Sandisk Corporation
A fast thinking anchor, not a thesis, a valuation or a recommendation. Every figure is from the filings listed below.
| company | Sandisk Corporation (Nasdaq: SNDK) |
|---|---|
| sources used | FY2026 Form 10-K (filed 17 Aug 2026, fiscal year ended 3 Jul 2026); Form 10-Q for the quarter ended 3 Apr 2026; Form 8-K 25 Mar 2026 (Nanya investment and DRAM supply); Form 8-K 5 Aug 2026 ($14.0bn repurchase authorization); Form 8-K 14 May 2026; DEF 14A 7 Oct 2025. |
| latest period | Fiscal year ended 3 July 2026 (53 weeks). Sandisk has existed as a standalone company only since 21 Feb 2025, so there are two full reported years of independent history and one of carve-out accounting. |
| missing or stale data | No earnings-call transcripts and no investor-day materials are in sources/. The earnings-release 8-Ks are cover pages only — Exhibit 99.1 is not in the folder — so management guidance, non-GAAP bridges and any bit/ASP outlook are unavailable. No Form 4 insider data (none filed). Quarterly figures for Q4 FY2026 marked (derived) are computed as full-year 10-K less the nine-month 10-Q, not separately reported. |
Source documents used
| Primary | FY2026 Form 10-K, filed 17 Aug 2026 — Item 1 Business, Item 1A Risk Factors, Item 7 MD&A, Note 4 Revenue, Note 10 Related Parties and Related Commitments. |
|---|---|
| Interim | Form 10-Q for the quarter ended 3 Apr 2026 — quarterly income statement, end-market revenue, customer concentration, Note 4 Revenue. |
| Current reports | 8-K 25 Mar 2026 (Nanya private placement and DRAM supply arrangement); 8-K 5 Aug 2026 ($14.0bn repurchase authorization); 8-K 14 May 2026; 8-K 18 Nov 2025 (annual meeting results); 8-K 30 Dec 2025 (board appointment). |
| Governance | DEF 14A filed 7 Oct 2025. |
| Not available | Earnings-call transcripts; investor-day and capital-markets materials; earnings-release exhibits (Exhibit 99.1); Form 4 insider filings (none filed). |
1Business in one line
Sandisk sells NAND flash storage into three end markets — enterprise SSDs (Datacenter, $5,153m of FY2026 revenue), OEM and embedded client storage (Edge, $12,160m) and retail cards and drives (Consumer, $2,935m). It buys 100% of its wafers from Flash Ventures, three joint ventures with Kioxia in which it holds 49.9% and pays cost plus a small markup, adding chip design, controllers, firmware, assembly and test. It earns the spread between the price of a gigabyte and a wafer bill it must absorb whether or not it wants the output.
Type: cyclical, and not as a judgement call. Gross margin was 16.1%, 30.1% and 71.5% across FY2024–FY2026 on net payments to Flash Ventures of $3.4bn, $3.4bn and $3.6bn. The cost of supply barely moved across three years in which the company went from a $672m net loss to $11,433m of net income.
2 · 3Bull and bear
The simple reason this stock could work is that AI has made NAND genuinely scarce for the first time in a decade, and Sandisk spent that scarcity buying multi-year customer commitments — $59.8bn of remaining performance obligations backed by $5.0bn of collateral — rather than fabs it would have to write down when the cycle turns.
A1Industry bit supply keeps growing more slowly than AI-driven demand
What must happen. NAND bit supply keeps growing more slowly than AI-driven demand for several more years, so pricing holds far above the FY2024–FY2025 range.
Why this company. Sandisk could not flood the market if it wanted to: all wafers come from Flash Ventures' eight Japanese fabs, and while the ventures run it may not have flash fabricated by third parties or fabricate it itself — a restriction extended on 29 Jan 2026 to 31 Dec 2034.
Evidence. FY2026 exabytes rose a mid-teens percentage while revenue rose 175%. Revenue per gigabyte rose almost 150% in Datacenter, almost 180% in Edge, a low-fifties percentage in Consumer. Sales incentives fell to 11% of gross revenue from 19%.
What to monitor. Revenue per gigabyte by end market, and exabytes shipped (10-K MD&A).
A2New Business Models convert spot exposure into contracted volume with real customer skin in the game
What must happen. NBMs become the predominant way Sandisk sells, and customers honour them into a weaker market rather than walking.
