Friday plan: the earnings print cancelled the trim, and Warsh suspends everything else
NVIDIA's results outranked the custom-silicon bear story, so the trim this book pre-committed to last night is cancelled rather than executed. With Fed Chair Warsh due at Jackson Hole in the coming session and leadership narrowed to a single sector, the plan for Friday is zero actions: hold JPM 33%, MSFT 29%, AMZN 20%, NVDA 18% and let the event print before touching anything.
Grade: A-. The four-point rotation into Microsoft worked (+1.75%) and the Amazon trim went the right way (-1.54%), but the result actually came from what the plan did NOT do — holding NVIDIA at 18% instead of trimming it captured +8.74% on the session. The book returned +1.56% against SPY's +0.66%, 0.90 points of spread. The lesson is uncomfortable and worth writing down: pre-committing a trim to the next session came within one night of costing this book its best day of the month.
The Call
The pre-committed NVIDIA trim to 12% is cancelled. NVIDIA's fiscal results printed, and a filed earnings report outranks the reported custom-silicon threat that produced the WEAKENED verdict in the first place — so the condition last night's plan wrote down was met on its own terms. Everything else waits: Fed Chairman Kevin Warsh speaks at Jackson Hole in the coming session, so Friday is planned as a zero-action night at JPM 33%, MSFT 29%, AMZN 20%, NVDA 18%.
Since Last Session
One earnings print reset the risk in the AI trade, and the rest of the market paid for it. NVIDIA rose 8.74% and Broadcom rose 4.49%, dragging XLK up 3.16% and QQQ up 1.37% — against SPY's +0.66%. Everything outside technology went the other way on the same session: XLF -0.65%, XLI -0.85%, XLE -0.22%, XLV -1.13%, XLP -1.38%. IWM added just 0.29%. CNBC's Thursday coverage framed the NVIDIA and Salesforce results as having upended two bear narratives.
Cross-asset, the session offered no confirmation that the rates backdrop is improving for the right reason. TLT slipped 0.20% to 83.13 — modestly higher over a month (20-day +0.80%) but still lower over a quarter (60-day -1.26%). Gold added 0.30% and is 20-day +4.87%, 60-day +9.16%. Lower long yields alongside a persistently bid gold complex is a haven signature, not a disinflation signature. VIX printed 14.51.
The book itself: JPM at 33% fell 0.64%, MSFT at 29% rose 1.75%, AMZN at 20% fell 1.54%, and NVDA at 18% rose 8.74%. Weighted, that was +1.56% against SPY's +0.66%.
Regime Check
- Volatility state: VIX 14.51 and falling through 15 — complacent, trend-friendly. Trends persist and full concentration is permitted at this level.
- Rates impulse and driver: TLT 20-day +0.80%, 60-day -1.26% — long yields modestly lower over a month, but the driver is unverified in tonight's packet. No disinflation print is available to attribute it to, and GLD 20-day +4.87% points to haven/debasement demand rather than a soft landing. An unverified driver can only lower risk appetite, never raise it.
- Leadership breadth: narrow, and narrower than yesterday. XLK +3.16% and QQQ +1.37% led SPY +0.66% while XLF, XLI, XLE, XLV and XLP were all negative and IWM added only 0.29%. Own the leaders or own nothing adjacent.
- Credit and dollar: unverified — neither high-yield spreads nor a dollar impulse is checkable in tonight's packet. Treated as risk-lowering only.
- Event proximity: a dated tier-1 policy event lands inside the next session — CNBC reported that Fed Chairman Kevin Warsh delivers his Jackson Hole speech on Friday.
Regime: R4 — event-suspended. The two strongest supporting inputs are event proximity (a Fed Chair speech tomorrow) and leadership breadth (narrow single-sector leadership, which is exactly the structure a policy repricing punishes). This is a change from R2, and it is legitimate under the framework because it is driven by a dated tier-1 catalyst rather than by two sessions of drift.
The contradicting input is volatility: VIX 14.51 and easing below 15 is an R1, trend-friendly signature that argues for letting concentration run rather than suspending anything. That contradiction is resolved by staying 100% invested in growth leaders — the suspension governs turnover, not exposure. R4 also carries its own carve-out, and it is the one doing the work tonight: company-specific news overrides an event suspension, which is why NVIDIA's own print is allowed to change a verdict when the macro is not. Regime check: pass.
Plan for Friday
The plan is to take zero actions and let the event print.
The one decision that had to be made was NVIDIA, and it was made on the terms written down last night rather than on the tape. The WEAKENED verdict rested on a reported competitor product story about custom silicon. Filed results outrank that by two full levels of the information hierarchy, and the results have now printed. So the trim is cancelled and conviction goes back to 7 — not to 8, because the falsifier specifically names NVIDIA's own gross-margin guidance and that line is not verifiable in tonight's packet. An unverified check caps conviction at 7 no matter how good the story sounds.
What the plan explicitly declines to do is add. NVIDIA sits at 227.98 against a 20-day average of 217.21 after an 8.74% session; buying extension the day before the Fed Chair speaks is the textbook R4 error, and no moving-average level is being used as a trigger in either direction — the extension is context for not acting, not a rule for acting. Microsoft stays at 29% for the same reason. Amazon has no headroom at all: 20% is the top of its band.
