Sandisk
Sandisk makes NAND flash, the memory chips that store data in phones, laptops and the drives of AI data centres.
The scores
Growth3/4 ↑
Revenue rose 175% in fiscal 2026 with no acquisition, but two-thirds of Q4's jump was price, and it grew slowest of the five makers.
Business quality2/4 ↑
Scale shared with four larger rivals selling interchangeable chips; fiscal 2023 lost $2.0B and the contracts, not the product, hold today's margins.
Financial position4/4 ↑
$4.8B of cash and no debt after $4.5B of buybacks; customers prepay and the factory commitments are covered by operating cash.
Execution record3/4 ↑
Beat the profit consensus in all seven quarters since listing and topped Q4 guidance, but high-bandwidth flash samples slipped a year.
Per share capture4/4 ↑
Share count fell to 145.3M from 146M while $4.5B bought back 2.8M shares in Q4; $15.5B of authorisation remains.
Potential3/4 ↑
$59.8B of obligations are filed and $93.9B claimed at floor, but floor prices are undisclosed and supply balances in 2027.
Risk profile3/4 →
No customer above 10% and no debt, but all output comes from shared Japanese fabs and Chinese capacity grows toward 19% of bits.
Valuation3/4 →
7.4 times annualised Q4 operating profit; the contracts and the Q1 guide cover the profit the price needs for several years, not indefinitely.
Evidence quality3/4 →
Statements, obligations and the warnings are filed and market share measured by TrendForce, but floors and contract margins rest on management.
No upside the evidence defends · if it fails you keep about 40%.
The Verdict
We rate Sandisk Neutral. It is a solid business at a fair price. The price is fair while today's record profits hold: it needs more than fiscal 2026's record profit every year. Long contracts with eight customers protect the next few years, but their floor prices are secret, and every past flash boom gave its margins back once supply caught up. As a bet, its new high-bandwidth flash has no customer or revenue yet, so there is no real upside to count, and the price already sits well above what the flash business alone would be worth at the low prices memory makers usually fetch.
The long caseAI inference needs more flash than the industry can build, and Sandisk has sold years of it in advance.
The evidence under this - 8 claims
The 10-K says the financial guarantees may not fully offset revenue lost if a customer fails to perform (S01).
- The floor prices and the margin at floor are not disclosed anywhere in the 10-K (S01).
- Guarantees of $16.5B cover about 18% of the $91.1B of obligations, and the 10-K says they may not fully offset lost revenue (S01).
On the Q1 guide the run-rate operating profit is near $33B a year, so the multiple is nearer 6 times on the next quarter (C05).
Management says actual NBM revenue will exceed the floor-based minimum and that margins are attractive even at floor (S03).
The company
Sandisk designs and sells NAND flash: chips that keep data with the power off. They go into phones, laptops and cars, into the solid-state drives that hold data in AI data centres, and into its own memory cards and USB sticks.
It was split off from Western Digital in February 2025. It is the fifth-largest flash maker, and all of its chips are made in Japanese factories it shares with Kioxia, a rival.
The evidence under this - 4 claims
Kioxia and Sandisk announced more than $31B of Japanese investment through 2032, which reads as new capacity rather than node migration alone (S10).
How it makes money
Sandisk sells flash by the gigabyte, so its revenue is the amount it ships times the price per gigabyte. In fiscal 2026 data centres were a quarter of sales, devices like phones and PCs 60% and its own branded products the rest.
Half its chips now go out under multiyear contracts with minimum prices and volumes. The other half sells at whatever the market pays, so when flash prices rise, almost all of the rise is profit.
The evidence under this - 4 claims
The 10-K says the financial guarantees may not fully offset revenue lost if a customer fails to perform (S01).
Datacenter exabytes rose almost 120% in fiscal 2026, a real volume gain (C04).
