/decisions · 2026-05-18 · investing · nvidia-adjacent-v1

Deploy NVIDIA-adjacent v1 paper trade (competitors + parallel-track split of AI-infra-chips bucket) — go / revise / defer / archive?

Founder gate on the second split of the prior "AI-infra-chips" bucket. Owns the competitive layer: AMD (MI300/MI400 GPU), Broadcom (custom ASIC for Google/Meta), Intel (Terafab + Gaudi 3 turnaround). All paper; no live capital at risk.

Second of two splits of the prior "AI-infra-chips" bucket. This basket owns the layer that competes WITH NVIDIA for the same hyperscaler accelerator wallet — capex routed to non-NVIDIA accelerators. AMD is the strongest signal: 2-manager NEW (ARK built from outside top-10 to #2 position over 3 quarters + Tepper persistent hold), the cleanest 2+ manager smart-money signal of the year. AVGO captures the custom-ASIC layer (Tiger Global $1.11B position, Google TPU + Meta MTIA). INTC sized small for double-count + turnaround uncertainty. Druckenmiller doctrine (no mechanical exits). R-weighted: AMD 1.5R + AVGO 1.0R + INTC 0.25R + cash reserve 1.25R = 4R bucket cap. INTC is double-counted with smart-money-mirror v1 — net INTC exposure across both strategies = 0.75R = $3,750. Paper only.

Why the split exists

See sister gate paper-trade-nvidia-supply-chain-v1-go for the full split rationale. TL;DR: supply-chain monetizes NVIDIA's revenue (vendor-agnostic); adjacent monetizes the alternative path. Different exit triggers, different fundamental anchors. Founder confirmed taxonomy 2026-05-18 (AMD belongs in adjacent).

Smart-money signal scoring (8 tracked managers)

TickerSignalSourceAnchorMirror overlap?
AMD2-manager NEWARK #2 at 4.29% AUM (built from 0 over 3Q); Appaloosa-Tepper persistentHIGH (MI300/350 shipping; MI400 announced; ROCm closing CUDA gap)NO
INTC2-manager 2026Q1 NEWDruckenmiller-Duquesne NEW + Tiger Global NEW (both 2026Q1)MEDIUM (Terafab + Gaudi 3; turnaround uncertainty)YES — at 0.5R in mirror
AVGO1-manager high-convictionTiger Global #7 at $1.11B (2026Q1); +912% in 2024Q3HIGH ($15B+ custom-ASIC run-rate; Google TPU + Meta MTIA)NO
MRVLnone — CUTZero smart-money; redundant with AVGOMEDIUMNO
Samsung (005930.KS)not tradeableKorean-listed; not AlpacaLOWNO

Bull case vs revision risk (the explicit tension)

The bull case: Hyperscaler custom-silicon programs reached material scale in 2025-2026 — Google TPU v5/v6 (Broadcom-designed), Meta MTIA (Broadcom-designed), Amazon Trainium 2/3 (Marvell-designed), Microsoft Maia (Marvell-adjacent), AMD MI300/MI350 at Meta + Microsoft + Oracle are no longer roadmap items, they ship at scale. Combined likely captures 25-40% of incremental hyperscaler AI compute spend in 2026 (up from <10% in 2023) — even at the low end the dollar is enormous because hyperscaler capex is ~$750B/yr. AMD specifically has the cleanest 2-manager NEW signal of the year — ARK (historically Nvidia-skeptical) built a 4.29%-AUM position over 3 consecutive quarters to make AMD their #2 holding. That is a meaningful pivot from a manager with public Nvidia-skepticism. Tepper holds AMD persistently. Two independent positive-sum allocators cross the AMD line. INTC turnaround signal is corroborated: Druckenmiller + Tiger both opened NEW INTC positions in 2026Q1 (first time INTC shows up on the smart-money board since 2021). Terafab + Gaudi 3 is real if uncertain. AVGO is the sleeper — Tiger Global's $1.11B position is one of their largest, and the custom-ASIC business at $15B+ run-rate with Google + Meta as named customers IS the hyperscaler-routes-around-NVIDIA thesis in clearest form.

The revision risk: Adjacent names lagged NVIDIA badly through 2023-2024 (NVDA ~10x while AMD went sideways, INTC down 30%+). They only re-rated in late 2024 / 2025 once MI300 shipped at scale. The thesis is implicitly that we're in the maturation phase of the AI ramp where competitors win share — but if NVIDIA's pricing power proves more durable than that (Blackwell Ultra commanding even larger premium, ROCm software gap NOT closing, CUDA moat re-expanding), the adjacent basket underperforms. ARK is the load-bearing 2nd-manager signal for AMD, and ARK is thematic-disruption-grade not quality-compounder-grade — per the ARK manager-note, ARK signal is "idea-generation" not "conviction-stacking." If ARK trims or exits AMD next quarter, the 2-manager threshold for AMD breaks and we're back to single-manager (Tepper). INTC turnaround risk is binary; Terafab either delivers or doesn't. AVGO is single-manager (Tiger) — if Tiger trims aggressively the smart-money corroboration weakens significantly. AMD concentration is 37.5% of bucket — single-name drawdown material. N=1 sanity-check is not a backtest.

