 for roughly $19 billion. It was spun back out as an independent public company on 21 February 2025. Within months, the separation looked like a mistake: in fiscal 2025 the company recorded a $1.83 billion goodwill impairment iGoodwill impairmentGoodwill is the premium paid above the value of a company's identifiable assets in an acquisition. Accounting rules require testing it regularly, and if the business is judged to be worth less than its book value, the difference must be written off as a loss. It is a non-cash charge — no money leaves the building — but it is a formal admission that assets are worth less than previously recorded. Here the trigger was Sandisk's own falling share price., triggered — per the release — by "the trading price of the Company's common stock and resulting market capitalisation." In plain terms, the market decided the business was worth less than its book value, and the accountants were obliged to agree.

Fiscal 2025 closed with a $1.641 billion net loss and an accumulated deficit of $1.784 billion.

One year later, retained earnings stand at positive $9.649 billion. The company earned $11.4 billion. The impairment that was written off because the market doubted the business has been earned back roughly six times over in twelve months.

That is not a normal corporate turnaround. That is a commodity cycle inflecting under a company with operating leverage.

The Numbers#↑ Contents

Fiscal fourth quarter (ended 3 July 2026):

Revenue rose 51% sequentially and 372% year over year. Management attributes the sequential increase to roughly one-third higher volume and two-thirds higher pricing.

Three derived figures the release does not print:

Q4 net margin was 77.0%. On $8.965 billion of revenue, $6.903 billion reached net income. There is no mature manufacturing business that earns 77 cents of net profit on a revenue dollar in normal conditions.

Q4 operating margin was 78.5%. Operating expenses of $545 million against $8.965 billion of revenue means the entire cost base — all R&D, all sales, all administration — consumed six cents of every revenue dollar.

Sequential incremental gross margin was 97%. Gross profit grew $2.920 billion on a $3.015 billion revenue increase. Essentially every additional dollar of revenue fell straight through to gross profit, because the incremental dollars were price, not volume.

A note on comparability: fiscal 2026 ended 3 July 2026 against 27 June 2025 for fiscal 2025, making FY26 a 53-week year. The extra week modestly flatters the annual growth figures. It does not remotely account for a 175% increase.

The Line That Explains Everything#↑ Contents

If you read one line of this filing, read the cost of revenue.

Sandisk spent less to produce its product in the fourth quarter of fiscal 2026 than it did a year earlier — while selling 372% more of it.

For the full year, cost of revenue rose 12.3% ($5.143B → $5.776B) against a 175% revenue increase.

That is what a pure pricing cycle looks like in an income statement. Volume is up, but the overwhelming driver of profit is that each bit of NAND now sells for dramatically more than it did. When the same physical output sells for multiples of last year's price, gross margin does not improve gradually — it detonates. 26.2% to 84.6% in four quarters.

The corollary deserves equal weight: this mechanism runs in both directions. A cost structure that cannot fall further is a cost structure that offers no protection when prices normalise. We will return to this.

Datacenter: From $960 Million to $5.2 Billion#↑ Contents

The end-market split is where the AI thesis lives:

And the quarterly view, which is starker still:

Datacenter revenue doubled sequentially and is roughly fourteen times what it was a year ago. CEO David Goeckeler described the year as having "established datacenter as a key growth pillar" — which is corporate understatement for a business line that went from 13% of revenue to 25% while the total tripled.

What Actually Sits in Each Segment#↑ Contents

Here is something worth stating plainly, because most coverage will gloss over it: the earnings release does not define Datacenter, Edge or Consumer anywhere. It presents the three-way revenue split under the heading "End Market Summary" and moves on. There is no product mapping, no customer-type definition, and no footnote.

That is a genuine disclosure gap, and readers should treat any confident description of these buckets — including ours — with the appropriate caution. What follows separates what the company actually says from what can reasonably be inferred.

What the company states#

The only segment-relevant language in the release is in the About section, and it does two useful things. First, it names three application areas that map one-to-one onto the reporting buckets:

"…our broad and ever-expanding portfolio delivers powerful flash storage solutions for artificial intelligence workloads in datacenters, edge devices, and consumer applications."

Second, it names the customer types and the product list:

"Our technologies enable everyone from students, gamers and home offices, to the largest enterprises and public clouds to produce, analyze, and store data. Our solutions include a broad range of solid state drives, embedded products, removable cards, universal serial bus drives, and wafers and components."

