⚠ EARNINGS ARE TOMORROW — READ THIS FIRST

Sandisk reports fiscal fourth-quarter and full-year FY2026 results on 2026-08-05 — ONE DAY after this dive was struck.

• Consensus for the print: revenue $8,439,742,000 and EPS $34.80 (earn_cal, last updated 2026-08-04; epsActual and revenueActual are both null).
• What that bar requires: +41.8% sequential revenue growth on the April quarter's $5.950B, and +48.7% sequential EPS growth on $23.41.
• The fiscal year being reported ended 2026-07-03 and is a 53-week year (10-Q, 2026-05-01). No filing in this archive covers it. The most recent filing-verified financials in this dive are as of 2026-04-03.
• The vendor's annual statements are ~13 months old. inc_a has only four rows and the newest is the year ended 2025-06-27. The entire FY2026 transformation is visible only through inc_q and the two 10-Qs.

This is not a "buy today" document and it will not pretend to be one. Writing a purchase recommendation into a print nobody in this room has seen, on a stock that rose 10.84% today and 3,354% in twelve months, is not defensible. What follows is a full valuation, an explicit list of the three numbers we would need to see in tomorrow's release, and a pre-registered set of prices and conditions at which the answer changes. The verdict is Watch, and the earnings date is an explicit input to it, not an afterthought.

Reference table

1. What the business is, and how the disclosure actually works

The 10-K describes Sandisk as "a leading developer, manufacturer and provider of data storage devices and solutions based on NAND flash technology" whose portfolio "delivers powerful flash storage solutions for AI workloads in datacenters, edge devices, and consumers." Chief executive David V. Goeckeler; approximately 11,000 employees across 33 countries — 73% in Asia Pacific, 19% in the Americas, 8% in Europe, the Middle East and Africa; headquarters at 951 Sandisk Drive, Milpitas, California; approximately 7,900 granted patents and 3,200 pending applications.

The separation — and why the history in this file is not what it looks like

This is the single most important structural fact about the data and it must be stated before any number is used.

Per the 10-Q filed 2026-05-01: Western Digital's board authorised the separation on 2023-10-30; it completed on 2025-02-21, when WDC distributed 116,035,464 shares — 80.1% of Sandisk — to its holders at a ratio of one-third of a Sandisk share per WDC share. Sandisk began trading on Nasdaq under "SNDK" on 2025-02-24. WDC retained 28,827,787 shares (19.9%), disposed of 21,314,768 (14.6%) on 2025-06-09 and a further 5,821,135 on 2026-02-18 through debt-for-equity exchanges, and as of 2026-03-19 retains only 1,691,884 shares, no longer restricted. The overhang is effectively gone.

Consequence for the statements. The 10-Q states directly: "Prior to the separation, the Company's historical consolidated financial information was derived from WDC's consolidated financial statements and accounting records and prepared as if the Company existed on a standalone basis." Fiscal years 2022, 2023, 2024 and the first two-thirds of fiscal 2025 are therefore carve-out allocations, not the reported results of this legal entity. The FY2022 figures — $9.754B of revenue and $1.064B of net income — are a construct describing a division inside another company. We use them for cyclical shape and explicitly not as a track record.

Consequence for the price history. The tech block reports a 52-week low of $40.69 and a twelve-month return of +3,354%. Both are computed from real post-spin closes: 2025-08-04, twelve months back, falls roughly five months after the shares began trading. There is no pre-2025-02-24 price history for this security and none is used. Anything a screen shows before that date belongs to the former SanDisk Corporation, a different company acquired by Western Digital in 2016 — and, as Section 8 documents, the vendor payload actually contains that other company's data.

What is sold, to whom

The FY2025 10-K reports three end markets — Cloud, Client and Consumer. The 10-Q renamed them: "multiple end markets of 'Datacenter' (formerly referred to as 'Cloud'), 'Edge' (formerly referred to as 'Client'), and 'Consumer'." The vendor's seg_prod block still uses the old names. We report both and use the new ones.

Reconciliation, performed and passed: the three lines sum to $7,355M against reported FY2025 revenue of $7,355M — an exact match. No double-counting, no missing line. But there is only ONE year of it, and it is FY2025. There is no FY2024 or FY2023 comparison and no FY2026 mix at all, which matters enormously because the FY2026 growth is wildly uneven across the three.

What the 10-Q tells us about FY2026 mix that the vendor cannot. For the three months to 2026-04-03 versus the prior year:

Read that table twice. The only place Sandisk is selling more physical memory is the datacentre. Everywhere else it is selling less memory at dramatically higher prices — consumers bought 40% fewer bits. The mix shift toward AI infrastructure is genuine and it is the strongest fact in the bull case. The demand destruction in Consumer and Edge is genuine and it is the strongest fact in the bear case. Both are in the same disclosure.

Geography (seg_geo, FY2025, sums verified). The six lines total $7,355M — an exact match to reported revenue.

Critical caveat, from the 10-K: "Net revenue is attributed to geographic regions based on the ship-to location of the customer." Combined China and Hong Kong is 45.4% of revenue by ship-to location — but Sandisk sells to contract manufacturers and distributors, so this overstates Chinese end demand and understates US end demand. It is a manufacturing-location map, not a demand map. We use it as such and draw no export-control conclusion from it that the filing does not support.

Customer concentration is unusually LOW for a semiconductor company, and this is a genuine positive. From the 10-K: "For 2025 and 2024 no single customer accounted for more than 10% of the Company's net revenue. For 2023, one customer accounted for 15%." Top ten customers were 40%, 41% and 47% of revenue in FY2025, FY2024 and FY2023. There is no single point of customer failure here, which distinguishes Sandisk sharply from the wafer-fab-equipment names.

Flash Ventures — the most important thing on the page and it is not on the balance sheet

Sandisk does not own the factories that make its product. From the 10-Q: "The Company procures substantially all of its flash-based memory wafers from its business ventures with Kioxia Corporation, which consists of three separate legal entities: Flash Partners Ltd., Flash Alliance Ltd. and Flash Forward Ltd., collectively referred to as 'Flash Ventures.'"

The terms, verbatim from the 10-K and 10-Q:

• Kioxia owns the facilities and provides wafer manufacturing services to Flash Ventures at cost, using equipment owned or leased by Flash Ventures.
• Flash Ventures accounts for approximately 80% of the total manufacturing capacity in the facilities owned by Kioxia.
• Sandisk and Kioxia each take approximately 50% of output, priced at cost plus a small markup.
• "In addition, we are obligated to pay for half of Flash Ventures' fixed costs regardless of the output we choose to purchase."
• Sandisk is "committed to fund 49.9% to 50.0% of each Flash Ventures entity's capital investments to the extent that the entity's operating cash flow is insufficient."
• Sandisk guarantees half of the outstanding obligations under Flash Ventures' Japanese equipment lease facilities, and a cancellation event "could result in an acceleration of the lease obligations and a call on our guarantees."
• The 49.9% interest in Flash Ventures' earnings is recognised one quarter in arrears, in Other expense, net.

On 2026-01-29 the structure was extended. Flash Alliance and Flash Partners were both extended from 2029-12-31 to 2034-12-31, so all three entities now co-terminate on 2034-12-31. In consideration, Sandisk agreed to pay Kioxia $1.2 billion directly across 2026 through 2029 "in consideration of Kioxia's manufacturing services and the continued availability of supply."

Why this matters more than any ratio in this file. It explains the entire financial shape of the company: property, plant and equipment was only $619M at FY2025 and quarterly depreciation and amortisation runs $36-38M against $5.95B of quarterly revenue. Sandisk looks asset-light and generates a spectacular free-cash-flow conversion — because the capital intensity of NAND manufacturing sits in an unconsolidated joint venture it does not control. It also means the operating leverage runs violently in both directions: when Sandisk cuts purchases, it still pays half the fixed costs. The 10-Q discloses exactly that happening — "During the nine months ended April 3, 2026, the Company temporarily reduced its utilization of its share of Flash Ventures' manufacturing capacity" — at a cost of $11 million charged to cost of revenue, against $24 million in the prior-year period and $75 million of underutilisation charges disclosed in the FY2025 10-K.

Competitors, per the 10-K: "We compete with vertically integrated suppliers such as Kioxia, Micron Technology, Inc., Samsung Electronics Co., Ltd., SK Hynix, Inc., Yangtze Memory Technologies Co., Ltd." Note that Kioxia is listed as a competitor and is simultaneously the manufacturing partner that takes the other half of the same wafers. That is not a contradiction the filing resolves, and it is worth holding in mind.

2. The one thing to understand about these numbers: it is price, not volume

Everything in this dive reduces to a single disclosed sentence, and we quote it exactly:

> "Net revenue increased 251% in the three months ended April 3, 2026 from the comparable period in the prior year, primarily due to a 248% increase in average selling prices ("ASP") per gigabyte. The exabytes sold remained flat from the comparable period in the prior year."

> — Sandisk Corporation, Form 10-Q filed 2026-05-01

And the margin consequence, from the same document:

> "Gross margin increased by 5,600 basis points and 3,000 basis points, respectively, for the three and nine months ended April 3, 2026... The higher ASP was primarily driven by favorable pricing conditions in the industry. As a result, for the three and nine months ended April 3, 2026, the increase in ASP has outpaced the movement in costs per gigabyte."

We verified the 5,600 basis points independently from the vendor's quarterly statements: gross margin was 382/1,695 = 22.5% in the March-2025 quarter and 4,662/5,950 = 78.4% in the April-2026 quarter. The difference is 55.9 percentage points, which is the filing's 5,600 basis points to within rounding. Vendor and filing agree exactly.

Now look at the cost line, because it is the part almost nobody checks. Cost of revenue went from $1,484M in the January-2026 quarter to $1,288M in the April-2026 quarter — it FELL 13.2% while revenue rose 96.7%. That is not a mistake. It is what happens when bit volume is flat-to-down, the joint venture's utilisation was deliberately reduced, and per-gigabyte cost fell with the node transition, all while the selling price tripled. A commodity producer whose costs fall while its price triples earns extraordinary money, and it earns it for exactly as long as the price holds.

Here is the same company four years earlier, from the annual series in this file:

Cumulative net income, FY2022 through FY2025: MINUS $3.392 billion. Revenue fell 37.6% in a single year in FY2023 and gross margin collapsed to 7.1%. This is the same business, the same joint venture, the same product — running the other way.

Three things follow, and they are the argument of the whole dive.

First, the current earnings are real. We tested the April quarter for every one-off in the defect taxonomy and found none. Section 8 sets out the tests. The only discrete item is a $46M loss on debt extinguishment from repaying the term loan early, which reduces reported earnings — so the clean number is better than the printed one.

