Portfolio Rebalance Sector IC Memo

Executive Summary & Action Plan

Verdict

REDUCE / DE-RISK — sell into strength this week; the book is not being paid for its tail. Twenty-two line items compress into roughly three risk buckets, with the AI-capex factor pricing ~74% of capital; weighted beta is 1.91 against a 1.4–1.6 guardrail, cash is 0% against a 12–18% target, and 4 of 5 house guardrails set on 7/29 are breached. Marked against IWANNAVY fair values, the weighted gap is −32%: only SpaceX (SPCX, +13% to mark) and NVIDIA (NVDA, +5%) trade at or below fair value, while the largest positions carry the widest gaps (TSLA −62%, MU −39%, AAPL −40%, SNDK −66%). The 12M scenario tree yields ≈ −0.5% expected return against a −35% tail — de-risking is positive-expected-value, not defensive. Sequence matters: Palantir (PLTR) prints tonight, SpaceX's first-ever report lands Tuesday 8/4 with a ~$116B lockup tranche Thursday 8/6, Sandisk (SNDK) prints 8/5, Rocket Lab (RKLB) 8/10 — about 15% of capital reports within five sessions.

Loading chart

Price $109.57 | Market cap $1.44T (13.17B FD sh) | Target $124.00 | Upside +13.2% IWANNAVY Fair Value (7/25, held into the 8/4 print — Flight 13 resolved 7/24: Ship success with first operational Starlink V3 deploy, booster lost at recovery; neither is bookable cash flow) · price as of 2026-07-31 close · Street $236.71 (n=21, reference) — the Street capitalizes AI-segment run-rate the S-1 shows burning $2.47B/quarter

Loading chart

Abstract

We marked all 22 positions against IWANNAVY fair values — nine inherited from live coverage (five re-underwritten in today's semiconductor-leaders report) and twelve underwritten fresh with SBC-expensed, cycle-normalized discipline. The result is uncomfortable: the weighted gap to marks is −32%, the book's factor structure collapses to roughly three independent buckets, and 74% of capital monetizes the same hyperscaler-capex line the house already treats with caution. The memory sleeve (MU, SNDK, SK hynix — 19.9%) sits 21–66% above through-cycle marks days after the 3Q26 contract momentum peak printed; TSLA, the single largest position, carries a −62% gap on a 16.3% auto gross margin and negative free cash flow; SNDK is a 5.2-beta position priced at consensus FY27 EPS that annualizes a 78% gross-margin spike. Scenario weights (Bear 30 / Base 45 / Bull 25) produce ≈ −0.5% expected 12M return with a −35% tail. The plan sells ~$26K into this week's strength — memory, TSLA, half of SpaceX pre-print — restoring 16% cash and a 1.5 beta, with staged re-entry ladders reserved for house entry zones. Principal risks to the plan: an LTA floor-price regime change extending memory economics, and a SpaceX short squeeze off a 25.6% short float.

Forecast Path

Portfolio value under the three scenarios, anchored to today's $157.5K. Probabilities mirror the house semiconductor weights set 2026-08-03 (Bear 30 / Base 45 / Bull 25); bands are judgmental. The expected path is flat — which is precisely the argument for cutting the tail rather than riding it.

Loading chart

Action Plan

Nine orders, sequenced around the event calendar, raise ~$26.3K (16.7% cash) and take weighted beta from 1.91 to ~1.50 while keeping every quality core (NVDA, AVGO, GOOG, GEV, half of SPCX) intact. Trims are sized by guardrail and inherited marks first; fresh single-pass marks set direction, not full magnitude.

Order	Action	From → To (weight)	Timing / trigger
1	SpaceX (SPCX): sell 39 of 104 sh, or zero-cost collar the block	7.2% → 4.5%	Before Tue 8/4 close — event sizing, not valuation; the +13% mark survives only if Connectivity margin holds near the S-1's 64%
2	Sandisk (SNDK): sell 3 of 4 sh	3.1% → 0.8%	Before the 8/5 print / into the 8/13 Investor Day — widest gap (−66%), 5.2 beta
3	Tesla (TSLA): sell 15 of 62 sh	12.5% → 9.5%	Immediate; guardrail breach + widest megacap gap (−62%)
4	Micron (MU): sell 6 of 19 sh	9.9% → 6.8%	Into 3Q contract-momentum strength; −39% to the $500 mark
5	SK hynix (000660.KS): sell 4 of 10 sh	6.9% → 4.1%	Into a ₩1,616,000 pivot reclaim if given, else staged this week; −21% to ₩1,231,000
6	Marvell (MRVL): sell 10 of 36 sh	4.2% → 3.0%	Before the 8/27 print; −47% gap, highest rework-probability mark
7	Coherent (COHR) 13→9 sh · Corning (GLW) 23→17 sh · Bloom (BE) 10→7 sh	6.6% → 3.9% combined	Into strength; fresh marks −49% to −60%, all satellite-sized
8	Hold unchanged: NVDA, AVGO, GOOG, AAPL, AMZN, TSM, GEV, VICR, PLTR, RDW, PL, RKLB, ITA	—	AAPL flagged (fresh mark −40%, single-pass) — re-mark before trimming below 9.5%
9	Cash raised ≈ $26.3K	0% → 16.7%	Deploy only at the re-entry ladders below