So the company itself pairs "largest enterprises and public clouds" against "students, gamers and home offices" — the two ends of its customer base — with edge devices sitting between them.

The reasonable mapping#

Combining those two sentences gives the following, which we label clearly as inference, not disclosure:

The open question that actually matters#

Where do "wafers and components" sit?

Sandisk sells raw NAND wafers and components to other companies who build their own products. The release lists this as a product line but never says which end market it lands in. This matters more than it sounds, because Edge is the largest segment by a wide margin — $12.160 billion, or 60.1% of fiscal 2026 revenue, larger than Datacenter and Consumer combined.

If a meaningful share of Edge is wafer and component sales, then "Edge" is partly a wholesale channel rather than a device-storage business, and its 195% growth is partly other manufacturers bidding for raw NAND. If it is genuinely embedded and client drives, it is a device-demand story. Those are materially different businesses with different durability, and the release does not let you distinguish them.

This is the single question we would put to management. Until it is answered, the honest position is that 60% of Sandisk's revenue sits in a category the company has not defined.

The brand-versus-business gap#

There is a nice irony in the numbers. To most people on earth, Sandisk is the red-and-white memory card in a camera and the USB stick in a drawer — one of the most recognised consumer storage brands in existence.

That business is now 14.5% of revenue. The other 85.5% is sold to enterprises, cloud providers and device manufacturers who never put the Sandisk name in front of a shopper. The brand is consumer; the business is overwhelmingly business-to-business, and became far more so this year.

Consumer: Repriced, Not Abandoned#↑ Contents

We initially read the consumer line as supply being deliberately pulled from retail. The full-year figures argue for a more careful reading, and it is worth showing the work.

For the full year, consumer revenue rose 29% — from $2.268 billion to $2.935 billion. That is not a business being abandoned. It grew.

But the quarterly path is genuinely odd:

(Q1 and Q2 are not disclosed separately; $780M is the implied average from the full-year total.)

Q4 came in 30% below the average of the preceding three quarters. The obvious explanation is seasonality — Sandisk's fiscal Q4 covers roughly April to early July, which is a quiet stretch for consumer electronics retail between the holiday and back-to-school builds.

The prior year does not support that explanation. In fiscal 2025, Q4 consumer revenue was $585 million against a Q1–Q3 average of $561 million — fiscal Q4 was slightly above average, not below it. One year of comparison is thin evidence, but it is the evidence available, and it points away from pure seasonality.

The more defensible reading — and the one the pricing data supports — is that consumer is being repriced rather than prioritised. Consider: NAND pricing rose enough during fiscal 2026 to take company gross margin from 30.1% to 71.5%. Against that backdrop, consumer revenue grew only 29% while datacenter grew 437%. If prices per bit rose substantially and consumer revenue rose modestly, then consumer unit volume very likely fell. The release does not disclose units, so this is an inference — but it is a well-grounded one.

That framing is more useful than "Sandisk is abandoning retail," and it explains something readers will recognise from their own shopping: memory cards and USB drives have been getting more expensive. Sandisk is not walking away from the consumer market. It is selling less of it, at much higher prices, while the incremental bit goes to whoever pays most — and right now that is a hyperscaler.

The honest caveat: with no unit disclosure and no segment definitions, allocation and demand cannot be fully separated from the outside. What can be said with confidence is that consumer went from 30.8% of revenue in fiscal 2025 to 14.5% in fiscal 2026 — not because it shrank, but because everything else grew far faster.

The New Business Model Agreements#↑ Contents

Sandisk disclosed that since announcing five "New Business Model" (NBM) agreements in April, it has signed five more — three with new customers, two expanding existing deals. Ten in total.

The release does not fully define NBM terms, but the balance sheet does the explaining:

Contract liabilities are cash customers have paid for product not yet delivered. They went from $25 million to $1.242 billion combined. The cash flow statement confirms the mechanism: management strips out $2.476 billion of "NBM prepayments and deposits" to arrive at adjusted free cash flow.

Customers are wiring Sandisk billions of dollars in advance to lock in future NAND supply.

That is the same signal we flagged in Caterpillar's record backlog — buyers reserving capacity years ahead — expressed through a different balance-sheet line. When customers prepay, they are not forecasting demand. They are defending against not getting any.