Where the money comes from, stream by stream
| Stream | Share of sales (FY 2026) | How it earns it |
|---|---|---|
| Datacenter | 25.4% | Enterprise SSDs for cloud and AI servers, much of it now under multiyear NBM contracts. |
| Edge | 60.1% | Flash and SSDs built into phones, PCs, game consoles and cars, sold to device makers. |
| Consumer | 14.5% | Sandisk- and WD-branded memory cards, USB drives and portable SSDs sold through retailers. |
Who runs it
David Goeckeler, chief executive and chair, has run the flash business since 2020, under Western Digital and since the split. Directors and officers own under 1% of the company, and almost all their sales in the past year were pre-arranged. The bonus rewards operating profit, margin and cash flow, and paid its maximum for fiscal 2026.
The evidence under this - 3 claims
Why it wins, and who could beat it
Why customers come back
Flash is hard to make: a factory costs billions and takes years, so only five companies make most of it. That scale keeps new rivals out, but it does not make Sandisk's chips special; a buyer can get similar ones from four others.
What keeps customers today is the shortage. Most new supply is already sold, so big buyers signed long contracts and paid deposits to be sure of getting flash. Those deposits, not the product, are what tie them to Sandisk.
The evidence under this - 5 claims
The 10-K says the financial guarantees may not fully offset revenue lost if a customer fails to perform (S01).
- The floor prices and the margin at floor are not disclosed anywhere in the 10-K (S01).
- Guarantees of $16.5B cover about 18% of the $91.1B of obligations, and the 10-K says they may not fully offset lost revenue (S01).
The threat
Samsung, SK hynix, Micron and Kioxia sell the same kind of chips, and all four grew faster than Sandisk in Q2 2026. Chinese makers are adding capacity toward about a fifth of world output. And new software from Nvidia and Google squeezes the memory an AI model needs for each question, which could slow demand.
The evidence under this - 4 claims
Sandisk says bits will stay on allocation beyond calendar 2027 (S03).
VDURA measured 30 TB QLC capacity at 22.6 times the cost of hard-drive capacity in Q1 2026, so hard drives still win on cost (S17).
The rivals: MU, NVDA, GOOGL, and 4 more that are private, in-house or listed abroad.
Who else does this, rival by rival
| Ticker | Rival | Where it competes | How far along | Priced against it |
|---|---|---|---|---|
| Samsung ElectronicsPriced against it: Listed abroad | Datacenter | On sale | Listed abroad | |
| Samsung Electronics | Consumer | On sale | ||
| SK hynix and its Solidigm unitPriced against it: Listed abroad | Datacenter | On sale | Listed abroad | |
| SK hynix | High Bandwidth Flash | On sale | ||
| MU | Micron TechnologyPriced against it: One segment only | Datacenter | On sale | One segment only |
| KioxiaPriced against it: Listed abroad | Edge | On sale | Listed abroad | |
| Private | Yangtze Memory Technologies (YMTC)Priced against it: Private | Edge | On sale | Private |
| NVDA · GOOGL | Nvidia's KV-cache compression and Google's TurboQuantPriced against it: Not checked | Datacenter | On sale | Not checked |
How big it could get
How big this could get
Management expects the whole flash market to top $300B in 2026, three times the year before, and approach $500B in 2027, with data centres the largest buyer. That is the company's own estimate.
Outside data agrees on direction: TrendForce sees data-centre drive demand up more than 80% in 2026 and contract prices still rising. No outside body sizes the market as far out as management does.
The evidence under this - 3 claims
TrendForce's Q2 figures imply about $315B a year at mid-2026 prices, consistent with 2026, but TrendForce expects supply to catch up in the second half of 2027, which a $500B 2027 does not assume (S06, S07).
Who has signed
Eight customers signed multiyear must-buy contracts. The annual report records $59.8B still to be delivered, about a fifth of it within a year. Management says all its contracts are worth at least $93.9B at the floor price, and cover over half of fiscal 2027's output and about two-thirds of fiscal 2028's.
The evidence under this - 2 claims
The 10-K says the financial guarantees may not fully offset revenue lost if a customer fails to perform (S01).
- The floor prices and the margin at floor are not disclosed anywhere in the 10-K (S01).
- Guarantees of $16.5B cover about 18% of the $91.1B of obligations, and the 10-K says they may not fully offset lost revenue (S01).