Second, the current earnings are not durable in any sense the word normally carries. They exist because a commodity price rose 248% against flat volume. Nothing about Sandisk's cost structure, competitive position or intellectual property changed between March 2025 and April 2026 to produce a 5,600 basis point margin expansion. The price of NAND changed.

Third, the market already knows this, which is why the multiple is 6.93x. A 6.93x forward P/E on a company with net cash is not a mispricing waiting to be discovered; it is the market's standing discount for peak-cycle commodity earnings. The bull case is not "it's cheap." The bull case is that this cycle does not end the way every previous one did — because AI infrastructure has structurally changed NAND demand. That is a real argument, several credible voices in our knowledge base make it, and it is unresolved. What it is not is a valuation argument.

3. Growth and margin trajectory

Quarterly, from the vendor income series, cross-checked line by line against earn_cal and the 10-Q MD&A:

Two mechanical notes before anyone reads a trend into this. First, Q1 FY2026 was a 14-week quarter — the 10-Q states fiscal 2026 comprises 53 weeks and ends 2026-07-03, with the first quarter running 14 weeks. Its sequential growth is flattered by roughly a week of trading. Second, the year-on-year comparisons for Q1 and Q2 FY2026 are against pre-separation carve-out quarters and are therefore not perfectly like-for-like at the expense lines.

The beat pattern is extreme and it matters for tomorrow. Sandisk has beaten the EPS consensus in all five prints in this file, and the recent beats are enormous: +522% (Aug 2025), +38% (Nov 2025), +71% (Jan 2026), +60% (Apr 2026). Revenue beats: +4.2%, +7.2%, +12.5%, +26.0% — accelerating. This is not analyst skill; it is analysts structurally unable to model a commodity price that is moving faster than their revision cycle. It cuts both ways. On the way down, the same modelling lag produces the same magnitude of miss.

Margin decomposition — recomputed by us from raw revenue and cost lines, not taken from vendor ratio fields (defect class 4 protocol):

Operating expenses are barely moving. R&D went from $285M to $337M — up 18.2% — while revenue rose 251%. SG&A went from $139M to $161M. Total operating expenses were 29.4% of revenue in the March-2025 quarter and 9.3% in the April-2026 quarter. That is the operating leverage, and it is the mirror image of the FY2023 disaster when the same fixed cost base sat on top of a 37.6% revenue decline.

The clean operating number for the April quarter. Reported operating income was $4,111M. Add back the $46M loss on debt extinguishment — a genuinely discrete item disclosed in the 10-Q as arising from "the write-off of the remaining unamortized issuance costs" on early repayment of the term loan — and clean operating income is $4,157M, a 69.9% operating margin. Tax-effected at the disclosed 12.0% rate, clean net income is approximately $3,655M, or $23.28 per diluted share on 157.0M diluted shares. The company's reported non-GAAP figure was $23.41. These agree to within 0.6%. The $23.41 is clean operating earnings. It is not a tax release, not a spin gain, not an inventory revaluation. That was the highest-value single check in this dive and it came back clean.

Research and development, in context. $337M in the April quarter, 5.7% of revenue — down from 16.8% a year earlier purely because the denominator exploded. In absolute terms the 10-Q attributes the $52M year-on-year increase to "a $42 million increase in compensation and benefits due to variable compensation associated with company performance." A material share of the R&D increase is bonus accrual on the profit spike, not incremental engineering. That is honest disclosure and it is worth knowing: R&D is not scaling with the opportunity.

4. Balance sheet and the enterprise-value rebuild

This section carries the largest vendor-versus-filing divergence in the dive, and the filing wins decisively.

What the vendor's bal_a says (as at 2025-06-27 — THIRTEEN MONTHS OLD):

Every number in that table is obsolete, and one of them describes an instrument that no longer exists.

The 10-Q filed 2026-05-01 settles it. At separation on 2025-02-21 Sandisk entered a $2.0B seven-year Term Loan B and a $1.5B five-year revolving credit facility, drew the full $2.0B term loan, and used part of it to make a $1.5B net distribution payment to WDC. Then:

> "On March 4, 2026, the Company settled in full the remaining outstanding principal amounts of the Term Loan Facility, plus accrued interest, using cash on hand. In connection with the early settlement of the Term Loan Facility, the Company recognized a loss on debt extinguishment of $46 million resulting from the write-off of the remaining unamortized issuance costs."

> "As of April 3, 2026, the Company has drawn no amounts under the Revolving Credit Facility."

> "As of April 3, 2026, the Company had no outstanding standby letters of credit, and the available capacity under the Revolving Credit Facility was $1.5 billion."

The vendor's balance sheet reports $1.829B of long-term debt and $561M of net debt. The 10-Q shows the term loan repaid in full and the revolver undrawn. We use the 10-Q.

We also searched both 10-Qs and all three 8-Ks specifically for convertible, capped call, senior notes, aggregate principal, term loan and revolving credit. There are no convertible notes, no capped call transactions and no senior unsecured notes anywhere in this archive. The financing risk the brief flagged as most likely does not exist here — and stating that as a verified negative is a finding. The financings that DO post-date every statement in the file are of a different kind, and there are three:

1. 2026-03-04 — the $2.0B term loan repaid in full, with a $46M extinguishment charge (10-Q).

2. 2026-03-25 / 2026-04-08 — a $972M equity investment in Nanya Technology Corporation. The 8-K of 2026-03-25 records the subscription agreement for approximately 139 million Nanya shares, roughly 3.9% of Nanya fully diluted, at a 15% discount to Nanya's 30-day average price, subject to a statutory three-year lock-up. The 10-Q's subsequent-events note records that on 2026-04-08 Sandisk remitted $972 million and received 138,685,000 shares on 2026-04-28. Concurrently the parties entered a multi-year strategic supply arrangement under which Nanya will supply Sandisk with DRAM products. This is a NAND company buying its way into DRAM supply, and the cash left the building five days after the last balance-sheet date in this file.

3. 2026-04-30 — a $6 billion share repurchase authorisation (8-K and 10-Q subsequent events), "expected to be funded by operating cash flows", with the standard non-obligation language.

The corrected position, as at 2026-04-03, rebuilt from km_ttm and ratios_ttm per-share fields and cross-checked against the 10-Q:

> Market cap $211.416B (148,089,758 shares × $1,427.62)

> + total debt ~$0.207B (finance leases; the term loan is gone)

> − cash and equivalents ~$3.735B

> = enterprise value $207.888B

Defect class 2 — tested and CLEAN on km_ttm. This is worth saying plainly because it is unusual: the vendor's trailing enterprise value for Sandisk is correct to the dollar. shortTermInvestments is genuinely $0, so the omission that broke META, AAPL and BRK-B cannot occur. The staleness sits in bal_a, not in km_ttm — using bal_a's $561M of net debt would have produced an enterprise value of $211.977B, an error of $4.089B (2.0%), and would have described Sandisk as a levered company when it is a net-cash one.

Two adjustments the vendor cannot make, which we make openly.

Adjustment A — the Nanya payment. $972M of cash left on 2026-04-08, five days after the balance-sheet date. Pro-forma net cash is ~$2.556B and pro-forma enterprise value is $208.860B. The offsetting asset is a Nanya equity stake locked up by Taiwanese statute for three years — a real asset, but not one that can be sold to fund anything. We report the multiples on the $207.888B EV and disclose the $972M as a known, dated, post-balance-sheet outflow.

Adjustment B — the diluted share count. Market capitalisation uses 148,089,758 shares outstanding, which matches the 10-Q cover page exactly. But the April quarter's EPS denominator was 157,000,000 diluted shares — 6.0% higher, and it has risen from 145M to 149M to 156M to 157M across four quarters, an 8.3% increase in three quarters. On a fully diluted basis market capitalisation would be $224.14B and enterprise value $220.61B, lifting EV/EBITDA to 39.79x. Neither figure is wrong; they answer different questions. We use the outstanding-share market cap as the anchor and disclose the diluted alternative rather than burying it. The $6B buyback authorisation — 2.84% of market capitalisation — would need to be roughly a third executed simply to offset the dilution already run.

What the balance sheet actually says. Net cash of $3.528B against $5.544B of trailing EBITDA, no funded debt, an undrawn $1.5B revolver to 2030, and disclosed compliance with the single leverage covenant. Interest coverage is 43.5x. Current ratio 4.78x. There is no solvency question here and the risk score of 9 has nothing to do with the balance sheet.

Two items that are genuine exposures and appear in no vendor field at all (defect class 10):

• The Flash Ventures lease guarantees. Sandisk "guarantee[s] half of all of the outstanding obligations under each lease agreement" on equipment Flash Ventures sells to and leases back from a consortium of Japanese financial institutions. The dollar amount sits in a table the extraction stripped, so we state the obligation and explicitly do not estimate its size.
• The Kioxia collaboration payments and building-depreciation prepayments. $1.2 billion payable directly to Kioxia across 2026-2029. Separately, the January 10-Q discloses $750 million recorded within Other non-current assets as building-depreciation prepayments to be credited against future charges, plus $387 million of additional committed prepayments and approximately $155 million of further building depreciation payments. These are real cash requirements attached to a manufacturing base Sandisk does not consolidate.

Goodwill is $4.999B against $12.985B of total assets at FY2025 — 38.5% of the balance sheet — after the $1.8B impairment. intangiblesToTotalAssetsTTM is 29.2%. The impairment trigger disclosed in the 10-K is worth quoting because it is a warning label: the company tested goodwill after identifying indicators "related to macroeconomic indicators, industry developments, the trading price of the Company's common stock and resulting market capitalization." Sandisk has already once written down $1.8B of goodwill because its own share price fell. At $1,427.62 that is not a live risk. It tells you what happens if the price goes back.

5. Cash flow and capital returns

Three consecutive years of negative free cash flow, FY2023 through FY2025, totalling roughly $1.527B of cash burn. Then a trailing-twelve-month swing to +$4.46B. That is a $5B swing in operating cash generation driven by one commodity price. No further comment is required on the cyclicality question.

Capex of ~$179M on ~$13.2B of trailing revenue is 1.36% of sales (capexToRevenueTTM 0.01358). Do not read that as capital efficiency. The 10-Q's investing-activities note for the nine months to 2026-04-03 lists "$165 million in net issuances from activity related to Flash Ventures and $134 million in capital expenditures." Sandisk's real capital intensity is funded through loans and equity contributions to an unconsolidated joint venture and through building-depreciation prepayments, none of which appear in the capex line. The company itself signals the direction: "For fiscal year 2026, we increased our capital investments as we transition to newer nodes to meet the demand and technology needs of our product portfolio." The reported free-cash-flow yield of 2.11% is therefore flattering, and we say so rather than quoting it as a virtue.

