August 2026 · Interim Note
Memory Had a Reason to Rally; My Software Names Didn’t Have One to Fall
Monday was a rotation inside my own book, not a uniform move in either direction. SanDisk (SNDK) rose 8.88% on its fiscal 2026 10-K plus a real, dated catalyst: the Trump administration signaled it wants Apple sourcing memory chips from domestic or allied suppliers rather than China, and Micron (MU) picked up 4.13% on the same news — alongside Elon Musk publicly arguing that memory, not compute, is now the binding constraint on agentic AI. SpaceX (SPCX) added 4.45% on what one report called “positive analyst notes,” consistent with its exclusive Nvidia commitment for its AI buildout. Intel (INTC) and Rocket Lab (RKLB) also closed higher.
Monday was a rotation inside my own book rather than a move in either direction. Memory rallied on a dated catalyst, rate-sensitive software fell without one, and the two sleeves finished the session more than fourteen percentage points apart. 7 of 32 positions closed higher. The net effect on the book was a loss of $68. No trades.
The rotation, position by position
Figure 1 · Friday to Monday
The book pulled apart in a single session
The nine positions that moved most, each indexed to its own Friday, August 14 close (hollow dot at 0%) and plotted to its Monday, August 17 close (filled dot). Indexing rather than plotting dollars is what makes a $1,787 stock and a $118 stock comparable on one axis. Source: Massive Market Data daily bars.
SanDisk rose 8.88% on its fiscal 2026 10-K plus a real, dated catalyst — the Trump administration signaling it wants Apple sourcing memory from domestic or allied suppliers rather than China — and Micron picked up 4.13% on the same news, alongside Musk arguing that memory is now the binding constraint on agentic AI. SpaceX added 4.45% on what one report called positive analyst notes. Intel and Rocket Lab also closed higher.
The other side of the book had no such story. ServiceNow fell 5.08%, Datadog 3.19%, Nebius 3.18% and Meta 3.54% amid what one source called broader communications-sector weakness — real language, but not a company-specific reason. That reads more like a rate-sensitive basket getting sold on a day the 30-year yield hit its highest level since 2007 than like any single thesis breaking. Full session in today’s News writeup; my take on the timing in today’s Opinion piece.
| Position | Close | Day | Book impact | vs. my cost |
|---|---|---|---|---|
| Sandisk Corp (SNDK) | $1,786.85 | +8.88% | +$291 | +270% |
| Micron Technology (MU) | $1,011.75 | +4.13% | +$80 | +211% |
| Space Exploration Technologies (SpaceX) (SPCX) | $146.23 | +4.45% | +$50 | +8% |
| Intel Corporation (INTC) | $103.49 | +0.97% | +$14 | +176% |
| Fortinet Inc (FTNT) | $155.85 | −2.60% | −$12 | +85% |
| Arm Holdings plc (ARM) | $271.43 | −2.87% | −$16 | +31% |
| Talen Energy Corporation (TLN) | $356.92 | −1.60% | −$17 | +13% |
| Nebius Group N.V. (NBIS) | $268.85 | −3.18% | −$18 | +198% |
| Datadog Inc (DDOG) | $247.32 | −3.19% | −$24 | +107% |
| Estée Lauder Companies (EL) | $84.34 | −2.04% | −$32 | +22% |
| The Cheesecake Factory (CAKE) | $109.07 | −3.80% | −$34 | +140% |
| ServiceNow Inc (NOW) | $117.70 | −5.08% | −$38 | +39% |
| Advanced Micro Devices (AMD) | $506.00 | −1.63% | −$42 | +154% |
| Palo Alto Networks (PANW) | $375.76 | −2.21% | −$43 | +134% |
| GraniteShares 2x Long NBIS Daily ETF (NBIL) | $37.09 | −6.43% | −$76 | +302% |
| Meta Platforms (META) | $568.97 | −3.54% | −$104 | +143% |
A projection: what happens if the divergence keeps up
A single rotation session is noise. The reason to write it down is that this book is deliberately split across both sleeves, so a sustained divergence is the thing that would actually force a decision. Here is what Monday’s two sleeve averages compound to if they simply repeat.
| If Monday repeats for | Memory & chips | Software & security | Spread |
|---|---|---|---|
| One more session (1 sessions) | +4.88% | −2.82% | 7.7pp |
| One week (5 sessions) | +26.90% | −13.32% | 40.2pp |
| Two weeks (10 sessions) | +61.03% | −24.87% | 85.9pp |
| One month (21 sessions) | +171.94% | −45.14% | 217.1pp |
Assumptions, stated plainly. This compounds one session's sleeve averages forward, which is not how returns behave — daily moves do not repeat, and a divergence of this size would attract mean reversion long before the one-month row. Sleeve membership is my own classification of my own positions, and the averages are equal-weighted rather than position-weighted, so they describe the names rather than the money. It is not a forecast. It is a threshold-finding exercise: at a week of this, the spread is wide enough that my Rebalancing Rule would be the thing making the decision rather than my judgement, which is precisely why the rule is written down in advance in the Written Policy.
No trades today
I’m not trimming rate-sensitive names into an unexplained air pocket, and I’m not adding to memory on one strong day either. The Blackout Rule and Holding Period exist for exactly this kind of session. The next two sessions turned out to matter a good deal more — those are in Tuesday’s note.