ScorecardAugust 26, 20269 min read
The Memory Trade Round-Tripped in Six Sessions. My Own Floor Broke.
On August 17 I published a projection saying that if the entire reshoring premium came out of SanDisk, the price landed at $1,528. Six sessions later it closed at $1,480.77 — 3.1% through that floor. This is what happened, why the floor was the wrong shape, and what I am changing because of it.
Nine days ago SanDisk closed at $1,786.85, up 16.9% in three sessions on a reported policy signal that Washington preferred domestic and allied memory suppliers. I wrote a projection that afternoon putting a floor under the move: strip out the entire policy premium and the price returns to $1,528.11. On Tuesday it closed at $1,480.77. The floor did not hold, and the reason it did not hold is more interesting than the miss.
01What I published, and what happened
The August 17 market wrap carried a projection table headed “how much of SanDisk’s run is the policy signal.” It took the three-session, 16.9% move from August 13 to August 17 and asked what the price would be if various fractions of it were given back. The bottom row — all of it — produced $1,528.11, and I wrote that this “establishes a floor for the argument.”
That was the wrong word, and I want to be precise about why. The arithmetic was not wrong: $1,528.11 is exactly where the price sits if you remove a 16.9% gain. What was wrong was calling it a floor, because the calculation contained a hidden assumption I did not state — that August 13 was a neutral starting point. It was not. August 13 was already inside a rally.
The move gave back more than it made, in half the time it took to make it
Reading this figure
- The run: $1,528.11 to $1,786.85 across three sessions, +16.93%.
- The projection floor I published on August 17 was the left edge of that shaded window.
- Tuesday’s close of $1,480.77 is 3.1% below that line.
- Six sessions down against three sessions up — and a deeper move.
02Why a round trip overshoots
A policy signal that is never written down does not decay to zero. It decays through zero, because what the market removes is not just the premium — it is the credibility of the class of argument that produced it. Here is the sequence, session by session.
-
A signal arrives with no document behind it
Reports said the administration signaled a preference for domestic and allied memory suppliers. No rule was published, no tariff schedule changed, no procurement order was filed. The market priced an expectation of policy, not policy.
Aug 17: SNDK +8.88%, MU +4.13% -
Rates reprice the whole long-duration complex
The following session, debt-financed AI infrastructure sold off hard as the market marked down anything whose value sits in distant cash flows. Memory went with it — not because the reshoring story changed, but because the discount rate applied to it did.
Aug 18: SNDK −9.01% -
The signal fails to produce a second data point
Days pass. Nothing is published. In the absence of confirmation, a position held because of an expectation becomes a position held on hope, and those get sold first when anything else goes wrong.
Aug 19–21: SNDK −1.7% net -
A risk-off session finds the weakest holders
Monday sorted the entire tape by speculativeness. Memory, which had the newest and least-documented story attached to it, was sold alongside quantum computing and small-cap space names rather than alongside its own sector.
Aug 24: SNDK −6.45%, MU −5.83% -
The price settles below where the story started
Which is the overshoot. The market has not merely removed the reshoring premium; it has discounted the base a little further, because the episode is now evidence that this name moves on unconfirmed narrative.
Aug 25: $1,480.77, 3.1% through the floor
A floor built by subtracting a rally assumes the level underneath the rally was fair. That is the assumption I did not write down, and it is the one that broke.
The specific error in the August 17 projection
03The two memory names did not move together
The other thing the round trip exposed is that SanDisk and Micron are not the same trade, which is easy to forget when they rise on the same headline. Indexed from August 13, the two separate materially after the peak.
Micron gave back the rally. SanDisk gave back more than the rally.
Reading this figure
- Both peak on August 17, on the same reported catalyst.
- Micron ends the window at 98.2 — slightly below its pre-rally base.
- SanDisk ends at 96.9, a materially deeper give-back.
- The gap is the part that is about SanDisk specifically, not about memory.
The mechanical explanation is position size and story dependence. SanDisk had two catalysts stacked in three days — an investor day and a 10-K — on top of the policy signal, so it had more narrative to unwind. Micron had one. And SanDisk is the larger, more crowded position in this book, which means it had further to fall before finding holders who owned it for the balance sheet rather than the headline.