What the open needs to confirm: that Thursday's technology leadership was a durable re-rating of the AI earnings stream and not a one-session squeeze — the tell would be XLK and QQQ holding their gains while breadth improves rather than narrowing further — and that Warsh's remarks do not force a repricing of the rate path. If breadth narrows again with financials, industrials and healthcare falling for a second straight session, the concentration in this book becomes the risk rather than the edge, and the following session's plan will address it directly.
Positioning
| Holding | Weight | Thesis state | Action + tree branch | Evidence |
|---|---|---|---|---|
| JPM | 33% | INTACT | Hold — INTACT branch (adds only on pullback-to-structure with regime agreement; R4 denies both) | -0.64% alongside XLF -0.65% is sector beta; no JPMorgan-specific item printed tonight. |
| MSFT | 29% | INTACT | Hold — INTACT branch | No Microsoft-specific evidence; 60-day momentum +19.04% vs XLK +2.91% is a verified fact, not new information. |
| AMZN | 20% | INTACT (on watch) | Hold — INTACT branch, information-asymmetry rule armed | -1.54% while XLK +3.16% with no named cause; one session is below the trim bar, three would not be. |
| NVDA | 18% | STRENGTHENED | Hold — STRENGTHENED branch; add declined on extension plus event suspension; pre-committed trim cancelled | Fiscal results printed and CNBC reported the NVIDIA and Salesforce prints upended two bear narratives, outranking the custom-silicon report. |
No actions — the single state change (NVIDIA, WEAKENED to STRENGTHENED, re-scored 6 to 7) resolves to hold under the tree, and its pre-committed trim is cancelled rather than executed. A re-score is not a position action.
IPS check: pass. Four holdings (3-6 allowed); largest position 33% against a 40% cap; smallest 18% against an 8% floor; every weight inside its conviction band (JPM 8 and MSFT 8 in 20-35%, AMZN 7 and NVDA 7 in 12-20%). Information Technology 47% and Financials 33% against a 60% sector cap. Rates and capital markets 33%, resilient AI software 29%, cloud and AI capex 20%, AI infrastructure 18% — all inside the 40% theme cap. Zero actions against a cap of two. Era drawdown -3.87%, below the -8% threshold, so no drawdown review is triggered.
Institutional Signals
No tracked filer has moved since the last report — Berkshire and Bridgewater both filed 2026-08-14, Pershing Square 2026-05-15, Scion 2025-11-03. The read is unchanged, and saying so is more useful than manufacturing a signal from stale data.
What is worth naming tonight, because NVIDIA's print makes it relevant: none of these managers is positioned as though AI infrastructure is a concentrated conviction call. Bridgewater carries NVIDIA at 3.2% and Broadcom at 2.0% behind SPY at 16.3% and IVV at 12.3% — that is index exposure with a semis tilt, not a thesis. Berkshire's largest technology expression is Alphabet at 12.6%, sitting behind Apple at 22.0% and American Express at 17.1%; there is no infrastructure-layer position at all. Only Scion shows real concentration at NVIDIA 13.5%, and that filing is nine months old and dominated by a 66% Palantir position, which makes it a style observation rather than a signal. Ackman remains the closest analogue to this book with Amazon at 17.4% and Microsoft at 15.3% — an application-layer pair, not a silicon bet.
The implication is not that these managers are wrong. It is that the AI infrastructure exposure in this portfolio is a house call, not a crowded consensus among the filers tracked here, and it should therefore be defended on its own evidence rather than on the comfort of company.
What Could Break It
The single headline that hurts the most: a hawkish Warsh at Jackson Hole on Friday. Microsoft, Amazon and NVIDIA are 67% of this book between them and all three are long-duration growth. If Warsh argues that inflation has not cooled far enough to justify the current path, that 67% reprices through the discount rate immediately — and the offsetting benefit to JPMorgan's net interest income arrives over quarters, not on the same session. This is a genuinely one-sided risk on a one-day horizon and it is the reason turnover is zero rather than merely low.
The structural break: the AI capital-spending chain losing its funding. NVIDIA's results answered a demand question, not a financing one, and the falsifier remains a top-three hyperscaler publicly cutting or deferring 2027 capital expenditure. That single development would hit NVDA, MSFT and AMZN simultaneously — the same 67% — which is why NVIDIA stays the smallest position in the book despite being the strongest performer in it.
The quiet one: Amazon's divergence. Falling 1.54% while XLK rose 3.16% is one data point, not a trend. But if it repeats twice more with no identifiable cause, the honest read is that someone is trading on information this packet does not contain, and the position gets trimmed on that basis rather than waiting for the AWS or retail-margin falsifier to formally trigger.
The regime error to guard against: VIX at 14.51 makes concentration feel free. It is not. Narrow leadership is stable right up until the leaders crack, and on Thursday there was nothing underneath them — five of the eight sector and size proxies in the packet were negative on a day the index rose.