Note also that management excludes these prepayments from adjusted free cash flow, describing them as "not indicative of the core underlying cash flows." That is conservative and correct: prepayments are borrowed-forward revenue, not earned cash. It is worth crediting a management team that adjusts a headline number downward.

Why the Cash Flow Is So Extreme#↑ Contents

Operating cash flow went from $84 million to $11.7 billion. But the number to sit with is capital expenditure of $177 million — 0.9% of revenue.

Semiconductor manufacturing is famously capital-hungry; fabs cost tens of billions. Sandisk's fab exposure sits inside Flash Ventures iFlash VenturesThe manufacturing joint venture Sandisk runs with Kioxia, the former Toshiba memory business. The two companies jointly own and operate the fabrication plants in Japan that actually produce the NAND wafers, and share the output. Because the fabs sit inside the JV rather than on Sandisk's own books, the enormous cost of building them does not appear in Sandisk's capital expenditure line — which is why a semiconductor manufacturer can report capex of just 0.9% of revenue. The JV framework runs through December 2034., its long-running joint venture with Kioxia (the former Toshiba Memory), which is accounted for outside the capex line — notes receivable and investments in Flash Ventures sit at $678 million on the balance sheet, with a separate $275 million of net Flash Ventures activity adjusted out of free cash flow.

The effect is an unusually asset-light income statement for a chip company: 56.8% free cash flow margin, and 43.2% on the more conservative adjusted basis. When the pricing cycle turns up, almost none of the incremental revenue is absorbed by reinvestment.

The same structure is a strategic constraint, though. Sandisk does not unilaterally control its own capacity expansion — that runs through the JV. In a shortage, being unable to add supply quickly is excellent for pricing and awkward for market share.

The Balance Sheet Transformation#↑ Contents

Four observations:

Debt is gone. Sandisk repaid $1.9 billion during the year and ended with zero long-term debt. For a company that was carrying an impairment charge eighteen months ago, that is a decisive change in financial position.

Receivables grew 341% to $4.708 billion, faster than revenue's 175%. Some of that is simply timing on a quarter that was back-end loaded. It is still the line to watch next quarter — receivables growing faster than sales is how a pricing boom can flatter reported profit ahead of collected cash.

A $1.777 billion equity portfolio appeared. Sandisk purchased $970 million of marketable equity securities during the year and recorded an $804 million gain in the fourth quarter alone — a roughly 83% mark-up in months. The release does not disclose the holdings. On an $8.965 billion revenue quarter it is not material to the operating story, but it is $804 million of non-operating income that a reader should mentally separate from the business. Management does exactly that in the non-GAAP reconciliation.

Capital returns began immediately. Sandisk repurchased $4.524 billion of stock in the fourth quarter, and the board authorised an additional $14 billion, taking remaining authorisation to $15.5 billion — approximately equal to total shareholders' equity. Treasury stock went from zero to $4.537 billion.

Buying back stock at roughly the size of your own book value is an emphatic statement. It is also the classic move that memory companies have historically made near cycle peaks, which is the next section.

Guidance: The Next Quarter Is Bigger Than Last Year#↑ Contents

The midpoint of $10.55 billion would represent 18% sequential growth — and would, on its own, exceed the entire $7.355 billion of fiscal 2025 revenue by 43%.

Guidance holds gross margin at 83–85%, meaning management sees no near-term pricing relief. At the EPS midpoint of $45.00, a single quarter would out-earn the $2.99 that all of fiscal 2025 produced on a non-GAAP basis, roughly fifteen times over.

The State of the Global NAND Industry#↑ Contents

Sandisk's filing tells you what happened inside one company. It does not tell you what is happening to the industry — and in memory, the industry is the investment case. Here is the wider picture, drawn from external market data rather than the release.

The market: $46 billion in a single quarter#

Global NAND revenue reached $46 billion in the first quarter of calendar 2026, up roughly 246% year over year and 90% sequentially, according to Counterpoint Research. That is the entire industry roughly tripling in a year.

Market share as of Q1 2026:

Two things jump out. First, Samsung remains dominant but no longer overwhelming. Second — and far more consequential — four companies are effectively tied for third place inside a single percentage point. Kioxia, Micron, Sandisk and YMTC are separated by rounding. This is not a stable oligopoly; it is a scrum.

Why prices exploded#

The mechanism is straightforward supply-demand arithmetic. NAND bit demand in 2026 is running at roughly 20–22% growth against supply growth of only 15–17%. That gap is what took NAND contract prices up between 33% and 60% in Q1 2026, per TrendForce.

