Overlap & Diversification Map
The honest correlation picture: where the book is secretly one trade, and where it is genuinely diversified. The most important risk-management note in 04 Portfolio Construction.
The Central Risk: The AI-Capex Factor#
The fund's biggest hidden exposure is that ~18 of 27 holdings share one factor, hyperscaler AI capital expenditure. A pause in capex from Microsoft/Amazon/Alphabet/Meta Platforms transmits to:
Hyperscaler capex ↓
→ NVIDIA orders ↓ → TSMC utilisation ↓ → Disco tool orders ↓
→ Micron HBM demand ↓ → Marvell ASIC/optics ↓
→ Constellation/Vistra power-demand narrative ↓
→ Snowflake/Datadog consumption ↓ (AI workloads)
→ CrowdStrike/Zscaler/ServiceNow/Salesforce AI-upsell ↓
One macro variable moves two-thirds of the portfolio. This is the price of value-chain completeness. The fund manages it via a shared risk budget for the spine + a deliberately diversifying Ring 3.
Overlap Clusters (high internal correlation)#
| Cluster | Holdings | Shared driver | |, -|, -|, -| | Silicon scarcity | NVIDIA · TSMC · Micron Technology · Marvell Technology · Disco Corporation | AI chip demand; also share Taiwan Strait Risk (all touch TSMC/Taiwan/Asia supply) | | Hyperscale cloud | Microsoft · Amazon · Alphabet · Meta Platforms | AI capex + ad/IT cycle; also compete with each other | | AI power | Constellation Energy · Vistra | Datacentre power demand + power prices | | AI software estate | Snowflake · Datadog · CrowdStrike · Zscaler · ServiceNow · Salesforce | Enterprise IT/AI budgets + rates (long-duration) | | Payments | Adyen · PayPal | Commerce volume | | China | Tencent · Li Ning · BYD | China-macro + policy + geopolitics | | Healthcare | BioNTech · IQVIA | Biopharma R&D cycle |
Within clusters, holdings are complements but also correlated, e.g., the silicon-scarcity cluster all benefit from AI but all suffer in a digestion + all carry Taiwan/Asia exposure.
Cross-Cutting Shared Risks (correlations that span clusters)#
| Shared risk | Holdings affected | |, -|, -| | AI Capex Cycle Risk | The entire spine + software estate (~18 names) | | Taiwan Strait Risk | NVIDIA, TSMC, Micron, Marvell, Disco (+ hyperscalers' custom silicon) | | Interest Rate Sensitivity | All long-duration growth (software estate, mega-caps, Roblox, Snowflake) | | China Regulatory Risk / Geopolitical Risk | Tencent, BYD, Li Ning, TSMC, Micron, NVIDIA (China sales) | | Advertising cycle | Alphabet, Meta, (Amazon, Netflix ad-tier), Tencent |
Where The Diversification Is Real#
The genuinely low-correlation exposures (the parts that aren't the AI trade):
- BioNTech: binary clinical outcomes; the single most uncorrelated holding.
- The China cluster (Tencent · Li Ning · BYD), driven by China-macro + policy, not US-AI capex (though they share their own China factor).
- MercadoLibre: LatAm digitisation + EM-macro.
- Payments (Adyen · PayPal), commerce volume + (PayPal) self-help.
- Netflix: consumer-defensive subscription.
- IQVIA: biopharma R&D cycle.
- Constellation Energy / Vistra: partial diversifier (AI-linked but via power markets, a different transmission).
The Diversification Verdict#
- Theme diversification: moderate. Real Ring-3 themes exist, but the book is AI-capex-dominant.
- Geographic diversification: good. US core + Taiwan, China, Japan, Europe, LatAm.
- Factor diversification: the weak point. Long-duration-growth + AI-capex factors dominate; in a rates-up / risk-off regime, much of the book correlates.
Implication for sizing: the spine must be risk-budgeted as a single position of correlated parts, and Ring 3 must be large enough to matter. → Position Sizing Framework, Risk Master Note.
Linked Notes#
- Portfolio Construction Overview · Return Drivers vs Hedges · Position Sizing Framework · Core vs Satellite Holdings
- Risk Master Note · AI Capex Cycle Risk · Taiwan Strait Risk · Interest Rate Sensitivity · Macro Master Note · Knowledge Graph