Post-plan book: beta ~1.50 · memory 11.7% · semis+storage 30.9% · top-5 ~46% — all guardrails restored except TSLA at the 9.5% edge
Re-entry ladders (house zones only): NVDA $175 / $165 · SK hynix ₩1,292K / ₩1,231K · SPCX $105 / $95 / $85 (stop $78) · Samsung (005930.KS) ₩221K–₩208K as a memory swap candidate (least-stretched name in the complex — swap for hynix/MU exposure, never a net memory add)
Prob-weighted 12M: ≈ −0.5% before rebalancing; after, the tail compresses to ≈ −19% with dry powder at the ladders — the EV gain lives in what the cash buys, not the trims themselves | Confidence: High on sizing/guardrail actions, Medium on fresh-mark magnitudes

Every Holding vs Fair Value

The ranking below is the per-name deliverable: upside to IWANNAVY fair value at frozen prices (US 7/31 close, KR 8/3 close). Inherited marks carry their as-of dates; the twelve fresh marks are first-pass (single underwriting, not yet adversarially verified) and are flagged as such in the appendix. ITA is an ETF held at NAV and excluded.

Loading chart

What Each Company Is — and Its Crux

Mega platforms (36.6%). Apple (AAPL, $183): record FQ3 (+16%) but Tim Cook's memory-cost "100-year flood" is compressing gross margin toward 46% while the stock pays a cycle-high 26.8x EBITDA — the bull catalyst is, ironically, the house's own memory-peak call. Tesla (TSLA, $121): a 16.3% auto-gross-margin maker with negative FCF whose equity is mostly a robotaxi/Optimus option the house probability-weights at 25%, not near-certainty. Alphabet (GOOG, $286 affirmed): Cloud +82% against a $195–205B capex bill that just turned quarterly FCF negative for the first time since IPO. Amazon (AMZN, $246): AWS +37% reacceleration — fastest in 18 quarters — against a $220B capex wall; closest megacap to fair. Palantir (PLTR, $105 affirmed): elite growth at ~160x EV/FCF; prints tonight with options implying ±12%.

AI semis & memory (40.3% incl. hynix). NVDA ($210), AVGO ($320), TSM ($310), MU ($500), SK hynix (₩1,231,000) carry today's sector-report marks — memory margins are a conventional-DRAM squeeze being normalized, compute is closer to fair. Marvell (MRVL, $96): real custom-silicon franchise priced at leader parity; 8/27 print is the mark's rework test. Sandisk (SNDK, $405): pure NAND at a 78% gross-margin spike the street annualizes into FY27 EPS of ~$207; NAND wafer spot has been rolling for five weeks — the same discipline that set MU at $500 lands here at −66%.

Power & components (10.7%). GE Vernova (GEV, $525): the genuine AI power bottleneck, but the 6% FCF yield is customer-deposit float, not earnings — the street capitalizes it as recurring. Bloom (BE, $85): $20B narrative backlog against <$1B firm RPO and related-party JV revenue; securities suits pending. Vicor (VICR, $132): sold-out AI power modules plus a royalty annuity, but the true share count puts it at ~82x EBITDA, not the screen's 57x. Coherent (COHR, $110): ~75% genuinely AI-linked, priced at peak transceiver margins with CPO cannibalization ahead. Corning (GLW, $72): only ~20% of sales are AI-linked; July's −46% began the de-rate of an AI multiple on a non-AI book.

Space (12.4%). SpaceX (SPCX, $124): Flight 13 gave the revenue-relevant first (operational Starlink V3 deploy, ship intact) and lost the booster — cost curve, not revenue; the 8/4 print and metered unlock are the real tests, with 34% of float short. Rocket Lab (RKLB, $20): the best operator of the trio at the worst price — $38.9B caps a Q4'26 Neutron binary at 52x forward revenue. Redwire (RDW, $5.60): backlog inflecting but growth is equity-financed; TTM dilution alone consumed a quarter of the market cap. Planet (PL, $12): real NATO demand, net cash, and a price that already pays 16x forward revenue for flawless execution.

Scenario Rebalancing Strategy

The table is the strategy: what changes, when, and what proves it wrong. The Base plan (above) executes this week regardless; Bear and Bull are conditional overlays with dated triggers.