The consumer-facing effect is visible in shops: 1TB consumer SSDs have gone from roughly $45 to nearly $90 since late 2025. If you have wondered why memory cards and drives suddenly cost what they do, this is why — and it is the same phenomenon that produced Sandisk's 84.6% gross margin.

There is a second, less obvious driver. Manufacturers have been redirecting wafer capacity and engineering attention toward high-bandwidth memory (HBM) for AI accelerators, which is DRAM rather than NAND. Capacity that might have expanded NAND supply went to a more profitable adjacent product instead, tightening flash supply as a side effect.

China: YMTC is the story nobody is pricing#

This is the part of the industry picture that deserves the most attention, and the part most equity coverage skips.

Yangtze Memory Technologies (YMTC) went from 8% to 13% global NAND share in twelve months. Its Q1 2026 revenue was roughly $2.6 billion, up almost 445% year over year (Caixin, TechNode). It is now tied with Sandisk and Micron, and it has publicly targeted 15% share by the end of 2026.

The strategic detail matters more than the share figure. YMTC has been under US export controls since 2022, cutting it off from American semiconductor equipment. Its response has been to build a production line using domestically produced Chinese fab tools, with reporting indicating domestic equipment content has crossed 50% (Tom's Hardware).

The capacity plan is the number that should concern every incumbent. YMTC's two existing Wuhan fabs produce a combined 200,000 wafers per month. Two additional Wuhan fabs would take total capacity toward 500,000 wafers per month — which would make YMTC the largest NAND wafer producer in the world by output (NineScrolls). Mass production at a new Wuhan fab is slated for the second half of 2026, and analysts cited in the reporting expect YMTC's bit output could roughly double by year end.

Sanctions were intended to cap Chinese memory capability. On this evidence they have instead produced a domestically tooled competitor that is scaling during the most profitable period in the industry's history — and one that, unlike Western firms, is not primarily optimising for margin.

This is the single largest structural risk to Sandisk's current economics, and it is not visible anywhere in the earnings release.

The capacity split: who is building and who is not#

Here the industry divides in a way that is unusually favourable to Sandisk in the near term.

Kioxia and Sandisk are expanding aggressively. Their combined capital expenditure is reported at roughly $4.5 billion this fiscal year, up about 41% year over year — the most aggressive NAND expansion among established players (TrendForce). Fab2 (K2) at the Kitakami plant in Iwate Prefecture, Japan began operation in September 2025, with meaningful output from the first half of 2026, partly subsidised by the Japanese government. On 2 July 2026 the partners began producing 10th-generation 3D flash there. The joint venture framework has been extended through December 2034.

Samsung and SK hynix are doing the opposite. Both are reported to be limiting NAND investment while redirecting capital toward DRAM and HBM, where AI accelerator demand is even more acute.

Read those two paragraphs together and you get the near-term setup: the two largest players are stepping back from NAND capacity precisely as the Japan–US alliance steps forward. That is why Sandisk's volumes grew at all in a shortage — and why its share held while YMTC's rose.

The technology race: layers versus density#

The engineering contest is genuinely interesting and is being reported slightly wrong in most places.

Kioxia and Sandisk's BiCS10 stacks 332 layers and reaches an industry-leading TLC density above 29 Gb/mm² iGb/mm² — areal densityGigabits of storage per square millimetre of silicon. This is the number that actually matters commercially, because a wafer costs roughly the same to process regardless of how much data you fit on it — so higher density means lower cost per gigabyte. Layer count gets the headlines, but two chips with the same layer count can have very different densities depending on how tightly the cells are packed laterally., with QLC versions reported above 37 Gb/mm² (StorageReview, Tom's Hardware). Samsung unveiled its 400-plus-layer V10 BV-NAND in early August 2026, at approximately 28 Gb/mm² TLC density (TrendForce).

So the company with fewer layers has higher density. Kioxia and Sandisk deliberately capped BiCS10 at 332 layers rather than chase past 400, on the reasoning that beyond that threshold power consumption rises sharply and thinner cell layers compromise charge retention and long-term reliability.

That is a real technical position, not marketing. It also matters commercially: layer count is what gets into headlines, but density per square millimetre is what determines cost per bit, and cost per bit is what wins datacenter contracts.