The picks: APPROVE locks in the competitive-displacement paper-trade shape at R-weighted sizing. REVISE walks back (e.g. drop INTC entirely from adjacent — rely on mirror coverage — and redeploy 0.25R into AMD; or reduce AMD concentration; or add MRVL despite zero smart-money). ARCHIVE concedes the maturation-phase thesis is wrong (NVIDIA pricing power durable). DEFER waits for Q2 13F to confirm ARK held AMD + Druckenmiller / Tiger held INTC.

Deploy parameters (R-weighted)

NameR$Convergent signals
AMD1.5R$7,5002-manager NEW signal (ARK + Tepper); ARK built from outside top-10 to #2 position over 3Q (meaningful pivot from a historically Nvidia-skeptical manager); MI300/MI350 shipping at named hyperscalers; MI400 announced for 2026. Strongest signal in basket.
AVGO (Broadcom)1.0R$5,000Tiger Global #7 at $1.11B (one of Tiger's largest positions); $15B+ run-rate compute-ASIC business with Google TPU + Meta MTIA. Custom-ASIC is the cleanest hyperscaler-routes-around-NVIDIA bet.
INTC0.25R$1,2502-manager 2026Q1 NEW (Druckenmiller + Tiger). Terafab + Gaudi 3 turnaround uncertainty. DOUBLE-COUNTED — mirror already holds 0.5R; net INTC = 0.75R combined. Sized lower than raw signal would suggest specifically for double-count + Druckenmiller's entry-exit-re-entry pattern reads opportunistic.
Cash reserve1.25R$6,250Reserved for: (a) anchor-strength add to AMD on >20% drawdown with bullish anchors, (b) MRVL if smart-money signal materializes in Q2/Q3 2026 13F, (c) Samsung if brokerage expands to Korean ADRs.
Total deployed2.75R$13,750Initial deploy; remaining 1.25R in reserve.
Bucket cap4R$20,000Paper, Alpaca sandbox.
ParamValue
ModePaper only (Alpaca paper sandbox)
1R unit$5,000
Initial deploy2.75R total = $13,750 (AMD 1.5R + AVGO 1.0R + INTC 0.25R)
Cash reserve1.25R = $6,250 held for anchor-strength adds + signal-confirmation new names
Bucket cap (adjacent total)4R ($20k)
Per-trade stopNone (thesis-stop, not price-stop)
Profit trimsNone — let winners run (Druckenmiller doctrine)
Exit trigger (single signal)Founder review via /decisions/ page within 7 days
Exit trigger (confirmation)2+ HIGH severity anchors confirmed 1-2Q = close ENTIRE bucket
Kill switchFounder channel "halt adjacent" or "pause investing" = immediate close
Review cadenceQuarterly (Ray runs, surfaces to founder only on anchor flip)
INTC double-count decisionSurfaced for founder explicit OK — net INTC = 0.75R across mirror + adjacent

Lock in

Approve — deploy paper per v1 R-weighted parameters

Ray fires 2.75R initial deploy ($13,750 paper) into Alpaca per the R-weighted table — AMD 1.5R + AVGO 1.0R + INTC 0.25R. Holds 1.25R cash reserve. Logs to positions/nvidia-adjacent-v1/. Sets up quarterly anchor watch. INTC net exposure across mirror + adjacent = 0.75R = $3,750 accepted. Use the field for any deploy notes or sizing overrides.

Approve + send
Revise — walk back specific parameters

v1 is close but a parameter needs adjusting (e.g. drop INTC entirely from adjacent → redeploy 0.25R to AMD bringing AMD to 1.75R; reduce AMD concentration; add MRVL despite zero smart-money; equal-weight instead of R-weighted). Name what; Ray re-files v1.1.

Revise + send
Archive — thesis does not deploy

If the competitive-displacement argument doesn't pencil — NVIDIA pricing power durable, hyperscaler custom-silicon programs stall on software-stack maturity, ROCm gap not closing. Ray files as status: archived-pre-deploy with reason. No paper trade. Adjacent bucket closes (mirror still holds INTC independently).

Archive + send
Defer — wait for Q2 13F confirmation

Hold for Q2 2026 13F (re-confirm ARK still holds AMD top-10, Druckenmiller + Tiger still hold INTC, Tiger still sized AVGO) before deploying. Q2 13F filings land mid-August.

Defer + send