Phase Space AI

SanDisk Valuation Analysis 2026-07-27

SanDisk [SNDK]

SanDisk Corporation [SNDK] — Valuation Analysis

Task 3 · investment-memo skill v1.4.0 · 2026-07-27 · price $1,278.23 · model of record: SanDisk_Financial_Model_2026-07-27.xlsx

---

2026-07-29 — METHODOLOGY MIGRATION AND DATA CORRECTION (supersedes parts of this document)

Written 2026-07-27 under the retired six-Gate / E[R]-vs-cash-hurdle framework; migrated onto the Criteria framework. The analysis below is retained as evidence; the following supersede it where they conflict. The authoritative valuation output for this name is now §3A/§3B of SanDisk_Trade_Construction_2026-07-27.md.

The data correction — this is the deliverable

	OLD	NEW
Revenue basis used by the screen / scan / rescore path	last fiscal year, FY2025 (ended 2025-06-27), 397 days stale	TTM, 4 quarters to 2026-04-03
Revenue	$7.355bn	$12.978bn — last-FY understated it by 76%
EV / Sales	21.6x (EV ÷ last-FY revenue)	12.22x
Demonstrated revenue CAGR	9.9% (annual-FY basis, same stale endpoint)	31.6% (TTM-consistent, FY2023 → TTM)
Implied-path margin	−34.2pp (rescore, on a hardcoded 20% terminal margin)	−4.8pp
Valuation Criteria	FAIL by 34 points	FAIL by 5 points

This document itself states no EV/Sales figure, so its headline conclusions were not directly contaminated — the last-FY figure entered through the screen / scan / rescore path, where it set the demonstrated growth rate and the multiple that drove the scoring. That path is what rejected the name, and it rejected it by roughly seven times the margin the corrected data supports. Corrected inputs come from reports/scan/SNDK_analysis.json and are used rather than recomputed.

What else is superseded

• The Valuation Criteria / E[R] line in §1 (−71.7% gross, against a 4.7% cash hurdle) is retired, along with the cash hurdle itself. Sensitivity now runs over the exit multiple, never over scenario probabilities.
• The house 12-month price target of $566.60 (−55.7%) is demoted. It is a DCF-and-scenario construction, not a multiple built on SanDisk's own trading history. Under references/valuation.md the 12-month target must be anchored on the name's own range — and SanDisk has only 123 daily observations of a trailing EV/Sales multiple (from 2026-01-30), spanning a period in which trailing revenue rose from $8.9bn to $13.0bn. The anchor is therefore UNIDENTIFIED and the 12-month target is reported as INDETERMINATE, with $566.60 retained as the best available figure and flagged. It is not replaced with a peer median.
• The implied-path test replaces the DCF as the primary long-horizon output, because terminal value far exceeds 60% of EV (SanDisk earns nothing on the filed FY2025).
• Price is now $1,095.78 (2026-07-28, scan), not the $1,278.23 used below.

Not superseded — retained and still load-bearing

The two-tranche contracted/merchant build, the RPO analysis (Note 4, $41.6bn), the NAND contract-price second-derivative work, the Flash Ventures fixed-cost argument, and the base-rate discipline are all retained. Nothing about the evidence weakened; what changed is the revenue denominator and the instrument used to test the price.

---

1. Summary

Every valuation approach in the workbook lands between $247 and $787. The current price is above all of them. The Street's average target is 2.9× the top of the house range.

---

2. Method 1 — Two-tranche DCF (the regime-change implementation)

references/regime-change-test.md Principle 4. Revenue is split into a contracted tranche modelled at contracted terms for exactly as long as the contracts run, and a merchant tranche that mean-reverts. The terminal value reverts entirely, because nothing evidences the post-contract period.

2.1 The contracted tranche (Tier 1)

2.2 The merchant tranche (mean-reverting)

Bits +15% p.a. (company guidance, via node transitions). ASP and margin are the scenario variables:

Base-rate check, stated explicitly. SanDisk's full-year gross margins were 7.1% (FY2023), 16.1% (FY2024), 30.1% (FY2025) — best-ever 30.1%. Micron's all-time-peak annual gross margin was 58.9%. Every merchant margin above ~45% in the table above is an above-base-rate assumption and is labelled as one. Even the Bear case's terminal blended margin of 35.2% sits above SanDisk's own best-ever full year.

2.3 Resulting Base-case model

The contracted share peaks at 36% in FY2029 and then declines as contracts roll off. Per Principle 4, that decline is the honest expression of uncertainty — a single blended assumption would hide it.

2.4 Discount rate

A 15.7% WACC is punitive. It is disclosed as such and the model is run across the full range, because it would be dishonest to let the discount rate carry the conclusion:

Even at an 8% cost of capital — implausibly low for a commodity semiconductor with 109% realised volatility — the Base case supports only 58% of the current price. That is the point: the load-bearing disagreement is the terminal margin, not the discount rate.

2.5 DCF sensitivity — value per share

Every cell is below $520. The current price is $1,278.23.

2.6 Reverse DCF — what $1,278.23 actually requires

Solving for the terminal assumption that would justify today's price, holding the Base FY2030 revenue base:

• At the base-case 15.7% WACC: a through-cycle gross margin of 183% — arithmetically impossible.
• At a 10% WACC: 111% — also impossible.

Because a margin solution does not exist, the price must instead be requiring a much larger terminal revenue base. At the base-case WACC and a 40% through-cycle margin, today's price requires terminal-year revenue of roughly $128bn — against a Base FY2030 of $27.9bn, an FY2025 actual of $7.4bn, and a whole worldwide NAND industry running at roughly $70–90bn at today's inflated prices.