Other regions#

Japan is the second pillar of Sandisk's manufacturing base and is actively subsidising it — the Kitakami Fab2 investment received Japanese government support under a plan approved in February 2024. Japan's industrial policy is effectively underwriting the main Western-aligned counterweight to Chinese NAND capacity.

South Korea remains the centre of gravity at 47% combined share between Samsung and SK hynix, but its incremental capital is going to DRAM and HBM rather than flash.

India is worth addressing carefully, because it is frequently overstated. India now has real memory packaging capability — Micron's $2.75 billion ATMP facility at Sanand, Gujarat was inaugurated in February 2026 and packages imported wafers into DRAM modules and NAND SSDs. Thirteen semiconductor projects have been approved across seven states. But ATMP is assembly, test, marking and packaging — not wafer fabrication. India is not currently a NAND manufacturing country and will not be one within this cycle. It is a downstream participant, and the distinction matters.

Europe has essentially no presence in NAND manufacturing, though it retains critical positions in equipment and materials upstream.

When does this end?#

The consensus among industry analysts is more specific than usual, and it is not comforting for anyone extrapolating today's margins.

NAND is expected to shift from supply-constrained to balanced, and then toward oversupply, by the second half of 2027 — diverging from DRAM, which is expected to stay tight for longer. New fabs commissioned during this boom are not expected to add meaningful supply until late 2027 or 2028, which is precisely when they arrive into a softer market.

Two demand-side warnings sit underneath that. Smartphone production is forecast to decline roughly 15–20% year over year and notebook shipments around 10% in 2026. Those are the volume markets that historically absorbed NAND bits. If consumer device demand keeps shrinking while bit supply expands through higher layer counts, enterprise SSD demand alone may not absorb the output.

The historical pattern is unambiguous: memory supercycles have consistently peaked and rolled over within one to two years, because elevated profitability triggers exactly the capital investment that ends them. Some analysts argue this cycle's correction will be gentler than past collapses given structurally stronger AI demand. That is a reasonable argument. It is not a guarantee, and it has been made near the top of every previous cycle.

What Could Break This#↑ Contents

We do not write victory laps, and this report demands the caveat section more than most.

1. Memory is the most cyclical industry in technology. Full stop. NAND and DRAM pricing has collapsed by 40–60% in multiple prior downturns — 2008, 2012, 2015-16, 2018-19, and 2022-23. Sandisk's own fiscal 2025, with a 30% gross margin and a $1.6 billion loss, was the last trough, and it ended barely twelve months ago. An 84.6% gross margin is not a new normal. It is the top of a cycle whose amplitude is the defining feature of the industry.

2. The cost structure offers no downside protection. The same arithmetic that produced a 97% incremental gross margin on the way up works identically in reverse. Cost of revenue is already near its floor. If NAND pricing halves, roughly half of that revenue decline lands directly on gross profit. Model the downside honestly.

3. Supply is coming, and the timeline is now fairly specific. Industry analysts expect NAND to move from constrained to balanced and then toward oversupply by the second half of 2027, with new fab output arriving in force through 2028. Sandisk's own joint venture is part of the problem: Kioxia–Sandisk capex is up roughly 41% to about $4.5 billion, and capacity commissioned at the top of a cycle has a long history of arriving into the bottom of the next one.

4. YMTC is the structural threat, and it is not in the filing. China's Yangtze Memory went from 8% to 13% global share in a year, is targeting 15%, and is building out toward a reported 500,000 wafers per month across four Wuhan fabs using increasingly domestic equipment. It is tied with Sandisk today. A competitor scaling aggressively with state backing and a lower cost of capital does not behave like a margin-maximising oligopolist — it behaves like a share-taker, which is how price floors break.

5. The consumer volume markets are shrinking. Smartphone production is forecast down 15–20% and notebooks down roughly 10% in 2026. Those markets historically absorbed the bits. If they keep contracting while layer-count advances expand supply, enterprise SSD demand has to carry more than it ever has.

6. Customer concentration and prepayment risk. Ten NBM agreements is a small number of relationships carrying an enormous amount of revenue, and hyperscaler capex plans can change quickly. Prepayments also mean some future revenue has already been collected — a headwind to reported growth when those contracts deliver.

7. Receivables and working capital. Accounts receivable grew twice as fast as revenue. Refund liabilities rose from $126 million to $1.5 billion. Neither is alarming in a quarter this size, but both are lines where a pricing peak first shows up as a collections problem.