Live chart — the names in this piece
TradingView, three-month daily. Switch symbols with the tabs.
Live data from TradingView, independent of the figures above — those are fixed to the August 13–25 window and will not change. Use this panel to see what has happened since publication. Charts are for context, not recommendations.
04What the underlying bet still requires
Price aside, the thing I actually own SanDisk for has not changed: management’s fiscal 2030 model calls for roughly 80% gross margin, and fiscal 2026 delivered 71.5% on $20.25 billion of revenue against $5.78 billion of cost of revenue. The August 17 piece made this point with a static table. It is better as something you can push on, because the whole argument turns on the relationship between two growth rates.
What the fiscal 2030 margin target needs from four more years
Drag either input. Everything below recomputes from SanDisk’s reported fiscal 2026 base of $20.25B revenue and $5.78B cost of revenue.
Compounded over four years to fiscal 2030.
Capacity additions push this up, not down.
Assumptions, stated plainly. Both inputs compound for exactly four years from the reported fiscal 2026 base, with no other change to the business — no acquisitions, no write-downs, no shift in product mix. The model compares a GAAP margin to management’s non-GAAP target, which are not directly comparable and which the original 10-K coverage flagged. Nothing here is a forecast: the default position of the sliders is not a prediction, it is a starting point. What the model is for is one structural fact you can verify by dragging it — at any cost growth above roughly 4% a year, there is no revenue growth rate in the plausible range that reaches 80%. The target is a bet on the price of memory, not on operating leverage.
05The full week, sortable
Every position in the book across the four sessions from the August 19 close to the August 25 close. Cells are shaded by the size of the move; the columns sort.
| Position | Aug 25 close | Thu 20 | Fri 21 | Mon 24 | Tue 25 | Week | vs cost |
|---|---|---|---|---|---|---|---|
| Rocket Lab USA RKLB | $66.91 | −3.81% | −0.52% | −5.91% | −2.01% | −11.77% | −15% |
| CrowdStrike Holdings CRWD | $185.38 | −5.60% | +0.85% | −0.66% | −2.78% | −8.06% | +96% |
| Intel Corporation INTC | $87.48 | −0.72% | −2.24% | −3.12% | +0.25% | −5.73% | +134% |
| Sandisk Corp SNDK | $1,480.77 | +2.02% | −0.28% | −6.45% | −0.83% | −5.62% | +206% |
| Palo Alto Networks PANW | $339.90 | −2.84% | +2.38% | −1.95% | −3.13% | −5.52% | +112% |
| WhiteFiber Inc WYFI | $20.40 | −1.03% | −2.08% | +0.29% | −1.83% | −4.58% | +53% |
| Talen Energy Corporation TLN | $307.67 | −1.66% | −0.80% | −2.84% | +0.70% | −4.55% | −2% |
| Datadog Inc DDOG | $222.99 | −0.42% | +1.32% | −4.18% | −1.23% | −4.51% | +87% |
| Symbotic Inc SYM | $40.41 | −1.33% | +0.89% | −4.30% | +0.95% | −3.83% | +5% |
| Arm Holdings plc ARM | $241.56 | +0.55% | −2.95% | −1.87% | +1.16% | −3.12% | +16% |
| CoreWeave Inc CRWV | $88.04 | −1.22% | −2.13% | −1.82% | +2.08% | −3.11% | +6% |
| Vistra Corp VST | $139.03 | −2.63% | −1.96% | −0.40% | +2.48% | −2.57% | +3% |
| GraniteShares 2x Long NBIS Daily ETF NBIL | $24.65 | −3.29% | −1.07% | −7.49% | +10.34% | −2.34% | +167% |
| Vertiv Holdings Co VRT | $255.75 | +1.39% | −1.01% | −2.66% | +0.31% | −2.01% | −16% |
| Broadcom Inc AVGO | $356.74 | +0.43% | +1.21% | −2.63% | −0.56% | −1.58% | +22% |
| Space Exploration Technologies (SpaceX) SPCX | $137.95 | −4.05% | +2.22% | −1.44% | +2.19% | −1.22% | +2% |
| Nebius Group N.V. NBIS | $221.97 | −1.69% | −0.45% | −3.75% | +5.24% | −0.86% | +146% |
| Invesco QQQ Trust QQQ | $710.72 | −0.72% | +0.35% | −1.00% | +0.62% | −0.75% | +27% |
| Micron Technology MU | $932.97 | +3.97% | −0.77% | −5.83% | +2.48% | −0.44% | +186% |