Why this company. Holding the scarce good is the only moment a memory supplier can demand cash up front. A year ago refund liabilities were $126m and contract liabilities $25m; they are now $1,500m of security deposits and $1,242m of advances, plus $5.0bn of collateral at third-party banks callable on default.
Evidence. Remaining performance obligations were $59.8bn at 3 Jul 2026, entirely NBM-related and $58.7bn unbilled. NBMs began in fiscal 2026.
What to monitor. RPO, the share expected within twelve months (19%), and the collateral balance (Note 4).
A3The Datacenter mix shift is volume, not just price, and qualification makes it sticky
What must happen. Enterprise SSD stays a quarter or more of revenue through the next down-leg rather than proving to be a one-year allocation artefact.
Why this company. This is the one line where bits, not price, did most of the work, and enterprise SSDs are qualified into platforms over long cycles — slower to lose than a retail shelf.
Evidence. Datacenter revenue went $325m → $960m → $5,153m across FY2024–FY2026, from 4.9% of revenue to 25.4%, on exabytes up almost 120% against a mid-teens company-wide increase.
What to monitor. Datacenter exabytes versus revenue per gigabyte, separated.
A4Capital intensity at the Sandisk line stays trivial, so the peak converts into distributable cash
What must happen. Sandisk's own capex stays immaterial against cash generation, with wafer capital staying inside Flash Ventures and its lease facilities.
Why this company. In its reported accounts Sandisk is close to fabless: it does not consolidate Flash Ventures, having concluded that at 49.9% it lacks power to direct the activities that most affect the ventures' economics.
Evidence. FY2026 capex was $177m on $20,248m of revenue against $11,671m of operating cash flow. The $2.0bn term loan was settled on 4 Mar 2026, leaving no long-term debt, $4,762m of cash and an undrawn $1.5bn revolver.
What to monitor. Capex, net payments to Flash Ventures, and any funding call under the 49.9%–50.0% capital commitment.
Why the market might be missing it: the economics sit across two sets of books. Screens see an almost fabless company with no debt and $11.7bn of operating cash flow, while the fab capital, the leases and the obligation to fund half of Flash Ventures' fixed costs sit outside the consolidation — and the $59.8bn backlog, roughly three times trailing revenue, is disclosed in a revenue footnote.
The most likely way I lose money is paying a peak-cycle price for peak-cycle earnings made almost entirely of price, in a business whose cost base does not fall when price does — while management converts that peak cash into its own shares at the top.
B1The entire earnings step-up is price, and the cost base is contractually fixed against it
Attacks A1 — Industry bit supply keeps growing more slowly than AI-driven demand
How it could fail. Revenue tripled while payments to Flash Ventures went from $3.4bn to $3.6bn, so gross profit is geared almost one-for-one to selling price in both directions. Put FY2025 pricing onto FY2026 volumes and margin does not settle in the middle — it falls toward the 16.1%–30.1% band, because Sandisk owes the variable cost of its share of output on a rolling three-month forecast and half of Flash Ventures' fixed costs regardless of what it takes.
What would confirm failure. Revenue per gigabyte falling while exabytes hold, and underutilization charges returning to cost of revenue — $249m in FY2024 and $75m in FY2025 against $11m in FY2026. Watch inventory alongside: days in inventory rose to 178 from 135 on $2,698m of stock carried against an NRV estimated from today's prices.
What it does to the economics. Gross margin first, then the operating line: opex is a $2,083m base against $1,540m in FY2024, much of the increase performance-linked. Sandisk posted operating losses in FY2024 and FY2025 on a smaller cost base.
Permanent or fixable. The cycle is fixable and always has been. The price paid for peak earnings is not.
B2The asset-light appearance is an accounting artefact, and the obligation runs one way
Attacks A4 — Capital intensity at the Sandisk line stays trivial, so the peak converts into distributable cash
How it could fail. Non-consolidation removes the fabs from the balance sheet but not the commitments from the company. Sandisk must fund 49.9%–50.0% of each venture's capital investment when venture cash flow falls short, guarantees half of Flash Ventures' equipment leases, and cannot source or fabricate elsewhere until the ventures expire on 31 Dec 2034.
What would confirm failure. Flash Ventures commitments of $6,559m, $2,627m due in FY2027; $4,902m of further purchase obligations; $923m of lease guarantees; $1.2bn payable directly to Kioxia over 2026–2029; maximum estimable loss exposure $2,897m.