Where the growth is coming from, business by business
Where the growth is coming from
| Business | Share of revenue | Growth: last quarter | Last half-year | Last year |
|---|---|---|---|---|
| Edge | 62% | +295.1% | - | - |
| Datacenter | 25% | +644.7% | - | - |
| Consumer | 14% | +43.6% | - | - |
| Whole company | 100% | +251.0% | - | - |
The filed split through 2026-04-03. These rates are not carried forward: the reading below says why.
How we read it: Datacenter grew 437% against Edge and Consumer far slower, and it is now a quarter of revenue; but prices rose in every market, so the blend is mostly a shortage lifting all three.
The results
The numbers
Sales grew 175% in the year to Jul 2026, faster than the 10% in the year before.
Operating margin is 61%, up from −33% in the year to Jun 2023.
Running costs take 10¢ of each sales dollar, down from 41¢ in the year to Jun 2023.
Shareholders' money earned 92% a year, up from −19% in the year to Jun 2023.
It turned 57% of sales into free cash, $11.5B over the year.
It holds $4.76B of cash against $0 of debt, $4.76B more cash than it owes.
The share count rose 6.9% in 3.0 years, and pay in shares costs 1.1% of sales.
All the numbers
Filed figures from the company's 10-K and 10-Q, read 2026-10-20. Fiscal years end in July.
| Jun 2023 | Jun 2024 | Jun 2025 | Jul 2026 | |
|---|---|---|---|---|
| Size and growth | ||||
| Sales | $6.09B | $6.66B | $7.36B | $20.2B |
| Sales growth | - | +9.5% | +10% | +175% |
| Profit | ||||
| Gross margin | 7.1% | 16% | 30% | 71% |
| Operating margin | −33% | −7.0% | −19% | 61% |
| Net margin | −35% | −10% | −22% | 56% |
| What growth costs, per sales dollar | ||||
| Research and development | 19% | 16% | 15% | 6.6% |
| Selling, general and admin | 9.2% | 6.8% | 7.8% | 3.3% |
| Pay in shares | 2.7% | 2.2% | 2.5% | 1.1% |
| Capital spending | 3.6% | 2.5% | 2.8% | 0.9% |
| Cash | ||||
| Cash from operations | −$713M | −$309M | $84.0M | $11.7B |
| Free cash flow | −$932M | −$475M | −$120M | $11.5B |
| Balance sheet | ||||
| Cash and short-term investments | - | $328M | $1.48B | $4.76B |
| Debt | - | $0 | $1.85B | $0 |
| Net cash | - | $328M | −$368M | $4.76B |
| Returns and shares | ||||
| Return on capital | - | - | - | 81% |
| Shares (diluted) | 145M | 145M | 145M | 155M |
What growth costs
Growth made Sandisk far more profitable, because costs barely moved while prices rose: research and overheads fell from about a quarter of sales to a tenth. The same lever works in reverse. In fiscal 2023, when prices collapsed, sales fell and the company lost $2.0B.
The evidence under this - 3 claims
The last report (5 Aug)
Quarter ended July 2026, filed figures.
In the fourth quarter of fiscal 2026, to July, sales rose 51% from the quarter before and nearly fivefold from a year earlier. Most of the jump was price, not volume. Gross margin reached 84.6%, against 26.2% a year earlier.
Management guided the next quarter higher again, to $10.3B to $10.8B of sales at a similar margin. The annual report filed two weeks later showed $59.8B of contracted sales still to deliver.
What it did to the thesisIt proved the shortage is real and that customers will sign years ahead at high prices. It left open what the contracts earn once prices stop rising, because the floors were not disclosed.
The quarter in detail
Sales growth from the previous quarter, filed figures.
Earnings per share $39.25 against $34.96 expected.
The evidence under this - 4 claims
The guided gross margin midpoint sits below Q4's 84.6% even with prices still rising, which analysts on the call questioned (S03).
The 10-K says the financial guarantees may not fully offset revenue lost if a customer fails to perform (S01).
What management did with the money
Sandisk repaid its loan early, paid $970M for a stake in Nanya, a Taiwanese maker of DRAM memory chips, and bought back $4.5B of its own shares in one quarter, with $15.5B more approved. It pays no dividend.