| SPDR S&P 500 ETF SPY | $765.91 | −0.84% | +0.41% | −0.29% | +0.32% | −0.41% | +17% |
| Vanguard S&P 500 ETF VOO | $704.02 | −0.83% | +0.39% | −0.27% | +0.31% | −0.41% | +21% |
| ServiceNow Inc NOW | $127.00 | +2.00% | −0.98% | −0.33% | −0.82% | −0.16% | +50% |
| Schwab U.S. Dividend Equity ETF SCHD | $35.11 | −0.74% | +0.80% | +0.28% | −0.28% | +0.06% | +9% |
| Fortinet Inc FTNT | $153.66 | −1.35% | +1.80% | −0.99% | +1.10% | +0.53% | +82% |
| American Airlines Group AAL | $13.95 | −2.45% | +2.22% | −1.37% | +2.35% | +0.65% | +37% |
| Alphabet Inc GOOGL | $346.96 | −1.17% | +1.22% | +0.94% | −0.32% | +0.65% | +26% |
| Netflix Inc NFLX | $82.23 | −0.10% | −0.69% | +0.53% | +2.77% | +2.51% | +4% |
| Advanced Micro Devices AMD | $479.18 | +0.65% | +0.81% | −3.49% | +4.91% | +2.74% | +140% |
| Meta Platforms META | $570.05 | −0.04% | +0.75% | +1.66% | +1.97% | +4.40% | +143% |
| Estée Lauder Companies EL | $104.85 | −1.90% | +6.02% | +2.15% | +0.69% | +6.98% | +52% |
| The Cheesecake Factory CAKE | $113.48 | +0.67% | +6.38% | +1.13% | −0.98% | +7.24% | +150% |
| Tempus AI Inc TEM | $68.73 | +8.82% | +9.06% | −8.97% | +3.87% | +12.21% | +38% |
Source: Massive Market Data daily bars. “vs cost” compares the August 25 close to my average cost basis, from the holdings file. Aug 22–23 was a weekend.
06What I am changing
- The projection
- $1,528.11Published Aug 17 as the “premium fully gone” level
- The outcome
- $1,480.77Aug 25 close, 3.1% below it
- Verdict
- Arithmetic right,
framing wrongThe number was correct; calling it a floor was not - Position change
- NoneThe Holding Period and Blackout Rule both apply
Three things change in how I write these, starting with this piece.
One: no more floors. A give-back calculation tells you where a price returns to if a specific gain reverses. It says nothing about whether the level underneath that gain was itself fair, and it cannot be a floor unless I have independently established that the base was. I will label these as give-back levels, which is what they are.
Two: unconfirmed catalysts get labelled as such in the figure, not just the prose. The August 17 chart plotted the rally without marking that its cause was a report of a signal rather than a published rule. That distinction belonged on the chart.
Three: I will score projections in public on a fixed schedule rather than when they happen to be interesting. A projection nobody checks is decoration. This piece exists because the number went against me; the discipline is worth nothing if it only runs in that case.
Methodology and sourcing
Every price in this article is a daily closing price from Massive Market Data grouped daily bars, for the sessions of August 13, 14, 17, 18, 19, 20, 21, 24 and 25, 2026. August 22 and 23 were a weekend.
Book impact is share count multiplied by the change in closing price. Share counts and average cost come from this site’s holdings file, which is the same source the Portfolio page uses.
The fiscal 2026 revenue and cost-of-revenue figures ($20.248B and $5.776B) are from SanDisk’s Form 10-K for the fiscal year ended July 3, 2026, filed August 17, 2026. The roughly 80% fiscal 2030 gross-margin figure is management’s stated long-term model, presented at its investor day, and is non-GAAP.
The reshoring reports are as covered by The Motley Fool on August 17, 2026. No rule, order or tariff schedule has been published as of this writing; that is the point of section 02.
The TradingView panel is live third-party data and is not used for any figure in this article.
This article is educational and reflects my own analysis. It is not investment advice. Prices and yields via Massive Market Data; filing figures as cited. Live chart data by TradingView.