Working capital. Days sales outstanding 79.2, days inventory outstanding 140.9, days payables outstanding 53.6, cash conversion cycle 166.5 days. The 10-Q explains the moving parts: DSO fell 11 days "primarily due to higher revenue during the period and continued strong receivables collections"; DIO rose 8 days "primarily due to inventory builds to meet demand." Building inventory into a price spike is the correct operational decision and the classic cyclical trap simultaneously — it is the same inventory that produced the write-downs of the FY2023 downcycle.

A genuinely bullish disclosure that deserves its own line. From the 10-Q:

> "In connection with entering into long-term agreements during the third quarter of fiscal 2026 with certain customers, we received customer advances that increased our available cash and cash equivalents... As of April 3, 2026, our contract liabilities were $511 million, which primarily relates to the remaining performance obligations under these long-term agreements."

Customers are pre-paying Sandisk for future supply. $511M of contract liabilities is 3.9% of trailing revenue. That is the single strongest piece of filing evidence that the shortage is a real physical supply constraint rather than a speculative price event, and it independently corroborates the knowledge base's "five-year contracts" claim (Santi, 2026-05-14). It is not proof the price holds — a locked-in contract at today's price is exactly what a customer signs at the top — but it is hard evidence and it belongs in the bull column.

Capital returns: there are none yet. lastDividend is 0, dividendYieldTTM is 0, dividendPayoutRatioTTM is 0. Sandisk pays no dividend and has never paid one. The only capital-return instrument is the $6 billion repurchase authorisation approved 2026-04-30 — 2.84% of market capitalisation — of which the file contains no evidence that a single share has been bought. The 8-K is explicit: "The approval of the Repurchase Program does not obligate the Company to repurchase any common shares." Shareholder yield in this dive is therefore 0.00%, and every dollar of expected return must come from earnings and the multiple.

6. Valuation — priced in or room?

At $1,427.62 (market cap $211.416B, 148,089,758 shares outstanding, rebuilt enterprise value $207.888B):

Estimate coverage, stated before any conclusion is drawn from it (defect class 9). FY2026: 15 revenue analysts, 11 EPS analysts. FY2027: 15 and 12. FY2028: 12 revenue analysts but only 5 on EPS — thin, and the EPS high/low spread of $490.46 to $160.39 is a 3.06x range, which is not a consensus in any meaningful sense. FY2029 rests on 2 analysts and FY2030 on ONE. We draw no conclusion whatsoever from FY2029 or FY2030, and we treat FY2028 EPS as directional only.

The shape of the estimate curve is itself the most interesting thing in the block. Revenue: $20.0B → $48.7B → $56.9B → $52.1B → $11.6B. The two analysts who model FY2029 have revenue declining 8.4%, and the single analyst who models FY2030 has it collapsing 77.8% to $11.566B — below FY2027 by a factor of four. The long-dated consensus, thin as it is, models a bust. We do not use those rows for valuation, but the fact that the only analysts willing to look past FY2028 model a cycle roll-over is a legitimate observation about how the street itself frames this name.

Sanity check on FY2026, because it is being reported tomorrow and we can test it. Three quarters of actual non-GAAP EPS are $1.22 + $6.20 + $23.41 = $30.83; adding the Q4 consensus of $34.80 gives $65.63, against the est row's $66.874 — a 1.9% difference. Revenue: $2.308B + $3.025B + $5.950B = $11.283B plus the Q4 estimate of $8.440B = $19.723B, against revenueAvg of $20.003B and revenueLow of $19.727B. The estimate block reconciles to the reported quarters. epsAvg is trustworthy here; we verified it rather than assuming it.

Basis note. epsAvg is struck on a non-GAAP basis — confirmed by comparing the est rows to earn_cal actuals, which are themselves non-GAAP ($23.41 versus GAAP $23.03). All forward multiples in this dive are non-GAAP. The trailing 49.62x is GAAP diluted. The gap between the two bases is small for Sandisk — 1.6% in the April quarter — which is unusual and creditable.

Peer context — and there effectively is none. The vendor peer set is Credo, Flex, HPE, Jabil, Keysight, MongoDB, Super Micro, Teledyne, Teradyne and Wipro. Not one of these is a memory manufacturer. MongoDB is a database company; Wipro is an Indian IT services firm; Keysight and Teradyne are test-and-measurement. The peer set is unusable and we discard it in full. The genuine comparables — Micron, SK Hynix, Samsung, Kioxia, Western Digital — are absent, and no peer multiples are supplied in the file for anyone, so no peer-multiple comparison is drawn anywhere in this dive. The only cross-name data point available is the batch context: Micron rose 7.6% today and Marvell 12.8%, against Sandisk's 10.84%.

6a. What today's price assumes (the inversion)

At $1,427.62 — 21.35x FY2026E and 6.93x FY2027E non-GAAP consensus — the price embeds the following falsifiable claims. Each carries a number and a date, and each is labelled by source.

• That FY2026 lands at roughly $20.0B of revenue and $66.87 of EPS, which requires tomorrow's print to deliver $8.440B and $34.80. (Consensus-derived; 15 and 11 analysts.) Falsifier date: 2026-08-05. This is the only assumption in the list that resolves within twenty-four hours. It requires 41.8% sequential revenue growth on a quarter that already grew 96.7% sequentially.
• That NAND average selling price per gigabyte does not fall through fiscal 2027 and 2028. (Our number, derived from the 10-Q's stated 248% ASP increase against flat exabytes.) This is the most fragile assumption in the price and it is not close. Consensus revenue of $48.670B in FY2027 against the $13.184B trailing and $20.003B FY2026E requires the price spike not merely to hold but to extend across a full further year. The company's own framing — "we expect these conditions to persist through calendar year 2026 and beyond" — is management voice and is half-weighted here per house rule.
• That net margin reaches 65.6% in FY2027 and 60.8% in FY2028. (Derived by us from revenueAvg and netIncomeAvg, which cross-check against epsAvg × diluted shares.) The April quarter actually delivered 60.8%, so this is not arithmetically impossible — it is an extrapolation of a single quarter across two years. For reference, this company's net margin was −22.3% in FY2025 and −35.2% in FY2023.
• That the market keeps paying at least ~7x forward earnings at every date between now and then. (Our number.) At 5x FY2027E the stock is $1,031; at 9x it is $1,855. The multiple, not the earnings, is where most of the variance lives — and unlike a compounder, a commodity producer's multiple normally contracts as earnings peak, because the market is discounting the reversion in advance.
• That the AI storage bid is structural, not a restocking cycle. (Knowledge-base derived, labelled as such: multiple independent claims in Section 7 argue memory has de-cyclicalised; at least two argue the opposite.) The best filing evidence for it is the $511M of customer advances under long-term agreements disclosed in the April 10-Q. The best filing evidence against it is that exabytes sold were flat — the shortage is not currently manifesting as more units shipped.

6b. The return bridge (why the multiple moves)

Expected return over the next twelve months decomposes as: EPS growth (+208% on the consensus path, FY2026E $66.87 to FY2027E $206.12) + multiple drift (COMPRESSION, from 21.35x on the current forward year to roughly 6.75x on the then-current forward year) + shareholder yield (0.00% — no dividend, buyback unexecuted).

Almost none of that arithmetic is meaningful in the usual way, and it is important to say why rather than to print it and move on. The 21.35x-to-6.93x step is not a de-rating; it is the forward-multiple rolldown that the expectations addendum explicitly warns against calling compression. The real question is what multiple the market pays on FY2027 earnings at each date, and today it pays 6.93x.

Our base case assumes the multiple HOLDS at roughly 6.75x on FY2027E, and that the return therefore comes entirely from whether the earnings arrive. We are explicit that this is a choice and that it can be attacked from both sides:

• The bear attack: a peak-cycle multiple should compress further, toward 4-5x, because the market discounts the reversion. At 5x FY2027E the stock is $1,031, −27.8%.
• The bull attack: if memory has genuinely de-cyclicalised — the Santi claim of 2026-05-14, that these names deserve "a re-rating" because of five-year contracts — then the multiple should expand toward the low teens that a structural-growth semiconductor commands. At 12x FY2027E the stock is $2,473, +73.2%.

The entire investment case is that single unresolved question, and we do not have an edge on it. What we can say with confidence is the asymmetry of the arithmetic: the earnings estimate has a 3.06x high-to-low spread on FY2028 and the multiple has plausibly a 2.4x range. Multiplied together, that is a distribution too wide to size a position against on the eve of a print.

If most of the return is earnings growth, say so. Here it is: on our base case, 100% of the expected return is earnings delivery and 0% is re-rating, with 0% of shareholder yield to cushion it. There is no dividend, the buyback is authorised but unexecuted, and diluted shares are rising. That is the opposite of a defensive return profile.

6c. Variant perception (where we differ, what would surprise)

• We differ from the street by a very large margin, and we hold that view with low confidence. Consensus is $2,088.57 (+46.3%) with 13 buy, 2 hold, 0 sell across 15 analysts and a low target of $1,200. Our base of $1,390 is 33.4% below consensus and only 15.8% above the street's lowest target. The source of the difference is not the earnings estimate — we use the street's own $206.12 — it is the multiple we are willing to pay for it. We are treating this as a commodity at a cyclical peak; the street is treating it as an AI infrastructure asset re-rating. We could be wrong, and we say so.
• We differ from the vendor on what the growth actually IS, and this is our most defensible divergence. Every screen reports +251% revenue growth. The filing reports that the same 251% came from a 248% ASP increase against flat exabytes. A screen cannot distinguish a company selling more from a company charging more, and for a commodity producer that distinction is the entire thesis. Watchable number: exabytes sold, disclosed each quarter in the 10-Q MD&A. The first quarter in which ASP growth decelerates while exabytes stay flat is the quarter revenue stops growing.
• We differ from the vendor on the balance sheet in a direction that HELPS the company. bal_a shows $1.829B of long-term debt and $561M of net debt. The 10-Q shows the term loan repaid in full on 2026-03-04 and the company holding $3.528B of net cash. Anyone screening on the vendor's annual balance sheet is looking at a capital structure that ceased to exist five months ago.
• Positive surprise that would force a re-rate: tomorrow's print delivering above $8.440B with gross margin at or above 78% and, critically, exabytes sold GROWING rather than flat — which would convert the story from a price spike to a volume ramp and would justify a double-digit multiple. Alternatively, disclosure that a meaningful share of the $6B buyback was executed below $1,500, or that the $511M of contract liabilities has grown materially, evidencing more customers locking in multi-year supply.
• Negative surprise that would break the thesis: gross margin below roughly 70%, which would say the ASP peak is behind us; any language retreating from "persist through calendar year 2026 and beyond"; a fiscal-2027 revenue guide implying materially less than the $48.670B consensus; an inventory build disclosed without a corresponding order book; or a Flash Ventures utilisation or capital-call disclosure implying Sandisk must fund capacity expansion into the spike. Any of these hits a stock that has risen 3,354% in twelve months.