Insiders in 2026Q3: 6 purchases and 86 sales, from their Form 4 filings.
The evidence under this - 3 claims
The price, and where we disagree
In short. Fair. Today's profit pays for the price for a few years, and the contracts cover most of that, but not beyond, and flash profits have never lasted.
The shares have fallen twice on fears about Chinese memory makers: by 55% in July and again by more than a quarter in October, while Sandisk's own sales kept rising. They now sit about 42% below their June high.
The next test comes on 29 October 2026, with results for the first quarter of fiscal 2027. They will show whether sales and margins hit the guide.
The evidence under this - 4 claims
The guided gross margin midpoint sits below Q4's 84.6% even with prices still rising, which analysts on the call questioned (S03).
How we read the multiple
Buyers pay about $207.5B for the whole company. Flash profits swing hard: fiscal 2026 earned a record $12.4B of operating profit, but in fiscal 2023 the same business lost $2.0B.
To earn 6 cents after tax for each dollar paid, a common rule of thumb for what a share should earn, Sandisk must:
- Earn about $14.6B of operating profit a year.
- That is more than the record fiscal 2026, every year.
- The latest quarter runs far above it, and the contracts cover the next few years, not beyond.
Against its own history. Sandisk has traded on its own for under two years, all of it in a boom, so its own normal says little about a whole cycle.
The evidence under this - 5 claims
On the Q1 guide the run-rate operating profit is near $33B a year, so the multiple is nearer 6 times on the next quarter (C05).
Management says actual NBM revenue will exceed the floor-based minimum and that margins are attractive even at floor (S03).
Fiscal 2028 consensus assumes revenue of $58.5B, above the contracted floor, which depends on market prices holding through 2028 (C23, C11).
Priced against its rivals
None of the listed rivals above sells the same thing to the same buyers as a large part of what it does. A price comparison needs at least two, so the price is judged against its own history instead.
Private or foreign rivals that cannot be priced: Samsung Electronics Co., Ltd., SK hynix Inc., Kioxia Holdings Corporation, Yangtze Memory Technologies Co., Ltd.
The bet: what it pays if it works
The bet is high-bandwidth flash: stacks of flash placed beside AI chips to hold more data than today's fast memory. A first chip is designed and an industry standard published, but samples have slipped to 2027 and no customer has signed.
Nothing yet supports a revenue figure, so there is no upside we can count. If it fails, no current profit is lost, but the price already sits well above what the flash business would be worth at a normal memory valuation.
The steps it has to take
- Done First high-bandwidth flash die taped out
- Done Open specification published through the Open Compute Project
- Slipped Inference product samples to customers due 2027-12
- Not started A named customer qualifies it for an AI accelerator due 2028-06
- Not started Volume revenue disclosed in the accounts due 2029-06
The detail behind the label
The evidence under this - 4 claims
The figures are vendor numbers on unreleased silicon, and industry expectations put commercialisation in late 2027 to 2028 with volume demand near 2030 (S11, S12).
On the Q1 guide the run-rate operating profit is near $33B a year, so the multiple is nearer 6 times on the next quarter (C05).
Where we disagree with the market
A memory boom near its peak: profits this high always draw new supply, so pay little for each dollar of today's profit.
The contracts are new to flash and protect the near years, but no flash boom has kept such profits once supply caught up.
The market has twice cut the shares hard on Chinese supply news while Sandisk's sales rose, pricing the cycle rather than the quarter. We agree the cycle matters, but the must-buy contracts are new: nothing like them held up past flash booms.
Against us: the floor prices are secret, a contract breach may not be fully covered, and supply is expected to catch up with demand in the second half of 2027.
What analysts say, and the evidence
The evidence under this - 7 claims
The 10-K says the financial guarantees may not fully offset revenue lost if a customer fails to perform (S01).
- The floor prices and the margin at floor are not disclosed anywhere in the 10-K (S01).
- Guarantees of $16.5B cover about 18% of the $91.1B of obligations, and the 10-K says they may not fully offset lost revenue (S01).