	Bear — 30%	Base — 45%	Bull — 25%
12M portfolio impact	−27% (range −22 to −33; tail −35%)	+3% (−5 to +8, violent chop)	+25% (+18 to +30)
Trigger conditions	Any two: 4Q26 DRAM contract flat/down · Fed hike (3 dissents live) · NVDA Q3 guide <$95B · a hyperscaler cuts 2027 capex	Contracts flatten, HBM4 2027 partially lands, Fed holds, SOXX rebuilds 50D Sept–Oct	2027 HBM4 signed near 2x ASP · NAND stabilizes 4Q26 · NVDA ≥$95B print + ≥$100B guide · PCE decelerates
Memory sleeve (now 19.9 → 11.7%)	Cut to ~8%: MU→5%, hynix→3%, SNDK exit	Hold 11.7%; swap hynix/MU into Samsung ₩208–221K allowed, no net add	Stop trimming; ride to trim ladders — hynix ₩1,616K→₩1,845K, MU into FQ4 strength
SpaceX (7.2 → 4.5%)	Cut to ~3% if print shows AI-burn shock (FCF worse than −$9.1B)	Hold 4.5%; re-add $105/$95/$85 post-unlock only if Connectivity EBITDA ≥60%	Trim remainder into $124–135 squeeze (25.6% short float is the fuel)
High-multiple duration (PLTR·BE·TSLA)	Cut PLTR/BE to zero on Fed hike; TSLA →7%	TSLA capped 9.5%; PLTR decision within 48h of tonight's print	Keep caps — duration rallies are for selling into under a hiking Fed
Cash (0 → 16.7%)	Build to 20%+; no deployment until SOXX >$534.30 with 3/5 breadth	Deploy only at house ladders (NVDA $175/165 · hynix ₩1,292K/₩1,231K · SPCX $105/95)	Deploy into quality enablers on pullbacks (NVDA/TSM/ASML), never into the memory spike
Beta (1.91 → ~1.50)	→1.3	hold ~1.5	≤1.6 cap even in strength

Catalysts & Risks

Catalyst	Timing	Book exposure	·	Risk	Prob	Impact
PLTR Q2 print (cons. $1.81B / $0.34; ±12% implied)	tonight 8/3 AMC	2.7%	·	Memory peak vs 19.9% sleeve (pre-plan)	High	High
SPCX first print (rev cons. ~$6.9B; FCF vs −$9.1B) → unlock 911.5M sh	8/4 → 8/6	7.2%	·	SPCX event stack + 74% share-count/mcap vendor discrepancy	High (certain)	Med-High
SNDK FQ4 + Investor Day (guide $7.75–8.25B)	8/5 → 8/13	3.1%	·	Beta 1.91 + 0% cash → forced-seller dynamics (pre-plan)	Med	High
RKLB Q2 (cons. ~$232M) + Neutron timeline	8/10 AMC	2.0%	·	Fed re-tightening tail (3 hike dissents, PCE 3.7%) hits the 114–160x duration cluster	Med	High
TSMC July revenue (≥NT$460B hot / <NT$400B digestion)	~8/10	~41% indirect	·	LTA floor-price regime change — invalidates memory trims (anti-plan)	Med	High
NVDA Q2 FY27 (≥$95B + ≥$100B guide = bull gate)	8/26	7.3% direct	·	KR dislocation + FX (KRW 1,427 vs 1,440 book)	Med	Med

Falsification & Review Rules

The de-risking is wrong if 2027 HBM4 contracts are signed near 2x 2026 ASPs with conventional floors intact (late-Oct Korean prints): halt memory trims, restore MU/hynix toward prior weights, and lift the memory marks per the sector-report rule.
The SPCX trim is wrong if the 8/4 print shows Connectivity EBITDA ≥64% with FCF burn improving and the 8/6 tranche absorbed on flat-to-up tape — re-add at the ladder, not the gap.
Fresh marks are provisional: AAPL/TSLA/AMZN and the nine mid-cap marks are single-pass; any position action beyond the guardrail sizes (orders 1–7) requires an adversarial verification round first. MRVL's mark is the most likely to move (8/27 print).
Mechanical review dates: within 48h of tonight's PLTR print; 8/5 SPCX post-print; 8/26 NVDA; late-Oct Korean Q3 prints (2027 HBM4 disclosure) — full re-mark of the memory sleeve either way.

References

Company prints & filings: SK hynix Q2'26 (7/29) · Samsung Q2'26 (7/30) · Apple FQ3'26 (7/30) · Amazon Q2'26 (7/30) · Tesla Q2'26 (7/22) · Bloom Q2'26 + guide raise (7/28) · GE Vernova Q2'26 (7/22) · SpaceX S-1
SPCX event cluster: CNBC — first earnings backdrop (8/3) · Starship Flight 13 outcome · Lockup mechanics (8/2) · Short interest 34% of float (7/31) · PLTR Q2 date (IR, 7/13) · RKLB Q2 date
Memory cycle & pricing: TrendForce 3Q26 server DRAM +13–18% (7/9) · Micron 5-yr price locks (6/25) · Hunterbrook on Bloom backlog (7/8)
Companion house reports: Semiconductor leaders — post-blow-off paths (2026-08-03) · Portfolio drawdown response (2026-07-29) · SpaceX company work (2026-07-25) · Alphabet (2026-07-23) · Global memory with China (2026-07-21)