Stated plainly: there is no combination of terminal margin and terminal revenue consistent with industry capacity that reaches $1,278 in a reverting model. The price is only reachable if the reversion does not happen at all.

---

3. Method 2 — Normalised earnings, plus the windfall

A discount-rate-independent cross-check.

Adding the windfall matters and is easy to omit: a through-cycle multiple values the steady-state business only, but the cash the peak actually throws off is real and lands on the balance sheet. Omitting it is the most common way a cyclical is under-valued at the top. It adds ~$98/share here.

Historical memory-cycle multiples on mid-cycle earnings have been roughly 8–14×. 15× is generous and credits the contracted tranche and the datacenter mix.

---

4. Method 3 — Comparables, and the value-trap signature

This table is the argument. The two companies running all-time-record gross margins — SanDisk at 78.4% and Micron at 84.6% — carry the lowest multiples on annualised earnings, at 9–10×, while peers whose margins have not inflected trade at 57–119×. That inversion is not a discount for quality. It is the market marking earnings as unsustainable. A single-digit P/E on a record-margin quarter is the textbook signature of peak-cycle earnings, not of cheapness — and it is exactly the trap the brief asked to be tested.

Cross-check on sales: SanDisk's EV of ~$185.6bn against an annualised Q4 FY2026 revenue of ~$32bn is 5.8×. Memory historically trades at 1.5–3× sales.

---

5. Triangulation and the house target

A single DCF at a 2.19-beta discount rate would produce a more extreme number than the evidence supports. The target is deliberately a blend.

---

6. Scenarios (the exact probabilities used in Valuation Criteria)

Probability-weighted value $361.64; scenario-weighted expected return −71.7% (DCF strand) or −55.7% (blended target). Bear is weighted above Base's complement because the reversion evidence is dated and quantified while the regime-change evidence is one quarter old.

Principle 5 check. A genuine contractual regime change should raise the bear case materially and the bull case little or not at all. Here the NBM floor does exactly that — the Bear case's 32% terminal margin is above SanDisk's own best-ever full year, which is the floor doing its job. It does not support the ceiling. The shape is consistent with a real but partial Tier 1 floor, not with a regime change.

---

7. Consensus bridge — required, because the house view diverges violently

7.1 The Street

7.2 Decomposition: is the gap about the numbers or the multiple? Both — and mostly the numbers.

This is unusual and worth stating clearly, because in most of this coverage the gap has been a multiple disagreement. Here it is not.

The numbers disagreement is the larger part. House FY2027 revenue of $31.1bn is roughly the Q4 FY2026 exit run-rate ($32bn annualised) held flat, with bits +15% and therefore ASPs declining ~13% through the year. Consensus at $41bn requires ASPs to rise again from the Q4 exit rate on top of mid-teens bit growth. That single assumption — the direction of ASPs through FY2027 — is the variant, and it is testable.

The multiple disagreement is secondary but real. Even accepting the Street's FY2027 EPS of $177, the house would not pay 12.4× for it, because FY2027 is the peak year: the house FY2029E is $45.93, and 12.4× that is $570 — almost exactly the house target. The two sides are not far apart on the multiple; they are far apart on which year's earnings deserve it.

7.3 What if the Street is right?

Named honestly. If NAND ASPs rise again through FY2027 and the NBM contracted share exceeds 50% of bits at re-struck (higher) floors, FY2027 EPS of $177–208 is achievable and the terminal question is deferred by a year or more. In that world the stock does not need the terminal to be resolved and can trade to $2,000+ on momentum and buyback alone. The house has no evidence against FY2027 being strong — the evidence (TrendForce) points to the rebalance arriving in 2H calendar 2027, which is SanDisk's FY2028. The house is early by construction, and says so.

7.4 Positioning

The inverted skew is a material fact for anyone considering a short: the options market is pricing a right tail, not a left one.

---

8. Factor & Anomaly Scorecard (required)

Direction assessed for a bearish/short view, since that is the direction the fundamental work points.

Synthesis — and this is uncomfortable, which is why it is stated rather than buried. The anomaly evidence runs strongly against the fundamental view. Momentum, PEAD, revisions, profitability, accruals, asset growth and F-score all point the wrong way for a short. references/trade-construction.md is explicit: "A Short whose scorecard shows clean earnings quality and strong profitability needs correspondingly stronger narrative evidence to pass Quality Criteria, and the memo must say so explicitly rather than ignoring the disagreement."

It is said. The narrative evidence offered against it is: (i) TrendForce's dated, quantified supply/demand reversal for CY2027; (ii) ~$40bn of NAND conversion WFE already pulled forward; (iii) the second derivative of NAND contract prices having already rolled from +72% to +12.5% q/q; (iv) flat bit shipments, meaning the entire earnings base is price; and (v) the mention-frequency finding that management's own capacity language collapsed as margin peaked. That is real evidence. It is not, however, evidence that FY2027 numbers will miss — and the scorecard is measuring exactly the horizon over which the house agrees the business is strong.

The honest resolution: the anomaly scorecard is right about the next two to three quarters and the fundamental work is about FY2028 and the terminal. That timing mismatch is itself the argument against acting now, and it is what drives the Task 5 decision.

---

Model of record: SanDisk_Financial_Model_2026-07-27.xlsx, verified by reading back computed cells from Excel via AppleScript after every edit; the balance sheet ties to zero in all forecast years. Price and options data: Alpaca (SIP feed), 2026-07-27. Fundamentals: SEC EDGAR XBRL, CIK 0002023554. Consensus: public aggregators (S&P Global, Simply Wall St, Yahoo Finance, published Goldman Sachs commentary), July 2026 — flagged where sourcing is a single aggregator.