Synthos fair values

All three anchors are stated arithmetic — a named multiple applied to a named estimate line. No discounted cash flow is used anywhere. Each anchor is cross-checked against a second, independent construction, and we only kept anchors where the two constructions agreed.

• Bear ~$600 — 9x the FY2026 consensus EPS of $66.874 (11 analysts). Independent cross-check: 10x the trailing tangible book value per share of $59.34 = $593. The two constructions agree to within 1.5%. The scenario: the NAND price spike peaks with tomorrow's print or shortly after, FY2027 never happens, and the market is left valuing FY2026's earnings power — which is itself elevated — at a normal cyclical multiple. Note that even at $600 the stock would trade at 10.1x tangible book and 43% above its own 200-day moving average; a genuine cycle trough for a memory company can go well below this, and FY2023's actual result was a $2.143B loss. We are not claiming $600 is a floor. We are claiming it is where the arithmetic lands if FY2027 does not arrive. −58.0%.
• Base ~$1,390 — 6.75x the FY2027 consensus EPS of $206.124 (12 analysts). Independent cross-check: 10x the annualised Q4 FY2026 consensus run-rate of $34.80 × 4 = $139.20, which gives $1,392. The two constructions agree to within 0.1%, which is the reason we chose this pair. Sensitivity, stated openly: 6x gives $1,237 (−13.3%); 7x gives $1,443 (+1.1%); 8x gives $1,649 (+15.5%). The whole answer lives in a 6-8x band and no greater precision is claimed. The scenario: consensus is roughly right about FY2027, the market continues to apply a peak-cycle discount rather than re-rating, and there is no dividend or executed buyback to add yield. −2.6% against spot.
• Bull ~$2,475 — 12x the FY2027 consensus EPS of $206.124. Independent cross-check: 10x the FY2028 consensus EPS of $247.562 = $2,476 (though FY2028 EPS rests on only 5 analysts and is used as a check, not an anchor). The two constructions agree to within 0.1%. The scenario: memory de-cyclicalises, the long-term customer agreements and $511M of advances prove to be the leading edge of contracted multi-year supply, Datacenter volume growth (+160% exabytes) becomes the company-level story rather than a segment story, the Nanya arrangement builds a second product line, and the market re-rates NAND toward a structural-growth multiple. This bull case requires BOTH the earnings AND a re-rating, which is why it is the bull case and not the base. +73.4%. For calibration: $2,475 is 6.0% above the 52-week high of $2,335.00 and 18.5% above the street's consensus target of $2,088.57, but 20.2% below the street's high target of $3,100.

Base is 2.6% BELOW spot. The bull is 4.1x the bear. That ratio — not the base case — is the honest headline of this section. A fair-value range that wide is not analytical laziness; it is the correct description of a commodity producer at a price extreme, one day before a print, whose forward EPS estimates carry a 3.06x high-to-low spread. When the range is this wide and the base is below spot, position sizing rather than direction is the decision, and the answer is Watch.

7. Knowledge base — a real lane, but a memory lane, not a Sandisk lane

Raw hits: 23. Entity matches: 19. Text-only matches: 4. Used: 21. Discarded: 2 (both quarantine-flagged for misattribution). Distinct channels: 11. Named speakers: 9.

This is one of the better-populated lanes we have written against — but the breadth is about the sector, and it is essential to say so before quoting any of it.

Searches run. The primary sweep covered SNDK, Sandisk and SanDisk (19 entity + 4 text-only). Supplementary single-term sweeps: NAND (0 entity, 11 text-only — all already in the primary set or about Micron/Samsung), flash memory (0 entity, 2 text-only — one is a China-technology claim with no Sandisk content), Kioxia (0 hits — the manufacturing partner that supplies substantially all of Sandisk's product has zero coverage in 51,928 claims), David Goeckeler (0 hits — no coverage of the chief executive), Western Digital (3 entity + 2 text-only, of which 2 duplicate the primary set).

Broad sweeps, reported with their discard reasons as required. memory returned 7 entity and 508 text-only matches. HBM returned 234 text-only. DRAM returned 4 entity and 227 text-only. Micron returned 48 entity and 170 text-only. storage returned 1 entity and 153 text-only. Total broad-sweep volume: roughly 1,350 matches, and we discarded essentially all of them. The reasons, stated concretely rather than gestured at:

1. The overwhelming majority of memory, HBM and DRAM hits are about DRAM and high-bandwidth memory — which Sandisk does not make. Sandisk is a NAND flash company. HBM is a DRAM product. A claim that HBM is sold out is not a claim about Sandisk, and treating it as one is precisely the "product claim is not an equity claim" error the attribution policy names.

2. Most of the rest are about Micron. Micron has 48 entity matches in its own right and is by a wide margin the most-discussed memory name in the base. A Micron claim is not a Sandisk claim — Micron makes DRAM, HBM and NAND, owns its own fabs, and has a completely different cost structure. Where a claim names both, we use it; where it names only Micron, we discard it.

3. storage collides badly. The single entity match dates to 2019-05-31 on a biotech channel and concerns biological storage. The 153 text-only matches are dominated by energy storage and data storage in a generic infrastructure sense.

4. The one genuine cross-sector claim we kept from the broad sweep is already in the primary set.

The 21 used claims. Grouped by what they actually argue.

Group A — the structural bull case (8 claims). These argue memory is the binding AI constraint.

> 2026-02-01 · bullish · conviction 88 · horizon thesis · channel: jordi_visser · NO NAMED SPEAKER · skill 2.0 · entities: Micron, SanDisk, SK Hynix, Samsung

> "Memory is a key AI bottleneck still in early innings; NAND/DRAM prices spiking, supply sold out, HBM revenue doubling — parabolic like silver."

> 2026-01-24 · bullish · conviction 80 · horizon thesis · channel: jordi_visser · speaker: jordi_visser · role: independent · skill 2.0 · entities: MU, WDC, SNDK

> "High-bandwidth/DRAM memory is scarce versus surging AI demand — Micron, SanDisk, Western Digital, SK Hynix, Samsung in 'banana zone'; a race to buy at size."

> 2026-06-17 · bullish · conviction 75 · horizon thesis · channel: jordi_visser · speaker: Jordi Visser · skill 2.0 · entities: MU, SNDK, WDC

> "Memory is a 5-year-minimum bull market — DRAM took off since September, humanoids will need lots of memory; Micron, SanDisk, Western Digital have gone up 3-10x."

> Test: "memory names remain a bull market over the next 5 years"

> 2026-05-14 · bullish · conviction 82 · horizon thesis · channel: empire · speaker: Santi · entities: SNDK, MU

> "Memory is de-cyclicalizing as agentic AI needs more memory; SanDisk/Micron trade sub-10x forward PE with 5-year contracts, a re-rating. Rotated cash in, fairly concentrated."

> Test: "Memory demand persists over next 5-10 years without a capex-driven oversupply and price collapse."

> 2026-05-15 · bullish · conviction 65 · channel: bankless · speaker: David · entities: DRAM, SNDK, MU

> "Memory is the new AI bottleneck — the 'AI agent phase' needing bigger context/longer runs; DRAM, SanDisk and Micron are the current inflection in AI stocks as compute demand squeezes memory supply."

Plus 2026-04-10 (forward_guidance, no named speaker, conviction 80, "Anything tied to compute demand is breaking out"), 2026-03-20 (money_of_mine, speaker Leon, guest, conviction 60, "the memory/on-device AI shift is a real theme now running; ways to express it are Micron, SanDisk and SK Hynix") and 2026-05-13 (anthony_pompliano_show, speaker Asher Genoot, guest, conviction 45, "You're seeing all these memory companies skyrocket in terms of their share price, like SanDisk and um and SK").

Group B — the one genuinely name-level claim (1 claim). This is the only claim in 51,928 that carries Sandisk as its sole entity and argues something specific about Sandisk's own demand.

> 2026-01-18 · bullish · conviction 68 · horizon thesis · channel: jordi_visser · NO NAMED SPEAKER · role: independent · skill 2.0 · entities: SNDK (only)

> "SanDisk NAND demand is genuinely incremental, not in prior forecasts; [a next-generation accelerator platform] shifts inference memory to persistent local storage, exploding NAND demand. Rerated on real news, not bubble."

This is the highest-value claim in the lane and also the one to be most careful with. It is a specific, mechanistic, falsifiable argument — inference workloads move memory into persistent local storage, which is NAND rather than DRAM, and that demand was absent from forecasts. It is exactly what the Datacenter disclosure in the 10-Q shows: +645% revenue on +160% exabytes. But it carries speaker: null and is attributed only to a channel, which under the 4-lane policy is the weakest sourcing we accept.

Group C — the bearish and cautionary counterweight (4 claims), which is what makes this lane usable.

> 2026-05-12 · neutral · conviction 65 · horizon principle · channel: compound_and_friends · speaker: Josh Brown · entities: MU, SNDK, STX, WDC

> "Don't mistake memory names' single-digit PEs for cheapness — they're the most cyclical names on the planet with wildly swinging operating margins, so they deserve a discount."

This is the single most valuation-relevant claim in the entire base for this name, it is named-speaker attributed, and it argues directly against the most obvious reason a screen would surface SNDK. It is quoted in Section 2 and it is a direct input to the Watch verdict.

> 2026-06-23 · bearish · conviction 62 · horizon fact · channel: compound_and_friends · no named speaker · entities: MU, WDC, SNDK

> "Memory/semi stocks severely overextended — semi index 68% above its 200-day MA, most stretched in 25 years; Micron 161% above; pullback severely overdue, need to sober up."

Sandisk closed today 66.9% above its own 200-day moving average. That claim was made six weeks ago and its central observation still describes this chart.

> 2026-06-09 · bearish · conviction 64 · horizon thesis · channel: arthur_hayes · speaker: Arthur Hayes · skill 1.2 · entities: SK Hynix, SanDisk, Samsung

> "The 50x memory names are priced for accelerating earnings, but AI capex's second derivative is decelerating (won't go 1T to 2T), so that premium deflates."

> 2026-07-28 · bearish · conviction 50 · horizon fact · speaker: Josh Brown · role: host · entities: MU, SNDK, WDC

> "A disgusting but necessary unwind in the memory/semi names they own; $46B of inflows into semiconductor ETFs this year is unsustainable, and investors are sick of the circular [a frontier AI developer]/Nvidia financing deals."

> [Redacted: the original claim names an AI laboratory; replaced per house rule.]