What it does to the economics. The $177m of FY2026 capex is a deliberately under-invested trough, not a structural feature. The 10-K says FY2027 capital investment rises for node transitions — into a year when the corporate alternative minimum tax begins to apply, against a FY2026 effective rate of 12%.
Permanent or fixable. Structural until end-2034. The risk that does not go away with the cycle.
B3Peak cash is being spent on peak-priced stock
Attacks A4 — Capital intensity at the Sandisk line stays trivial, so the peak converts into distributable cash
How it could fail. Sandisk repurchased 3 million shares for $4,537m in FY2026 — roughly $1,512 a share — and on 5 Aug 2026, the day it reported the year, authorised a further $14.0bn on top of April's $6.0bn, funded from operating cash flow that was $84m two years earlier.
What would confirm failure. Repurchases continuing at pace through a quarter in which revenue per gigabyte declines, or alongside disclosed funding calls and higher capex.
What it does to the economics. If FY2027 earnings normalise toward the FY2024–FY2025 range, $20.0bn of authorisation — $15.5bn unspent at year-end — will have been converted into shares priced off a 71.5% gross margin. No leverage argument remains: the term loan was retired in March.
Permanent or fixable. Permanent. The one place here where value is destroyed outright rather than swinging with the cycle.
B4The $59.8bn backlog is a price bet dressed as a floor
Attacks A2 — New Business Models convert spot exposure into contracted volume with real customer skin in the game
How it could fail. NBM pricing has fixed and variable components, and the variable ones were struck in the tightest NAND market in a decade. Only about 19% of the RPO — roughly $11.4bn — falls in the next twelve months against a Q4 FY2026 run-rate near $8,965m per quarter (derived), so most current revenue is still uncontracted spot. Collateral covers 8% of the RPO.
What would confirm failure. RPO falling without a matching transfer into revenue; the twelve-month share declining; collateral drawn. The 10-K states a customer breach may leave Sandisk unable to resell affected volumes 'at comparable prices, or at all.'
What it does to the economics. Revenue, then worse: volumes committed to a defaulting customer become excess inventory or underutilisation exactly when spot is weak. NBMs also 'constrain a portion of our available supply,' capping what a spike can earn.
Permanent or fixable. Fixable in principle — but a broken NBM destroys the argument that this cycle is different, which is the argument doing most of the work.
B5Datacenter 'volume growth' is largely reallocation of a fixed wafer supply, and it concentrates the customer base
Attacks A3 — The Datacenter mix shift is volume, not just price, and qualification makes it sticky
How it could fail. Total exabytes rose only a mid-teens percentage while Datacenter exabytes rose almost 120% and Consumer exabytes fell a mid-teens percentage — so much of the Datacenter volume story is the same constrained supply pointed at a better-paying customer, a trade available once. It also concentrates the book: top-10 customers were 46% of revenue in the quarter ended 3 Apr 2026 against 41% a year earlier, one above 10%.
What would confirm failure. Datacenter exabytes growing while total exabytes stay flat; Consumer revenue falling absolutely — roughly $556m in Q4 FY2026 against $820m in Q3 (derived); concentration rising further in the 10-Q.
What it does to the economics. A reallocated mix cannot repeat, so the FY2026 growth rate is not extrapolable — and it starves the retail brand and channel the 10-K names as a competitive strength while handing pricing power to a few hyperscalers as supply loosens.
Permanent or fixable. The mix shift is fixable; a vacated retail shelf and customers who have learned they are the marginal buyer are closer to permanent.
Each bull assumption with the bear risks that attack it. Open both sides and read across.
A1Industry bit supply keeps growing more slowly than AI-driven demand
What must happen. NAND bit supply keeps growing more slowly than AI-driven demand for several more years, so pricing holds far above the FY2024–FY2025 range.
Why this company. Sandisk could not flood the market if it wanted to: all wafers come from Flash Ventures' eight Japanese fabs, and while the ventures run it may not have flash fabricated by third parties or fabricate it itself — a restriction extended on 29 Jan 2026 to 31 Dec 2034.
Evidence. FY2026 exabytes rose a mid-teens percentage while revenue rose 175%. Revenue per gigabyte rose almost 150% in Datacenter, almost 180% in Edge, a low-fifties percentage in Consumer. Sales incentives fell to 11% of gross revenue from 19%.
What to monitor. Revenue per gigabyte by end market, and exabytes shipped (10-K MD&A).