8. Governance and disclosure gaps. An undisclosed $1.777 billion equity portfolio producing an $804 million gain is unusual for a semiconductor manufacturer. It is not a red flag on this evidence — but it is worth understanding what it is before treating it as recurring.

The honest framing: this is an outstanding business moment, captured by a company with real technology and an asset-light structure, in an industry that has never once sustained these economics. Both halves of that sentence are true.

Peers and the Memory Complex#↑ Contents

Sandisk sits inside a small oligopoly. Watching the others is how you tell whether this is a Sandisk story or an industry story — and it is almost certainly the latter.

Micron Technology (NASDAQ: MU) — The only US-headquartered manufacturer of both DRAM and NAND, and the closest listed comparison. Micron carries high-bandwidth memory (HBM) exposure that Sandisk does not, which makes it a broader AI memory play. Run by Sanjay Mehrotra, a Sandisk co-founder — a detail worth savouring. Next earnings: [ADD DATE]

SK Hynix (KRX: 000660) — The HBM leader and, through its Solidigm subsidiary (the former Intel NAND business), a direct enterprise SSD competitor. Its results are the single best read on whether AI memory demand is broad or concentrated. Next earnings: [ADD DATE]

Samsung Electronics (KRX: 005930) — The largest memory manufacturer on earth and the swing factor in every NAND cycle. Samsung has historically been willing to add capacity into weakness to defend share, which is the mechanism that has ended prior upcycles. Its capex commentary matters more than its earnings. Next earnings: [ADD DATE]

Kioxia Holdings (TYO: 285A) — Sandisk's Flash Ventures joint venture partner, so their manufacturing economics are directly linked. Reads as confirmation rather than competition. Next earnings: [ADD DATE]

Western Digital (NASDAQ: WDC) — The former parent, now a pure hard-disk-drive business after the separation. HDD still dominates cold storage for hyperscalers, so WDC's results indicate whether AI storage demand is lifting all media or shifting toward flash. Next earnings: [ADD DATE]

Seagate (NASDAQ: STX) — The other HDD major. Same read-across as WDC. Next earnings: [ADD DATE]

The demand side: Microsoft, Amazon, Google, Meta and Oracle capital expenditure guidance. As we noted in Microsoft's FY26 filing, a single hyperscaler put $115.9 billion into property and equipment in one fiscal year. Sandisk's revenue is a downstream function of that spending. If hyperscaler capex guidance flattens, the memory complex is where it will show up second — right after the accelerator names.

Two tests worth running:

1. Does Micron report comparable NAND gross margin expansion? If yes, this is an industry-wide pricing cycle. If Sandisk is meaningfully ahead, part of this is company-specific execution and mix.
2. Do Samsung and SK Hynix announce NAND capacity additions? That is the leading indicator for when this ends. Watch capex announcements, not earnings.

Financial Gurkha Takeaways#↑ Contents

1. This is a pricing cycle, not a volume story — and the filing says so. Management attributes two-thirds of sequential growth to price. Cost of revenue fell year over year. Understand the mechanism and you understand both the upside already realised and the downside not yet priced.

2. Datacenter is now structural. From $960 million to $5.153 billion in a year, with a doubling in the last quarter alone. Whatever happens to NAND pricing, AI has permanently changed the composition of Sandisk's demand.

3. Consumer being cut 32% sequentially is the scarcity proof. Companies with capacity do not fire their retail customers. That single line is worth more than any management adjective.

4. The prepayments are the forward indicator. $1.242 billion of contract liabilities and $2.476 billion of NBM cash. Customers are paying in advance to guarantee supply — and management is conservatively excluding it from adjusted free cash flow.

5. The asset-light structure is the reason the cash flow is this extreme. 0.9% capex intensity via Flash Ventures means almost nothing absorbs the upswing. It also means Sandisk cannot rapidly add its own capacity.

6. Respect the cycle. Fiscal 2025 — a $1.6 billion loss, a $1.8 billion goodwill impairment, 30% gross margin — was fourteen months ago. The same company. The same assets. An 85% gross margin business and a 26% gross margin business can be the same business in consecutive years, and in this industry they routinely are.

Frequently Asked Questions#↑ Contents

What were Sandisk's fiscal 2026 results? Sandisk reported fiscal 2026 revenue of $20.248 billion, up 175% from $7.355 billion, with GAAP net income of $11.433 billion ($73.76 diluted EPS) versus a net loss of $1.641 billion in fiscal 2025. Non-GAAP diluted EPS was $70.88.