Group D — recent tape-reading and sector-rotation colour (7 claims), used as context only. Includes 2026-07-31 (darius_dale, speaker Adam Turnquist, bullish, conviction 62: "Memory-name sell-off (MU, SanDisk) was leveraged-positioning unwind, not the memory trade ending; 50%+ drawdowns after historic rallies now showing technical evidence of a bottom"), 2026-06-23 (andreas_steno, speaker Andreas Steno, bullish 72), 2026-03-20 (compound_and_friends, bullish 50), 2026-01-06 (compound_and_friends, neutral 60: "Pockets of nonsense (SanDisk +25%) always exist, but the presence of a circus doesn't mean the whole market is a circus" — note that this one uses Sandisk as an EXAMPLE OF FROTH), 2026-01-25 (jordi_visser, bullish 83), and two text-only claims (2026-03-24, SanDisk named among 2026 leaders; 2026-07-07, "Despite the memory-name unwind (SanDisk −30%, WDC −28%, Micron −22%), guessing this isn't the ultimate top").

The two discarded claims, and why.

> 2026-03-02 · bearish · conviction 60 · channel: quarantine_misattributed · QUARANTINED: SUMMARY_OF_VISSER · entities: Micron, Sandisk

> "Skeptical the memory-shortage melt-up in Micron and Sandisk goes straight up; markets follow the software disruption signal instead."

> 2026-05-23 · bullish · conviction 70 · channel: quarantine_misattributed · QUARANTINED: NOT_VISSER · speaker: Dan · entities: MU

> "Memory is in a super cycle; multiples are high versus history but the street underestimates the demand growth in memory/components (Micron, Sandisk)."

Both are excluded from the conviction pool. They carry explicit quarantine flags indicating the distillation pipeline could not establish who actually said them — one is a summary of a voice rather than the voice itself, the other was attributed to a voice it does not belong to. This is the attribution failure the 4-lane policy exists to catch, and it is directly relevant here because both quarantined claims involve the same channel that supplies the largest single block of usable claims in this lane. Note that the discard is directionally neutral: one bull and one bear were removed.

The attribution finding, stated plainly because it is the most important thing in this section.

The jordi_visser channel supplies 5 of the 18 usable entity claims — 27.8% of the lane — at a skill weight of 2.0, the highest in the base. Of those five, only ONE (2026-06-17) carries a named speaker. The other four are channel-attributed with speaker: null. And two further claims sit in quarantine precisely because they were misattributed relative to that same voice. A single voice therefore drives roughly a quarter of the lane at double weight with mostly channel-level attribution. That is the exact pattern the attribution policy was written to flag, and we flag it rather than letting the conviction number absorb it silently. The compound_and_friends channel supplies a further 5 entity claims plus 2 text-only — another 27.8%. Two channels account for over half of this lane.

No claim in the lane carries a speaker_role indicating company management, so the half-weight management discount does not apply to any knowledge-base claim. It does apply to the company's own forward statements quoted from the filings — "we expect these conditions to persist through calendar year 2026 and beyond" — and those are half-weighted throughout.

Conclusion. The knowledge base has a strong, broad, well-populated view on MEMORY as an AI-infrastructure theme, and it is net bullish. It has one claim about Sandisk specifically. The theme-level view supports the FY2027 revenue ramp in a general way and is independently corroborated by the Datacenter disclosure in the 10-Q. What it does not support is paying up for SNDK rather than any other expression of the same trade — the claims almost invariably list SNDK alongside MU, WDC, SK Hynix and Samsung, and several of the most credible are explicitly about Micron. The two most recent claims split in direction. The single most valuation-specific claim is a caution against exactly the multiple that makes this stock screen well. Conviction: Medium. Breadth: 11. Net: positive-medium. This is a real lane and it does not get us to Buy.

8. Data integrity — what we tested, what we rejected, and why

Fourteen findings. Every confirmed defect class was tested; the clean results are reported alongside the failures, because a verified-clean check is a finding.

1. bal_a is 13 months stale and describes a capital structure that no longer exists — REJECTED for all leverage and enterprise-value work. The newest balance sheet in the file is 2025-06-27, showing $1.829B of long-term debt, $2.042B of total debt and $561M of net debt. The 10-Q filed 2026-05-01 states that the $2.0B Term Loan Facility was "settled in full" on 2026-03-04 using cash on hand, with a $46M extinguishment charge, and that the $1.5B revolver was undrawn with full availability at 2026-04-03. The vendor's annual balance sheet reports $561M of net debt; the filing shows the company in a net CASH position; we use the filing. Using bal_a would have produced an enterprise value of $211.977B against the correct $207.888B — a $4.089B error — and would have characterised a net-cash company as levered.

2. inc_a has only FOUR annual rows and the newest is 13 months old — the entire FY2026 transformation is invisible in it. The annual series runs FY2022-FY2025 and ends at 2025-06-27. It does not contain FY2026, which ended 2026-07-03 and is being reported tomorrow. Every FY2026 figure in this dive comes from inc_q (three quarters, through 2026-04-03), cross-checked against earn_cal and the two 10-Qs. The fourth quarter is unreported and is not estimated anywhere in this dive.

3. The $1.8B goodwill impairment is classified INCONSISTENTLY between the vendor's annual and quarterly statements — the quarterly classification is wrong and the filing settles it. The FY2025 annual row shows operatingExpenses of $3,589M against R&D of $1,132M plus SG&A of $573M — a $1,884M residual — producing reported operating income of −$1,377M. The March-2025 quarterly row instead shows normal operating expenses of $424M and dumps −$1,859M into totalOtherIncomeExpensesNet, producing operating income of only −$42M. The two cannot both be right, and the quarterly sum (+$484M of FY2025 operating income) differs from the annual figure (−$1,377M) by exactly $1,861M. The 10-K settles it: the impairment "resulted in the recognition of a $1.8 billion impairment charge during the third quarter of the year ended June 27, 2025 which was recorded in the accompanying Consolidated Statements of Operations", and the 10-Q lists "Goodwill impairment" among the operating expense line items. The annual treatment is correct; the quarterly treatment is wrong. We use the annual classification and report FY2025 operating margin on both bases: −18.7% as reported, +6.6% excluding the impairment.

4. est.ebitdaAvg and est.ebitAvg carry a FABRICATION SIGNATURE and are REJECTED in full. In every one of the five forward years, ebitdaAvg is exactly 14.44% of revenueAvg and ebitAvg is exactly 10.78% — FY2026, FY2027, FY2028, FY2029 and FY2030 without deviation. This is the same class of signature the data contract documents for MU (exactly 50.000% and 33.000%): a fixed ratio applied mechanically to the revenue row, not an analyst aggregate. It also produces an arithmetic impossibility in every year: netIncomeAvg EXCEEDS ebitdaAvg — FY2027 net income of $31.948B against EBITDA of $7.028B, FY2028 $34.601B against $8.214B. Net income after tax and interest cannot exceed earnings before interest, tax, depreciation and amortisation. Both rows are discarded entirely. All forward valuation in this dive runs on epsAvg and revenueAvg, which we verified against reported quarters (Section 6).

5. The FY2030 estimate row is nonsensical and is REJECTED; FY2029 is too thin to use. revenueAvg for FY2030 is $11.566B against FY2029's $52.103B — a 77.8% collapse — on ONE analyst. FY2029 rests on two. Neither row is used for any conclusion. We report the shape (the long-dated consensus models a cycle roll-over) as an observation about analyst framing and explicitly not as a forecast.

6. seg_geo contains data from a DIFFERENT COMPANY — three rows belong to the pre-2016 SanDisk Corporation. Alongside the FY2025 row the vendor supplies rows dated 2014-12-28 (fiscal 2015), 2013-12-29 (fiscal 2014) and 2011-01-02 (fiscal 2011), with a December fiscal year-end and regional labels ("TAIWAN, PROVINCE OF CHINA", "Other Asia-Pacific") that do not appear in any Sandisk Corporation filing. The 2014-12-28 row sums to $6.628B. Sandisk Corporation (CIK 0002023554) was incorporated as a Delaware holding company for the Western Digital NAND business and had no public existence before 2025-02-24; its fiscal year ends on the Friday nearest 30 June, not in December. These rows belong to the former SanDisk Corporation, acquired by Western Digital in 2016 — a different registrant with a different CIK. All three are discarded. Only the 2025-06-27 row is used.

7. profile.ipoDate is 1995-11-08 — the former SanDisk Corporation's IPO, not this company's listing — REJECTED. Sandisk Corporation began trading on Nasdaq under "SNDK" on 2025-02-24, per both the 10-K and the 10-Q. The vendor's IPO date is off by thirty years and belongs to the predecessor entity. Related and consistent with finding 6.

8. Share count — TESTED and CLEAN, and the check mattered at this price level. Market capitalisation of $211,415,888,895 divided by $1,427.62 implies 148,089,750 shares. The 10-Q cover page states: "As of the close of business on April 24, 2026, 148,089,758 shares of common stock, par value $0.01 per share, were outstanding." Agreement to 0.000005%. Independent third check: the 8-K of 2026-05-15 describes a mini-tender for 100,000 shares as "less than 0.07% of the Company's common stock as of April 24, 2026", implying at least 142.9M shares. No share-class error, no split, no partial-unit problem. Vendor and filing agree exactly. Disclosed separately (not a defect): the diluted EPS denominator in the April quarter was 157,000,000 — 6.0% above shares outstanding — and has risen from 145M to 157M in four quarters. Fully diluted market capitalisation would be $224.14B.

9. Currency — TESTED and CLEAN. inc_a[0].reportedCurrency is USD; bal_a and cf_a are USD; profile.currency is USD; the exchange is NASDAQ Global Select. No currency mixing. Every multiple in this dive is struck in a single currency. The only foreign-currency exposure disclosed is operational — Flash Ventures investments are yen-denominated and a hypothetical 10% adverse move would produce a $29 million fair-value loss at 2026-04-03.

10. km_ttm.netDebtTTM and enterpriseValueTTM — TESTED and CLEAN. This is the defect that broke META ($45.7B), AAPL (sign inverted) and BRK-B ($321.4B), and it does not occur here. shortTermInvestments is genuinely $0, so there is nothing to omit. Our independent rebuild — market cap $211.416B plus ~$0.207B of debt less ~$3.735B of cash — reproduces the vendor's $207.888B enterprise value exactly, and the net-cash sign is correct. We use the vendor's EV, corrected only for the disclosed post-balance-sheet Nanya outflow, and say so.