B1The entire earnings step-up is price, and the cost base is contractually fixed against it
Attacks A1 — Industry bit supply keeps growing more slowly than AI-driven demand
How it could fail. Revenue tripled while payments to Flash Ventures went from $3.4bn to $3.6bn, so gross profit is geared almost one-for-one to selling price in both directions. Put FY2025 pricing onto FY2026 volumes and margin does not settle in the middle — it falls toward the 16.1%–30.1% band, because Sandisk owes the variable cost of its share of output on a rolling three-month forecast and half of Flash Ventures' fixed costs regardless of what it takes.
What would confirm failure. Revenue per gigabyte falling while exabytes hold, and underutilization charges returning to cost of revenue — $249m in FY2024 and $75m in FY2025 against $11m in FY2026. Watch inventory alongside: days in inventory rose to 178 from 135 on $2,698m of stock carried against an NRV estimated from today's prices.
What it does to the economics. Gross margin first, then the operating line: opex is a $2,083m base against $1,540m in FY2024, much of the increase performance-linked. Sandisk posted operating losses in FY2024 and FY2025 on a smaller cost base.
Permanent or fixable. The cycle is fixable and always has been. The price paid for peak earnings is not.
A2New Business Models convert spot exposure into contracted volume with real customer skin in the game
What must happen. NBMs become the predominant way Sandisk sells, and customers honour them into a weaker market rather than walking.
Why this company. Holding the scarce good is the only moment a memory supplier can demand cash up front. A year ago refund liabilities were $126m and contract liabilities $25m; they are now $1,500m of security deposits and $1,242m of advances, plus $5.0bn of collateral at third-party banks callable on default.
Evidence. Remaining performance obligations were $59.8bn at 3 Jul 2026, entirely NBM-related and $58.7bn unbilled. NBMs began in fiscal 2026.
What to monitor. RPO, the share expected within twelve months (19%), and the collateral balance (Note 4).
B4The $59.8bn backlog is a price bet dressed as a floor
Attacks A2 — New Business Models convert spot exposure into contracted volume with real customer skin in the game
How it could fail. NBM pricing has fixed and variable components, and the variable ones were struck in the tightest NAND market in a decade. Only about 19% of the RPO — roughly $11.4bn — falls in the next twelve months against a Q4 FY2026 run-rate near $8,965m per quarter (derived), so most current revenue is still uncontracted spot. Collateral covers 8% of the RPO.
What would confirm failure. RPO falling without a matching transfer into revenue; the twelve-month share declining; collateral drawn. The 10-K states a customer breach may leave Sandisk unable to resell affected volumes 'at comparable prices, or at all.'
What it does to the economics. Revenue, then worse: volumes committed to a defaulting customer become excess inventory or underutilisation exactly when spot is weak. NBMs also 'constrain a portion of our available supply,' capping what a spike can earn.
Permanent or fixable. Fixable in principle — but a broken NBM destroys the argument that this cycle is different, which is the argument doing most of the work.
A3The Datacenter mix shift is volume, not just price, and qualification makes it sticky
What must happen. Enterprise SSD stays a quarter or more of revenue through the next down-leg rather than proving to be a one-year allocation artefact.
Why this company. This is the one line where bits, not price, did most of the work, and enterprise SSDs are qualified into platforms over long cycles — slower to lose than a retail shelf.
Evidence. Datacenter revenue went $325m → $960m → $5,153m across FY2024–FY2026, from 4.9% of revenue to 25.4%, on exabytes up almost 120% against a mid-teens company-wide increase.
What to monitor. Datacenter exabytes versus revenue per gigabyte, separated.
B5Datacenter 'volume growth' is largely reallocation of a fixed wafer supply, and it concentrates the customer base
Attacks A3 — The Datacenter mix shift is volume, not just price, and qualification makes it sticky
How it could fail. Total exabytes rose only a mid-teens percentage while Datacenter exabytes rose almost 120% and Consumer exabytes fell a mid-teens percentage — so much of the Datacenter volume story is the same constrained supply pointed at a better-paying customer, a trade available once. It also concentrates the book: top-10 customers were 46% of revenue in the quarter ended 3 Apr 2026 against 41% a year earlier, one above 10%.
What would confirm failure. Datacenter exabytes growing while total exabytes stay flat; Consumer revenue falling absolutely — roughly $556m in Q4 FY2026 against $820m in Q3 (derived); concentration rising further in the 10-Q.
What it does to the economics. A reallocated mix cannot repeat, so the FY2026 growth rate is not extrapolable — and it starves the retail brand and channel the 10-K names as a competitive strength while handing pricing power to a few hyperscalers as supply loosens.