Sandisk says bits will stay on allocation beyond calendar 2027 (S03).
What could go wrong, and how we would know
The risks
- Supply catches up TrendForce expects the shortage to end in the second half of 2027, and every past glut sent margins down hard.
- Contracts may not hold The annual report warns a customer breach may not be fully covered by guarantees, and the volume may not resell at the same price.
- Factories shared with a rival All its chips come from ventures shared with Kioxia; Sandisk pays half the fixed costs whatever it orders.
- Smallest of five It grew slowest of the five makers in Q2 2026, and Chinese capacity keeps growing.
- Smarter software Compression from Nvidia and Google cuts the memory each AI question needs, which could slow data-centre demand.
The evidence under this - 6 claims
Sandisk says bits will stay on allocation beyond calendar 2027 (S03).
VDURA measured 30 TB QLC capacity at 22.6 times the cost of hard-drive capacity in Q1 2026, so hard drives still win on cost (S17).
What we still do not know
- The floor prices in the contracts, and what margin Sandisk earns at them.
- Whether customers sign more contracts covering fiscal 2029 and beyond on similar terms.
- Whether high-bandwidth flash finds a buyer, and when.
The evidence under this - 3 claims
- The floor prices and the margin at floor are not disclosed anywhere in the 10-K (S01).
- Guarantees of $16.5B cover about 18% of the $91.1B of obligations, and the 10-K says they may not fully offset lost revenue (S01).
Management says actual NBM revenue will exceed the floor-based minimum and that margins are attractive even at floor (S03).
What would change our view
- Neutral→AttractiveThe floors are shown to hold
It needs all of:
- ○Sandisk discloses its contract floor prices, or a gross margin at the floor of 60% or more.checked at the next results, due by 29 Oct
- ○Contracts cover two-thirds of fiscal 2029 output on similar terms.checked at the next results, due by 29 Oct
- Neutral→AttractiveA price that pays for the contracted years alone
It needs:
- ○The company's value falls to 5 times its latest quarter's operating profit over a year, with the margin held.checked daily
- Neutral→UnattractiveThe cycle turns on Sandisk's own numbers
It needs any of:
- ○Gross margin falls below 70% while flash contract prices fall.checked at the next results, due by 29 Oct
- ○A customer cuts, defers or defaults on a contracted volume.watched as it happens
We would also be wrong if
- Kioxia and Sandisk deadlock on investing in their shared factories, or the factories' equipment leases are called.
What is coming
- 29 October 2026 Q1 fiscal 2027 results: Whether revenue and margin hit the guide and how contracted obligations moved.
Every dated event ahead
- 29 October 2026 Q1 fiscal 2027 results: Whether revenue and margin hit the guide and how contracted obligations moved.
- 31 December 2026 Western Digital disposes of its last Sandisk shares: The end of a known seller; the remaining stake is about 1.7M shares or fewer.
- 31 January 2027 Q2 fiscal 2027 results: Whether contract prices still rose in calendar Q4 2026 and whether new NBMs were signed.
- 31 December 2027 First HBF inference product samples: Whether the early line reaches a customer's hands on the revised schedule.
Since we wrote this
- 20 OctThesis writtenNeutral, medium convictionWritten nine days before Q1 fiscal 2027 results, after the shares fell 27% in October on Chinese memory news.
Sources 19
Sources - 19
Every claim above cites one of these, or states the reasoning that derived it. The quote is frozen at the date it was read: a link can rot, and an agent fetch returns a reading of a page rather than the page. Backs counts the whole claim ledger, so it can name a claim this page does not surface.