11. seg_prod and seg_geo sums — TESTED and CLEAN, but labels are STALE and coverage is one year. seg_prod (Client Devices $4.127B + Cloud $960M + Consumer $2.268B) sums to $7,355M against reported FY2025 revenue of $7,355M — exact. seg_geo sums to $7,355M — exact. No double-counting, no disclosure-basis merge, no 2.05x XOM-style failure. What IS wrong is the labelling and the coverage: the 10-Q renamed the segments to Datacenter, Edge and Consumer and the vendor still uses the old names; and only FY2025 is supplied, so there is no FY2026 mix and no prior-year comparison. We relabel to the filing's vocabulary, use the FY2025 figures, and take the FY2026 mix from the 10-Q's MD&A percentages rather than estimating it.

12. Zeroed and null fields — SCANNED, with three real findings. shortTermInvestments is $0 in all three balance-sheet years — verified as genuine, not a null masquerading as zero, because longTermInvestments is populated ($734M/$510M/$617M) and the company holds cash in money-market funds per the 10-Q. retainedEarnings and additionalPaidInCapital are $0 for FY2024 and FY2023 with the entire equity balance sitting in commonStock — a carve-out artefact, correct for a business that had no separate equity before the spin, and it makes the FY2023-FY2024 equity composition uninformative. dividendPerShareTTM, dividendYieldTTM and dividendPayoutRatioTTM are all 0 — verified genuine: Sandisk pays no dividend. longTermDebtToCapitalRatioTTM is 0 — verified genuine: the term loan was repaid.

13. quote.yearHigh/yearLow disagree with tech.hi52/lo52 — we use tech. The quote block reports $2,354.39 / $40.53; the computed technical block reports $2,335.00 / $40.69. Discrepancies of 0.8% and 0.4%. We use the tech figures throughout, for internal consistency with the moving averages, drawdown and relative-return figures computed from the same close series. Note that the 52-week window covers only the post-spin period; there is no pre-2025-02-24 price history for this security and none is used.

14. Non-equity tripwire — CHECKED and PASSED, and at a $1,427.62 price the check was warranted. SNDK is common stock, $0.01 par value, confirmed by the 10-Q cover page, listed on Nasdaq Global Select, CUSIP 80004C200, ISIN US80004C2008. The price is not par-like ($25, $100 or $1,000). Beta is 5.1932, roughly seventeen times the sub-0.3 threshold that flags a fixed-income-like instrument. There is no dividend at all, fixed or otherwise. Volume was 13,742,730 shares — roughly $19.6B of turnover — against a 13,210,000 average. And the 52-week band of $40.69 to $2,335.00 is a 5,638% range, the opposite of a narrow band. This is unambiguously common equity.

Verified clean and worth recording, because these are the fields that most often fail elsewhere: ratios_ttm.grossProfitMarginTTM (0.56038) reproduces exactly from our own sum of quarterly gross profit and revenue ($7,388M / $13,184M); operatingProfitMarginTTM (0.40928) reproduces exactly; netProfitMarginTTM (0.34185) reproduces exactly; effectiveTaxRateTTM (12.49%) reproduces from $643M of tax on $5,150M of pretax income; interestCoverageRatioTTM (43.5x) reproduces from $5,396M of operating income on $124M of interest expense; incomeQualityTTM (1.029) correctly reflects ~$4.64B of operating cash flow against $4.507B of net income; priceToEarningsDilutedRatioTTM (49.62x) reproduces exactly from $1,427.62 on trailing diluted EPS of $28.77. revenuePerShareTTM of $89.081 confirms the per-share metrics are struck on 148.0M shares, consistent with the market-cap denominator. Sandisk's derived-metric block is, on the whole, one of the more internally consistent in this batch — the failures are concentrated in staleness (findings 1, 2, 11), wrong-entity contamination (findings 6, 7) and the fabricated est EBIT/EBITDA rows (finding 4).

8-K extraction quality — reported as a finding. All three 8-Ks in this archive carry substantive Item text, not bare cover pages, which is better than the norm and is worth recording: 2026-03-25 (Nanya equity investment and DRAM supply arrangement — full Item 1.01 narrative), 2026-04-30 (Item 2.02 earnings reference plus Item 8.01 disclosing the $6B repurchase authorisation in full), 2026-05-15 (mini-tender rejection, full narrative). What IS absent is Exhibit 99.1 in each case — so the 2026-04-30 8-K tells us the third quarter was announced but contains no financial figures whatsoever. The April-quarter numbers in this dive therefore come from the 10-Q filed the following day, 2026-05-01, and from the vendor's quarterly income statement, cross-checked against earn_cal.

9. Technicals

• Price $1,427.62. −38.86% from the 52-week high of $2,335.00, and +3,408.5% above the 52-week low of $40.69. Position within the annual range: 60.5th percentile.
• Moving averages, a split decision at an extreme: −16.26% below a 50-day average of $1,704.83; +66.91% above a 200-day average of $855.31. The gap between the two moving averages is itself the story — the 50-day sits 99.3% above the 200-day. A chart in which the short average is double the long average is a chart in the late stage of a vertical move, in either direction.
• RSI 44.3 — neutral, neither overbought nor oversold, which is itself informative after a 3,354% year. MACD −136.19 — deeply negative, the largest absolute negative reading in this batch, though its scale is a function of the price level.
• Maximum drawdown from peak over the trailing year: −38.86% — i.e. the current price is the worst point of the drawdown, having recovered nothing net.
• Relative performance, and the numbers are not typographical errors: 3-month +13.7% vs SPY +7.6% and QQQ +7.7%; 6-month +114.6% vs SPY +11.1% and QQQ +15.7%; 12-month +3,354.2% vs SPY +24.3% and QQQ +30.8%. SNDK outperformed the S&P 500 by roughly 3,330 percentage points over twelve months.
• Beta 5.1932. Treat this number with care. It is computed over a period in which the security has only existed for about seventeen months and has moved in one dominant direction. It is directionally right — this instrument is extraordinarily volatile — but it is not a stable estimate of forward sensitivity and should not be plugged into a risk model as though it were.
• Sentiment: 13 buy, 2 hold, 0 sell across 15 analysts; consensus target $2,088.57 (+46.3%), median $1,975, high $3,100 (+117.1%), low $1,200 — 15.9% BELOW spot. Zero sell ratings on a stock up 3,354% in a year. The vendor's composite rating is A− / 4, with sub-scores of 5 on return on equity and return on assets, 4 on debt-to-equity, and 2 on both price-to-earnings and price-to-book — the rating machinery is telling you the same thing this dive is: excellent returns, no leverage, expensive on asset-based measures.

Today's move and what it does to the entry

SNDK closed 2026-08-04 at $1,427.62, up 10.84% or $139.59 from a previous close of $1,288.03. It opened at $1,357.68 — a 5.41% gap — traded a $1,340.00 to $1,446.62 range, and closed 1.31% below the day's high on 13,742,730 shares, roughly 4% above its average volume.

This was a sector event and the batch context confirms it. The whole memory and semiconductor complex moved together: MRVL +12.8%, INTC +10.9%, LRCX +7.9%, MU +7.6%, KLAC +7.0%, AMD +7.0%, AMAT +5.5%, ASML +4.2%. Sandisk's 10.84% is at the top of that distribution but is not an outlier within it, and no company-specific news for 2026-08-04 appears anywhere in this file. The last company event was the 8-K of 2026-05-15. The gap is REAL — it is not a data artefact, a split adjustment or a stale quote — and the quote timestamp of 1785873601 confirms a 2026-08-04 20:00:01 UTC close print.

The honest read, and it is genuinely two-sided. Unlike a stock gapping into new highs, SNDK gapped from 38.9% below its 52-week high with RSI at 44.3 and MACD deeply negative — this is a bounce inside a substantial correction, with the 50-day average 19.4% overhead at $1,704.83. In pure chart terms it resolves nothing. The knowledge base has claims on both interpretations from within the last six weeks: one (2026-07-31, Adam Turnquist) reads the sell-off as "leveraged-positioning unwind, not the memory trade ending" with "technical evidence of a bottom"; another (2026-06-23, compound_and_friends) reads the whole complex as "severely overextended... pullback severely overdue."

But the chart is not the reason to decline the entry, and neither is the gap. The reason is the conjunction of three facts:

1. Our base fair value of $1,390 is BELOW the closing price. There is no positive expected return in our own arithmetic at $1,427.62.

2. Earnings are tomorrow, against a bar requiring 41.8% sequential revenue growth, and we have no informational edge on that print.

3. You would be paying a 10.84% one-day advance for the privilege of taking that risk.

What we are telling the reader, plainly: do not buy this today. Not because the business is bad — the April quarter was extraordinary and it is filing-verified — but because an entry the day before an unseeable print, after a double-digit gap, at a price above our own fair value, has no edge in it in any direction. The setup that changes this is set out in Section 11, and the cheapest of the flip conditions is simply waiting twenty-four hours.

10. Insiders — eight transactions, two people, zero purchases, and every sale above today's price

All eight insider transactions in this file are DISPOSALS, and they involve only two officers.

The reading, stated carefully.

Two of the eight are F-InKind — shares withheld by the issuer to satisfy tax on vesting equity. These are not sales and carry no signal, exactly as in the KLA case. One is a gift (2,694 shares, $0 price), which is estate or charitable planning and also carries no directional signal.

Five are S-Sale — genuine open-market disposals — totalling 3,200 shares. In absolute terms that is small: roughly $5.6 million at the transaction prices, against a $211B market capitalisation. The pattern (600 shares, repeated, on the first business day of consecutive months) is consistent with a Rule 10b5-1 pre-arranged plan, and Sandisk's own 8-K language references such plans in the buyback context. We do not read routine plan sales as a bearish signal and we say so.

What IS worth noting, and it is the only genuinely informative thing in this table. Every single open-market sale executed at a price ABOVE today's close — at $1,736.00, $1,755.31, $1,757.00, $1,758.40 and $2,088.00, i.e. 21.6% to 46.3% above the $1,427.62 close. The stock has traded down through all five execution prices since. That is not insider prescience; it is simply a record that the last time insiders transacted, the price was materially higher than it is now. It is a data point about the range, not about their conviction.

Two absences matter more than the eight transactions present.

First, there are ZERO open-market purchases. Not one insider in this file has bought a share on the open market at any price, including during the 38.9% drawdown from $2,335. In a stock this volatile, a discretionary purchase into weakness would have been a real signal. There is none.

Second, chief executive David Goeckeler does not appear anywhere in this file. Neither does the chief financial officer. The eight most recent transactions on record involve only the chief legal officer and the chief technology officer. The file contains no information about the chief executive's holding, his equity vesting schedule, or whether he has transacted at all. We report that as an absence of data, not as an absence of activity — the payload supplies only the last eight transactions, and a larger CEO grant or sale outside that window would not appear.

Overall reading: NEUTRAL, leaning very mildly negative. Small routine plan sales, no purchases, no chief-executive visibility, and an aggregate direction that is entirely one way. It changes nothing about the verdict and it is reported at that weight.

11. Verdict, kill-criteria and flip conditions

Watch.