Permanent or fixable. The mix shift is fixable; a vacated retail shelf and customers who have learned they are the marginal buyer are closer to permanent.
A4Capital intensity at the Sandisk line stays trivial, so the peak converts into distributable cash
What must happen. Sandisk's own capex stays immaterial against cash generation, with wafer capital staying inside Flash Ventures and its lease facilities.
Why this company. In its reported accounts Sandisk is close to fabless: it does not consolidate Flash Ventures, having concluded that at 49.9% it lacks power to direct the activities that most affect the ventures' economics.
Evidence. FY2026 capex was $177m on $20,248m of revenue against $11,671m of operating cash flow. The $2.0bn term loan was settled on 4 Mar 2026, leaving no long-term debt, $4,762m of cash and an undrawn $1.5bn revolver.
What to monitor. Capex, net payments to Flash Ventures, and any funding call under the 49.9%–50.0% capital commitment.
B2The asset-light appearance is an accounting artefact, and the obligation runs one way
Attacks A4 — Capital intensity at the Sandisk line stays trivial, so the peak converts into distributable cash
How it could fail. Non-consolidation removes the fabs from the balance sheet but not the commitments from the company. Sandisk must fund 49.9%–50.0% of each venture's capital investment when venture cash flow falls short, guarantees half of Flash Ventures' equipment leases, and cannot source or fabricate elsewhere until the ventures expire on 31 Dec 2034.
What would confirm failure. Flash Ventures commitments of $6,559m, $2,627m due in FY2027; $4,902m of further purchase obligations; $923m of lease guarantees; $1.2bn payable directly to Kioxia over 2026–2029; maximum estimable loss exposure $2,897m.
What it does to the economics. The $177m of FY2026 capex is a deliberately under-invested trough, not a structural feature. The 10-K says FY2027 capital investment rises for node transitions — into a year when the corporate alternative minimum tax begins to apply, against a FY2026 effective rate of 12%.
Permanent or fixable. Structural until end-2034. The risk that does not go away with the cycle.
B3Peak cash is being spent on peak-priced stock
Attacks A4 — Capital intensity at the Sandisk line stays trivial, so the peak converts into distributable cash
How it could fail. Sandisk repurchased 3 million shares for $4,537m in FY2026 — roughly $1,512 a share — and on 5 Aug 2026, the day it reported the year, authorised a further $14.0bn on top of April's $6.0bn, funded from operating cash flow that was $84m two years earlier.
What would confirm failure. Repurchases continuing at pace through a quarter in which revenue per gigabyte declines, or alongside disclosed funding calls and higher capex.
What it does to the economics. If FY2027 earnings normalise toward the FY2024–FY2025 range, $20.0bn of authorisation — $15.5bn unspent at year-end — will have been converted into shares priced off a 71.5% gross margin. No leverage argument remains: the term loan was retired in March.
Permanent or fixable. Permanent. The one place here where value is destroyed outright rather than swinging with the cycle.
4Signals to monitor
Each signal traces to a named assumption or risk. Hover a tag for its title.
Revenue per gigabyte, by end market
Exabytes shipped, total and Datacenter
Net payments to Flash Ventures vs revenue
Underutilization charges in cost of revenue
Remaining performance obligations and 12-month share
Customer cash held: deposits, advances, collateral
Days in inventory
Repurchases vs Flash Ventures commitments and capex
5External challenge notes
6Bottom line
- The stock could work because AI has made NAND scarce against a supply base that cannot respond quickly, and Sandisk converted that scarcity into $59.8bn of contracted obligations and $11.7bn of operating cash flow without owning the fabs.
- For that to hold, industry bit supply must keep lagging demand long enough for the NBM book to become the predominant way Sandisk sells, and Datacenter volume — not just Datacenter price — must keep compounding.
- It most likely breaks on the fact that the entire step-up was price against a wafer bill that moved only from $3.4bn to $3.6bn, so normalisation takes gross margin back toward the 16%–30% band while Sandisk still owes half of Flash Ventures' fixed costs and cannot source elsewhere before 2035 — with $20.0bn of buyback authorised at roughly $1,512 a share.
- I would change my mind after two or three quarters in which Datacenter exabytes keep growing while revenue per gigabyte falls and gross margin holds well above 40%, with RPO and collateral rising rather than draining — which would mean the contracted book, not the spot price, is setting the earnings.