What was Sandisk's Q4 FY2026 revenue and EPS? Fourth quarter revenue was $8.965 billion, up 51% sequentially and 372% year over year. GAAP diluted EPS was $43.97 and non-GAAP diluted EPS was $39.25.

Why did Sandisk's gross margin increase so much? Gross margin rose from 26.2% to 84.6% year over year in Q4 because NAND flash pricing increased sharply while production costs stayed flat. Cost of revenue actually declined slightly year over year ($1.403 billion to $1.383 billion) even as revenue rose 372%, so nearly all incremental revenue fell through to gross profit. Management attributed roughly two-thirds of sequential growth to price and one-third to volume. Check DRAM and NAND Prices from Stanford DAM

How much of Sandisk's revenue comes from data centers? Datacenter revenue was $5.153 billion in fiscal 2026, up 437% from $960 million, representing about 25% of total revenue. In the fourth quarter alone it reached $2.977 billion, roughly fourteen times the $213 million recorded a year earlier.

Why did Sandisk's consumer revenue decline? For the full fiscal year, consumer revenue actually rose 29% to $2.935 billion. The decline is a fourth-quarter effect: $556 million, roughly 30% below the average of the preceding three quarters. Seasonality is the obvious explanation, but in fiscal 2025 the equivalent quarter ran slightly above that year's average, which argues against it. Because NAND pricing rose sharply during the year while consumer revenue grew only modestly, consumer unit volumes very likely declined — the segment is being repriced rather than abandoned. Sandisk does not disclose unit volumes, so this is an inference.

What is included in Sandisk's Datacenter, Edge and Consumer segments? Sandisk does not define these end markets in its earnings release, which is a genuine disclosure gap. Based on the company's own product and customer descriptions, Datacenter most likely covers enterprise and hyperscale SSDs sold to "the largest enterprises and public clouds"; Edge most likely covers embedded products and client drives built into other manufacturers' devices, and possibly raw wafers and components; and Consumer covers the retail Sandisk-branded memory cards, USB flash drives and portable SSDs. The placement of "wafers and components" is unconfirmed and matters, because Edge alone is 60.1% of revenue.

How much of Sandisk's revenue is consumer versus business? Consumer was $2.935 billion of $20.248 billion in fiscal 2026, or 14.5% — down from 30.8% the prior year. Despite Sandisk being one of the most recognised consumer storage brands in the world, roughly 85% of its revenue now comes from datacenter and edge customers rather than retail shoppers.

What are Sandisk's New Business Model (NBM) agreements? NBM agreements are long-term supply arrangements that include customer prepayments and deposits. Sandisk has signed ten to date. They are visible on the balance sheet as contract liabilities, which rose from $25 million to $1.242 billion, and in the cash flow statement as $2.476 billion of NBM prepayments and deposits that management excludes from adjusted free cash flow.

What is Sandisk's guidance for Q1 fiscal 2027? Revenue of $10.30–10.80 billion, non-GAAP gross margin of 83.0–85.0%, non-GAAP operating expenses of $520–540 million, and non-GAAP diluted EPS of $44.00–46.00 on approximately 155 million diluted shares.

How much stock is Sandisk buying back? Sandisk repurchased $4.524 billion of common stock in the fourth quarter. The board approved an additional $14 billion authorisation, bringing total remaining authorisation to $15.5 billion — approximately equal to the company's $15.736 billion of shareholders' equity.

Does Sandisk have any debt? No. Sandisk repaid $1.9 billion of debt during fiscal 2026 and ended the year with zero long-term debt and $4.762 billion of cash and cash equivalents.

Is Sandisk an AI stock? Directly, yes. NAND flash storage is required infrastructure for AI training checkpoints, inference serving and retrieval datasets, and Sandisk's datacenter segment grew 437% in fiscal 2026 on that demand. Its revenue is a downstream function of hyperscaler capital expenditure.

What is the relationship between Sandisk and Western Digital? Western Digital acquired Sandisk in 2016 and spun it back out as an independent public company on 21 February 2025. Western Digital retained the hard disk drive business; Sandisk holds the NAND flash business.

What is Flash Ventures? Flash Ventures is Sandisk's manufacturing joint venture with Kioxia Corporation, the former Toshiba Memory business. Because fabrication sits inside the JV, Sandisk's own capital expenditure was just $177 million on $20.2 billion of revenue — roughly 0.9% — which is the main reason its free cash flow margin reached 57%.