And the timing is not incidental to that word — the company reports in less than twenty-four hours.

What is genuinely extraordinary, and none of it is in dispute. In the three months to 2026-04-03 Sandisk generated $5.950B of revenue, up 251.0% year on year, at a 78.4% gross margin and a 69.1% operating margin, producing $3.615B of net income and $23.03 of GAAP diluted EPS against a −$13.33 loss in the same quarter a year earlier. We tested that quarter against every one-off in the taxonomy and found none — the effective tax rate of 12.0% comes from disclosed Malaysian tax holidays, not a valuation-allowance release; there is no spin gain; there is no inventory revaluation; the only discrete item is a $46M debt-extinguishment charge that makes the clean number better. The balance sheet is now pristine: the $2.0B separation term loan was repaid in full on 2026-03-04, the $1.5B revolver is undrawn to 2030, and the company holds roughly $3.5B of net cash. Customer concentration is genuinely low — no customer above 10% of revenue in FY2025 or FY2024, top ten at 40%. Customers are pre-paying for supply, producing $511M of contract liabilities under long-term agreements. And the Datacenter business is growing in units, not just price: exabytes sold there rose 160% year on year.

What we are declining to pay for. 6.93x FY2027 consensus EPS of $206.12 — a number that requires the price of NAND flash to stay at or above a historic spike for two more years. A consensus that has revenue rising 143% in FY2027 to $48.670B and net margin reaching 65.6%, from a company that lost $3.392B cumulatively across the four fiscal years in this file and whose revenue fell 37.6% in a single year in FY2023. A supply chain Sandisk does not own, in which it must pay half the fixed costs regardless of what it buys and guarantee half the joint venture's Japanese lease obligations. A beta of 5.19. No dividend, an unexecuted buyback, and diluted shares up 8.3% in three quarters — so shareholder yield is 0.00% and every dollar of return must come from earnings and multiple. A base fair value of $1,390 that is 2.6% BELOW the closing price and 33.4% below the street's $2,088.57. And an entry price that includes a 10.84% one-day gap on the eve of a print.

The distinction that matters, and it is different from the usual one. We are not saying the earnings are fake — we checked, in detail, and they are real. We are not saying the AI storage demand is imaginary — the Datacenter volume disclosure and the customer advances both argue it is not. We are saying that a commodity producer earning a 78.4% gross margin on flat volume because its selling price tripled is, definitionally, a business at a price extreme, and that our own arithmetic puts fair value slightly below the market price. In the Synthos frame, a name where our base fair value sits below spot and the decisive catalyst is one day away is a Watch by construction. The one thing we would not do is manufacture urgency where the calendar supplies a free option: waiting one day costs nothing and resolves the largest single uncertainty in the file.

Pre-registered KILL criteria — what would take this to Avoid:

• Gross margin below roughly 70% in the 2026-08-05 print or the following quarter. The April quarter was 78.4%. A break of 70 says the ASP peak is behind, and at 6.93x peak earnings there is nothing underneath.
• Exabytes sold declining at the company level while ASP growth decelerates. Volume was already flat in April. Flat volume plus falling price is a revenue collapse, and it is exactly what FY2023 was.
• Any retreat from the disclosed expectation that favourable pricing will "persist through calendar year 2026 and beyond" — half-weighted as management voice, but a withdrawal of it would be a hard signal.
• A fiscal-2027 revenue guide implying materially below $40B against the $48.670B consensus. That is the FY2027 ramp failing.
• A Flash Ventures capital call or a disclosed increase in Sandisk's funding obligation into the spike, which would confirm that expanding supply is expensive and that Sandisk does not control the timing.
• An inventory write-down, or a return to the underutilisation charges disclosed in FY2025 ($75M) and the nine months to April 2026 ($11M).
• A break below the 200-day moving average of $855.31 on company-specific news rather than sector beta.

Pre-registered FLIP conditions — what would take this to Buy — Tactical:

• The 2026-08-05 print delivering above $8.440B of revenue AND above $34.80 of EPS AND showing exabytes sold GROWING rather than flat. This is the single most likely path to a Buy and it resolves tomorrow. Volume growth is what converts a price spike into a franchise, and it is the one disclosure that would move our multiple assumption from 6.75x toward double digits.
• Price below approximately $1,050 — roughly 5.1x FY2027E and 15.7x FY2026E — with no deterioration in the disclosed fundamentals. At $1,050 our base of $1,390 is +32% and the risk/reward becomes genuinely asymmetric even on peak-cycle earnings. That is 26.4% below today's close and well inside the range this stock has traversed in a month.
• Evidence that the $6B buyback is actually being executed at prices near or below current levels. Today it is an authorisation, not a return. Executed at $1,400 it is 2.84% of the company retired at a peak-cycle multiple — which is either excellent or a mistake, and knowing which requires knowing the cycle.
• A material increase in contract liabilities beyond the $511M disclosed, evidencing more customers locking in multi-year supply — the strongest available filing evidence for the de-cyclicalisation thesis.
• Genuine name-level knowledge-base conviction accumulating. At present exactly ONE of 21 used claims is Sandisk-specific and it carries no named speaker. A second and third independent, named-speaker, Sandisk-specific claim would be worth roughly one tier.

Where SNDK fits in the Synthos Framework Portfolio. The memory and storage sleeve, at 0% today, with a 1.0-1.5% target — deliberately smaller than we would size a non-commodity name at similar conviction, because a beta of 5.19 means a 1.5% position carries the market risk of a much larger one. On the batch overlap question, which is acute here: SNDK, MU, WDC and STX are one trade with four expressions, and our own knowledge base makes the point for us — the claims in Section 7 almost never name Sandisk alone; they name Sandisk alongside Micron, Western Digital and SK Hynix. Holding more than one of these is concentration dressed as diversification. Within that group, Sandisk is the purest NAND expression and therefore the highest-beta one — no DRAM business to diversify the price cycle (the Nanya arrangement is a supply agreement and a locked-up minority stake, not a DRAM business), and no hard-disk business either. If the memory trade is to be owned at all, Sandisk is the maximum-torque way to own it, which is an argument for owning less of it, not more. Logged as a tracked Synthos call (Watch) as of 2026-08-04 at $1,427.62, with the 2026-08-05 print as the first graded checkpoint.

Single biggest risk: the price of NAND flash. Every dollar of the beat, every basis point of the 5,600 basis point margin expansion, and the entire FY2027 consensus rests on an average selling price per gigabyte that rose 248% year on year against flat bit volume. Sandisk does not set that price. It does not own the fabs that make the product. It takes a contractually fixed 50% of a joint venture's output and must fund half the fixed costs regardless. And its partner in that venture, Kioxia, is listed in its own 10-K as a competitor. At 6.93x forward earnings there is no valuation cushion in the conventional sense — because a 6.93x multiple on earnings that halve is a 13.9x multiple, and a 6.93x multiple on earnings that go to zero is not a multiple at all. That is the difference between a cheap compounder and a cheap cyclical, and it is why the risk score is a 9.

Most fragile assumption in the price: that FY2027 revenue reaches $48.670B — a 143% increase — with a 65.6% net margin. Every other assumption in Section 6a is either observable quarterly or resolves tomorrow. This one is a two-year extrapolation of a single quarter's pricing environment, made by 15 analysts who have collectively under-forecast this company by 26.0% on revenue and 60.1% on EPS in the most recent quarter — which tells you their models are chasing, not leading. The same modelling lag that produced four consecutive enormous beats will produce enormous misses when the price turns. If FY2027 revenue lands at $30B instead of $48.7B with a 50% net margin, EPS is roughly $101 rather than $206, and at 8x — a higher multiple than today's — that is $808 per share, 43% below the current price. That arithmetic, not a chart pattern, is why the bear case is $600 and why this is a Watch.