- S01primarycompany filing2026-08-17the business, competition, NBM terms and obligations, Flash Ventures, end markets, cash flow, Nanya, risk factorsbacksC01C02C04C06C08C09C16C18C19C26C30C31C32C35
- S02primarycompany release2026-08-05Q4 and full-year results, end markets, balance sheet, cash flow and the Q1 fiscal 2027 guidebacksC02C03C05C16C43
- S03primarytranscript2026-08-05Sandisk fiscal Q4 2026 earnings call transcript (FMP), held 2026-08-05data/cache/transcripts/SNDK-2026Q4.jsonNBM customers, floors, guarantees and duration; market size; bit growth; capex; twelve analysts' questions on margins and durability
The total expected revenue from all our NBMs we have signed is a minimum of $93.9 billion, assuming floor pricing.
backsC03C07C13C15C17C18C29C34C37 - S04primarycompany filing2026-10-06ownership, the share count on 4 September, pay designbacksC18C26C36C37
- S05secondaryvendor data2026-10-20Research pack frozen beside this thesis, 2026-10-20content/SNDK/research-pack.jsonprice, consensus, earnings history, short interest, price-move splitbacksC20C21C24C25C39C41C43
- S06primaryindustry data2026-08-18NAND revenue and share by maker in calendar Q2 2026
relatively conservative bit shipment growth limited its revenue increase to 50.7% QoQ, reaching nearly $8.97 billion and trailing the growth of its peers.
backsC10 - S07primaryindustry data2026-07-21the 2026 deficit, the 2027 balance, China's share of bits, device forecasts, KV-cache compression
the supply-demand balance is expected to turn positive in the second half of 2027, allowing supply constraints to gradually ease.
backsC11C33C40 - S08primaryindustry data2026-09-30Q4 2026 contract price outlook and enterprise SSD demand
Bit demand for enterprise SSDs is projected to grow by more than 80% YoY in 2026 as CSPs continue to expand AI inference infrastructure.
backsC12 - S09secondarynews2026-08-1324/7 Wall St., 'Sandisk Unveils Multi-Year Financial Model and Growth Strategy, Memory Stocks Soar' 24/7 Wall St.the Investor Day long-term model; Business Wire's release refused the repo's fetcher and EDGAR carries no 8-K for it
including mid-to-high teens revenue growth, approximately 80% non-GAAP gross margins and approximately 50% adjusted free cash flow margins.
backsC14 - S10secondarynews2026-08-29The Motley Fool, 'Sandisk and Kioxia Plan to Invest More Than $31 Billion in Japanese Memory Plants' The Motley Foolthe joint investment plan and the venture extension
said Thursday that they plan to invest more than $31 billion in Japan through 2032.
backsC27 - S11secondarynews2026-08-14StorageReview, 'Sandisk tapes out its first HBF memory die, targets 2027 for inference product samples' StorageReviewthe HBF tapeout, roadmap and Sandisk's own performance claims
As of the August 2026 Investor Day, the first HBF inference product samples are listed as coming soon in 2027.
backsC28C29 - S12secondarynews2026-08-04the OCP specification, who joined, and the commercial timeline
Industry expectations put actual commercialization somewhere between late 2027 and 2028
backsC28 - S13primarycompany filing2025-10-03Micron Technology Form 10-K for the fiscal year ended 28 August 2025, Competitive Conditions Micron Technologya rival naming Sandisk among its competitorsbacksC30
- S14secondarynews2026-10-1224/7 Wall St., 'SanDisk Drops 6% in Memory Selloff on China's DRAM Progress; Seagate Slides 4%' 24/7 Wall St.the stated reason for the October fall
ChangXin Memory Technologies, a Chinese memory maker that isn't listed in the U.S., said it has made key progress on a denser memory-cell design.
backsC25 - S15tertiarynews2026-07-27the headline the July fall was reported under; the article itself is behind a paywall and only the headline is usedbacksC24
- S16primaryindustry data2025-10-14flash replacing hard drives in AI storage
As a result, the average price per GB has increased from US$0.012-0.013 to US$0.015-0.016, diminishing HDD’s main cost advantage.
backsC40 - S17secondaryindustry data2026-04-08the cost gap that limits flash replacing hard drives
by Q1 2026, 30 TB QLC SSD capacity cost 22.6 times as much as equivalent 30 TB HDD capacity.
read, but no claim rests on it - S19primarycompany filing2026-10-20who sold, how much and whether under a planbacksC26
- S20primarycompany filing2026-10-20fiscal 2023 to 2026 revenue, margins, costs and cash flowbacksC35