Who are the largest NAND flash manufacturers in 2026? As of the first quarter of calendar 2026, global NAND market share was approximately: Samsung 29%, SK hynix 18% (including Solidigm), Kioxia 14%, Micron 13%, Sandisk 13% and YMTC 13%. The global NAND market reached a record $46 billion in that quarter, up roughly 246% year over year, according to Counterpoint Research.

Why are NAND flash and SSD prices rising in 2026? NAND bit demand is growing roughly 20–22% in 2026 against supply growth of only 15–17%, driven by AI datacenter storage requirements. Contract prices rose between 33% and 60% in the first quarter of 2026. Manufacturers have also redirected capacity and engineering resources toward high-bandwidth memory for AI accelerators, further constraining flash supply. Consumer 1TB SSD prices have roughly doubled since late 2025.

Is China a threat in the NAND flash market? Yes, and it is the most significant structural competitive development. China's YMTC grew global NAND share from 8% to 13% in a year, with Q1 2026 revenue up almost 445% to about $2.6 billion, and has targeted 15% share by the end of 2026. Despite US export controls since 2022, YMTC has built a production line using domestically produced Chinese equipment, with reported domestic tool content above 50%. Plans for two additional Wuhan fabs would take capacity toward 500,000 wafers per month, potentially making YMTC the world's largest NAND wafer producer by output.

Which NAND manufacturers are expanding capacity? Kioxia and Sandisk are expanding most aggressively, with combined capital expenditure reported at roughly $4.5 billion, up about 41% year over year, centred on Fab2 at the Kitakami plant in Japan, which began 10th-generation production in July 2026. Samsung and SK hynix are reported to be limiting NAND investment while prioritising DRAM and high-bandwidth memory. YMTC is expanding rapidly in China.

What is BiCS10 and how does it compare to Samsung's V10 NAND? BiCS10 is the tenth-generation 3D NAND jointly developed by Kioxia and Sandisk. It stacks 332 layers and achieves TLC density above 29 Gb/mm², with QLC versions reported above 37 Gb/mm². Samsung's V10 BV-NAND, unveiled in August 2026, uses more than 400 layers at approximately 28 Gb/mm² TLC density. Kioxia and Sandisk deliberately capped layer count at 332 because beyond roughly 400 layers power consumption rises sharply and thinner cell layers compromise charge retention and reliability. Density per square millimetre, not layer count, determines cost per bit.

When will the NAND shortage end? Industry analysts generally expect NAND to move from supply-constrained to balanced and then toward oversupply by the second half of 2027, diverging from DRAM which is expected to remain tight longer. New fab capacity commissioned during the current boom is not expected to add meaningful supply until late 2027 or 2028. Smartphone production is forecast to decline 15–20% and notebook shipments about 10% in 2026, which reduces the volume demand that historically absorbed NAND output.

Does India manufacture NAND flash memory? No. India has memory packaging and assembly capability but not wafer fabrication. Micron's $2.75 billion ATMP facility at Sanand, Gujarat, inaugurated in February 2026, imports finished wafers from Micron's global fabs and packages them into DRAM modules and NAND SSDs. ATMP is assembly, test, marking and packaging — a downstream activity distinct from manufacturing the memory itself.

What are the main risks to Sandisk's results? Memory is the most cyclical sector in technology, with NAND pricing having fallen 40–60% in multiple past downturns. Sandisk's own fiscal 2025 produced a $1.6 billion loss at a 30% gross margin. With cost of revenue already near its floor, a pricing reversal would flow almost directly to gross profit. Additional risks include competitor capacity additions from Samsung, SK Hynix, Micron and YMTC, customer concentration across ten NBM agreements, and hyperscaler capital expenditure plans.

About This Analysis#↑ Contents

Financial Gurkha is an independent markets research desk run by Kanchan Sharma from New York City. We read the filings other coverage summarises — every figure above is sourced to Sandisk's fiscal Q4 and full year 2026 earnings release, and where we have computed or derived a figure, we have shown the arithmetic.

Need a full intrinsic valuation — for a public company, a private target, or your own business? Schedule a consultation with the Financial Gurkha desk.

Reported from New York City. Source: Sandisk Corporation fiscal Q4 2026 earnings release (SEC Exhibit 99.1, filed 5 August 2026).