Provenance & disclosures

• EARNINGS TIMING — stated first because it governs the verdict. Sandisk reports fiscal fourth-quarter and full-year FY2026 results on 2026-08-05, ONE DAY after this dive was struck. The earn_cal entry carries epsEstimated: 34.80 and revenueEstimated: 8,439,742,000 with epsActual and revenueActual both null. The fiscal year being reported ended 2026-07-03 and was a 53-week year; no filing in this archive covers it. The verdict of Watch is set with that timing as an explicit input, and Section 11 pre-registers the three numbers in tomorrow's release that would change it.
• Traceability: 23 raw knowledge-base hits on the primary sweep (SNDK, Sandisk, SanDisk) across 51,928 distilled claims — 19 entity matches, 4 text-only, 21 used, 2 discarded. The two discards are both flagged quarantine_misattributed (2026-03-02, SUMMARY_OF_VISSER, bearish; 2026-05-23, NOT_VISSER, bullish) and are excluded from the conviction pool because the pipeline could not establish the speaker; the discard is directionally neutral, removing one bull and one bear. Breadth 11 distinct channels; 9 named speakers. Attribution concentration is the key finding: the jordi_visser channel supplies 5 of the 18 usable entity claims (27.8%) at a 2.0 skill weight, of which only ONE carries a named speaker, and two further claims sit in quarantine precisely because they were misattributed relative to that same voice; compound_and_friends supplies a further 5 entity claims plus 2 text-only. Two channels account for over half the lane. Exactly ONE of the 21 used claims (2026-01-18) carries Sandisk as its sole entity; every other claim baskets SNDK with Micron, Western Digital, SK Hynix or Samsung, and that is stated as a finding rather than dressed as name-level conviction. No claim in the lane carries a management speaker_role, so the half-weight management discount applies only to company statements quoted from the filings. Supplementary sweeps: NAND (0 entity / 11 text-only), flash memory (0 / 2), Kioxia (0 hits total — the sole manufacturing partner has no coverage), David Goeckeler (0 hits — no chief-executive coverage), Western Digital (3 entity / 2 text-only, 2 duplicating the primary set). Broad sweeps and their discard reasons: memory (7 entity / 508 text-only), HBM (234 text-only), DRAM (4 / 227), Micron (48 entity / 170 text-only), storage (1 / 153) — roughly 1,350 matches, essentially all discarded, because the great majority concern DRAM and high-bandwidth memory, which Sandisk does not manufacture, or concern Micron specifically, or collide on storage in an energy and biological sense (the single storage entity match is a 2019 biotech claim). A NAND-cycle or memory-cycle claim that does not name Sandisk is sector colour, not name-level conviction, and is treated as such throughout. All quotes are verbatim from the stored claim text, with one redaction noted in square brackets where a claim named an AI laboratory.
• Data as-of: annual statements — income, balance sheet and cash flow for fiscal 2025, year ended 2025-06-27, i.e. THIRTEEN MONTHS OLD and not containing fiscal 2026 · quarterly statements through 2026-04-03 · most recent filing-verified financials 2026-04-03, from the 10-Q filed 2026-05-01 · estimates 2026-08-04 · prices 2026-08-04, quote timestamp 1785873601 = 2026-08-04T20:00:01Z ($1,427.62, +10.84%; 50-DMA $1,704.83; 200-DMA $855.31; RSI 44.3; MACD −136.19) · knowledge-base claims 2026-08-04. Fiscal/calendar note: Sandisk's fiscal year ends on the Friday nearest 30 June. FY2025 ended 2025-06-27 (52 weeks); FY2026 ends 2026-07-03 and comprises 53 WEEKS, with the first quarter running 14 weeks. The est block dates every forward year to a 27 June, which does not match the company's actual fiscal calendar; the mismatch is immaterial to the arithmetic and is noted for accuracy. All figures come from the Synthos vendor data file for SNDK or from the SEC filings in the SNDK archive; no figure comes from memory, recall or external retrieval.
• Corporate-history note, load-bearing for every historical figure: Sandisk Corporation was spun out of Western Digital on 2025-02-21 and began trading on Nasdaq as "SNDK" on 2025-02-24. WDC distributed 116,035,464 shares (80.1%) at one-third of a Sandisk share per WDC share, retained 28,827,787 (19.9%), and has since disposed of all but 1,691,884 shares via debt-for-equity exchanges on 2025-06-09 and 2026-02-18. The 10-Q states that pre-separation financial information "was derived from WDC's consolidated financial statements... and prepared as if the Company existed on a standalone basis" — so fiscal 2022, 2023, 2024 and most of 2025 are CARVE-OUT ALLOCATIONS, not the reported results of this registrant. They are used in this dive for cyclical shape only and never as a track record. There is no price history for this security before 2025-02-24 and none is used. (A drafting note in the 10-K cover page renders the ticker as "SDNK"; the correct symbol, used consistently elsewhere in the same filing and in the 10-Qs, is SNDK.)
• Filing archive contents and its gaps: 10-K filed 2025-08-21 (fiscal 2025, year ended 2025-06-27); 10-Q filed 2026-01-30 (quarter ended 2026-01-02); 10-Q filed 2026-05-01 (quarter ended 2026-04-03); 8-K filed 2026-03-25 (Nanya private placement and DRAM supply arrangement); 8-K filed 2026-04-30 (Q3 results announcement plus the $6B repurchase authorisation); 8-K filed 2026-05-15 (rejection of an unsolicited mini-tender by Tutanota LLC at $1,150.00 per share for up to 100,000 shares). All three 8-Ks carry substantive Item text rather than bare cover pages — better than the norm and recorded as such — but Exhibit 99.1 is ABSENT in each case, so the 2026-04-30 earnings 8-K contains no financial figures whatsoever. Numeric tables are stripped from the extracted filing text throughout, which is why the Flash Ventures lease-guarantee amounts, the customer-concentration table in the 10-Q and the balance-sheet detail at 2026-04-03 are stated as disclosures rather than as figures. We state those limits rather than estimating around them. No filing in this archive covers fiscal 2026.
• Where the filings contradicted or corrected the vendor (detailed in Section 8): bal_a's $1.829B of long-term debt and $561M of net debt REJECTED — the 10-Q states the $2.0B Term Loan Facility was "settled in full" on 2026-03-04 with a $46M extinguishment charge and the $1.5B revolver was undrawn at 2026-04-03, so the company is in a NET CASH position of $3.528B; using bal_a would have overstated enterprise value by $4.089B. The $1.8B goodwill impairment is classified inconsistently between the vendor's annual statement (in operating expenses — correct) and its quarterly statement (in other income — wrong); the 10-K states it was recorded in the Consolidated Statements of Operations and the 10-Q lists it among operating expenses; we use the annual classification and report FY2025 operating margin on both bases (−18.7% reported, +6.6% ex-impairment). est.ebitdaAvg and est.ebitAvg REJECTED IN FULL as fabricated — exactly 14.44% and 10.78% of revenue in every one of five forward years, and netIncomeAvg exceeds ebitdaAvg in every year, which is arithmetically impossible. The FY2030 estimate row REJECTED (revenue collapsing 77.8% to $11.566B on ONE analyst) and FY2029 excluded (2 analysts). Three seg_geo rows (fiscal 2011, 2014 and 2015) REJECTED as belonging to the FORMER SanDisk Corporation, a different registrant acquired by Western Digital in 2016 — December fiscal year-ends and regional labels that appear in no Sandisk Corporation filing. profile.ipoDate of 1995-11-08 REJECTED for the same reason; this company listed on 2025-02-24. seg_prod labels corrected from the 10-K's "Cloud/Client/Consumer" to the 10-Q's "Datacenter/Edge/Consumer". quote.yearHigh/yearLow ($2,354.39/$40.53) rejected in favour of tech.hi52/lo52 ($2,335.00/$40.69) for internal consistency. The vendor peer set — Credo, Flex, HPE, Jabil, Keysight, MongoDB, Super Micro, Teledyne, Teradyne, Wipro — DISCARDED IN FULL: not one is a memory manufacturer, and no peer-multiple comparison is drawn anywhere in this dive. Where vendor and filing AGREED — worth recording: the share count (10-Q cover 148,089,758 at 2026-04-24 against 148,089,750 implied by market cap ÷ price — agreement to 0.000005%); the 5,600 basis point year-on-year gross-margin expansion in the April quarter (vendor quarterly statements reproduce it exactly); seg_prod and seg_geo each summing to $7,355M against reported FY2025 revenue of $7,355M; km_ttm.netDebtTTM and enterpriseValueTTM, which our independent rebuild reproduced to the dollar; and the entire ratios_ttm margin, tax-rate, coverage and income-quality block, every element of which we recomputed from raw lines and matched.
• Financings post-dating every statement in the file — hunted for specifically and reported in full: (1) the $2.0B term loan repaid in full on 2026-03-04; (2) a $972 million equity investment in Nanya Technology Corporation, agreed 2026-03-25, paid 2026-04-08 — five days after the last balance-sheet date — for 138,685,000 shares (~3.9% of Nanya fully diluted) subject to a three-year Taiwanese statutory lock-up, alongside a multi-year DRAM supply arrangement; (3) a $6 billion share repurchase authorisation approved 2026-04-30, equal to 2.84% of market capitalisation, with no evidence in this file that any shares have been repurchased. We searched all five filings specifically for convertible, capped call, senior notes, aggregate principal, term loan and revolving credit: there are NO convertible notes, NO capped call transactions and NO senior unsecured notes. That verified negative is reported as a finding.
• Off-balance-sheet items in no vendor field (defect class 10), hunted in the filings: Sandisk guarantees half of all outstanding obligations under Flash Ventures' Japanese equipment lease facilities, with a cancellation event capable of accelerating them — the dollar amount sits in a table the extraction stripped and is NOT estimated here; $1.2 billion payable directly to Kioxia across 2026-2029 under the January 2026 Agreement to Enhance Collaboration; $750 million of building-depreciation prepayments in Other non-current assets plus $387 million of further committed prepayments and approximately $155 million of additional building depreciation payments (January 10-Q); an obligation to fund 49.9%-50.0% of each Flash Ventures entity's capital investment where JV operating cash flow is insufficient; an obligation to pay half of Flash Ventures' fixed costs regardless of output purchased; $253 million of unrecognised tax benefits, of which ~$244 million could become cash; and a $131 million tax indemnification liability to WDC under the Tax Matters Agreement.
• Basis note: analyst estimates in est and the actuals in earn_cal are non-GAAP; the income statements are GAAP. The gap is unusually small — April-quarter non-GAAP EPS of $23.41 against GAAP diluted $23.03, a 1.6% difference, entirely explained by the $46M debt-extinguishment charge. All forward multiples in this dive are struck on consensus NON-GAAP EPS; the trailing 49.62x is GAAP diluted. No conclusion turns on the choice.
• Estimate coverage: 15 analysts on FY2026 revenue and 11 on FY2026 EPS; 15 and 12 on FY2027; 12 and only 5 on FY2028 (with a FY2028 EPS high-to-low spread of $490.46 to $160.39 — a 3.06x range). FY2029 rests on 2 analysts and FY2030 on ONE; both are excluded from every conclusion. The FY2028 EPS row is used as a cross-check only, never as an anchor.
• Fair-value caveat: the $600 / $1,390 / $2,475 anchors are stated arithmetic — a named multiple applied to a named estimate line — not a discounted cash flow. Bear: 9x the FY2026 consensus EPS of $66.874, cross-checked at 10x tangible book value per share of $59.34 = $593. Base: 6.75x the FY2027 consensus EPS of $206.124, cross-checked at 10x the annualised Q4 FY2026 consensus run-rate of $139.20 = $1,392. Bull: 12x the FY2027 consensus EPS of $206.124, cross-checked at 10x the FY2028 consensus EPS of $247.562 = $2,476. Base sensitivity disclosed openly: 6x gives $1,237, 7x gives $1,443, 8x gives $1,649 — the whole answer lives in a 6-8x band and no greater precision is claimed. The base assumes the multiple HOLDS rather than expanding or compressing, and 100% of the expected return is earnings delivery — there is no dividend, the buyback is unexecuted, and diluted shares are rising, so shareholder yield is 0.00%. The bull is 4.1x the bear, and that width is the honest description of a commodity producer at a price extreme one day before a print.
• Peer note: the vendor peer set contains no memory manufacturer and is discarded in full. The genuine comparables named in Sandisk's own 10-K — Kioxia, Micron Technology, Samsung Electronics, SK Hynix and Yangtze Memory Technologies — are absent from the file, and no peer multiples are supplied for anyone, so no peer-multiple comparison is drawn. The only cross-name context used is the batch move on 2026-08-04.
• Timing: fiscal fourth-quarter and full-year FY2026 results are released 2026-08-05 — ONE DAY after this dive. The last reported quarter ended 2026-04-03 and was announced 2026-04-30. 2026-08-04 was a violent sector-wide semiconductor and memory up-day (MRVL +12.8%, SNDK +10.8%, INTC +10.9%, LRCX +7.9%, MU +7.6%, KLAC +7.0%, AMD +7.0%, AMAT +5.5%, ASML +4.2%); SNDK's 10.84% move carried no company-specific news in this file and is treated as sector beta. The gap is real, not a data error, and the question of whether the reader should pay up for it is answered explicitly in Section 9: no.
• Accessibility note: no information in this dive is conveyed by colour. All emphasis is carried by bold text, table structure and explicit labelling. The earnings warning at the top is a labelled text block, not a coloured banner.
• Not investment advice. Independent research, educational and informational only, never personalised. No recommendation to buy, sell or hold any security is made to any person.
• Version: 2026-08-04-full.