August 2026 · Interim Note
CoreWeave Fell 12%, My Leveraged Nebius Position Fell 15%, and I Made Zero Trades
Tuesday was the roughest session I’ve logged since mid-July, and unlike that one it landed almost entirely inside the sleeve of my book I’m most exposed to. The S&P 500 fell 0.68%, the Nasdaq-100 fell 1.69%, and semiconductors fell 4.09%. My book fell $1,481. Five of my 32 positions closed higher.
Where the money actually went
Figure 1 · Dollar attribution
Seven names account for most of the day
Share counts × the change in each position’s closing price, August 17 to August 18, 2026. Percentages and dollars diverge sharply here — NBIL fell the most in percentage terms but Sandisk cost more in dollars, because Sandisk is a far larger position. Source: Massive Market Data closes, applied to my own share counts.
Sandisk is the point of that chart. It fell 9.01% — a smaller percentage move than NBIL’s 15.23% — and cost nearly twice as much, because it is my largest position at close to its 10% single-name cap. The name that hurts most in a bad session is almost never the one at the top of the percentage-decliner list, and I’d rather look at this chart than that list.
The names, and what they cost
| Position | Close | Day | Book impact | vs. my cost |
|---|---|---|---|---|
| Sandisk Corp (SNDK) | $1,625.78 | −9.01% | −$322 | +236.45% |
| GraniteShares 2x Long NBIS Daily ETF (NBIL) | $31.44 | −15.23% | −$170 | +240.63% |
| Micron Technology (MU) | $940.76 | −7.02% | −$142 | +188.73% |
| Meta Platforms (META) | $543.67 | −4.45% | −$127 | +132.02% |
| Talen Energy Corporation (TLN) | $317.66 | −11.00% | −$118 | +0.98% |
| Advanced Micro Devices (AMD) | $484.39 | −4.27% | −$108 | +143.00% |
| Intel Corporation (INTC) | $96.69 | −6.58% | −$95 | +158.31% |
| CoreWeave Inc (CRWV) | $93.17 | −12.10% | −$64 | +12.33% |
| Vertiv Holdings Co (VRT) | $272.54 | −6.80% | −$60 | −10.61% |
| WhiteFiber Inc (WYFI) | $27.07 | −10.81% | −$46 | +103.08% |
| Nebius Group N.V. (NBIS) | $248.43 | −7.60% | −$41 | +175.67% |
| Arm Holdings plc (ARM) | $253.32 | −6.67% | −$36 | +21.89% |
| Invesco QQQ Trust (QQQ) | $717.51 | −1.69% | −$25 | +28.52% |
| Space Exploration Technologies (SpaceX) (SPCX) | $143.34 | −1.98% | −$23 | +6.18% |
| The Cheesecake Factory (CAKE) | $106.36 | −2.48% | −$22 | +134.27% |
| SPDR S&P 500 ETF (SPY) | $767.45 | −0.68% | −$16 | +17.13% |
The last column is the one that keeps a bad day in proportion: after a 9.01% decline, Sandisk is still up 236% on my cost basis.
How much cushion is actually left
“Still ahead of my cost basis” is easy to say and hard to size, so here it is sized: how far each of the day’s big movers would have to fall from Tuesday’s close to reach what I actually paid, and how many repeats of Tuesday that would take.
| Position | Close | My cost | Fall to reach cost | Repeats of that day |
|---|---|---|---|---|
| Sandisk Corp (SNDK) | $1,625.78 | $483.21 | −70.3% | 13 |
| Micron Technology (MU) | $940.76 | $325.83 | −65.4% | 15 |
| Intel Corporation (INTC) | $96.69 | $37.43 | −61.3% | 14 |
| Meta Platforms (META) | $543.67 | $234.32 | −56.9% | 19 |
| GraniteShares 2x Long NBIS Daily ETF (NBIL) | $31.44 | $9.23 | −70.6% | 7 |
| Talen Energy Corporation (TLN) | $317.66 | $314.57 | −1.0% | under 1 |
| Vertiv Holdings Co (VRT) | $272.54 | $304.90 | +11.9% | — already below |
Assumptions, stated plainly. The last column compounds that single session’s percentage move repeatedly until the position reaches my average cost. Nothing about markets works that way — daily returns do not repeat, and a run of that length would long since have triggered a rethink rather than a spreadsheet. It is not a forecast and carries no probability. It is a unit of measurement: it converts “still well ahead of cost” from a reassuring phrase into a countable number of bad days, which is the only form in which that phrase is worth anything. Talen is the number that surprised me: after an 11.00% day it sits 1.0% above my cost, one more bad session from joining Vertiv under water. That is the position this exercise flagged that the dollar chart did not.
What happened, in one paragraph
CoreWeave, my largest AI-infrastructure position by cost, fell 12.10% to $93.17 on reporting that tied the decline to its debt load — $9.4 billion in second-quarter capital spending against nearly $30 billion in long-term debt — being repriced as the cost of that debt rises. Nebius fell 7.60% on the same mechanism, and because I hold NBIL, a 2× daily-leveraged fund on Nebius, that position fell 15.23%: almost exactly double, which is leverage doing what leverage does rather than a separate story. The full writeup is on the News page; my read on the mechanism is in the Opinion piece.
One thing I said that day that was wrong
I originally described this as yields “moving higher” on Tuesday. Checked against the constant-maturity series, the 10-year actually fell a basis point on the day, after rising nine basis points over the three prior sessions. The mechanism holds; the timing I gave it didn’t. Corrected here and in both linked articles rather than quietly edited away.
The one number I want on the record
Vertiv slipped below my cost
−10.62%Vertiv closed at $272.54 against my $304.90 average cost, 10.62% under water — deepening a drawdown that had already started before this week rather than beginning it. It is now the only position in the book more than 5% below what I paid.
The rest of the book, after the worst day
29 of 3229 of 32 positions closed Tuesday above my cost basis, several by multiples. The other two below water are Netflix, at 1.84% under, and Tempus AI, which slipped 0.74% below cost on Tuesday itself. Neither is material. That is context, not comfort, but it is the difference between a drawdown and a loss.
No trades today
The Blackout Rule and Holding Period exist for exactly this kind of session — one bad day, even a genuinely bad one, is data, not a decision. Nothing about Tuesday changes a thesis I’ve written up on any of these names; it moved the price I’d be transacting at, not the case for owning them. I made no trades. The follow-up, when the same thing happened again the next day, is in Wednesday’s note.