Construction

Portfolio Construction Overview

How the 27 holdings are assembled into a coherent portfolio. Read after The AI Value Chain. Companion notes: Core vs Satellite Holdings, Return Drivers vs Hedges, Overlap & Diversification Map, Position Sizing Framework.

The Construction Philosophy#

PCA SOF is built as a value-chain-complete, concentrated, barbelled thematic portfolio. Three principles govern it:

  1. Completeness over selection. Rather than bet on the single AI winner, own every layer of The AI Value Chain so the fund captures value wherever it migrates.
  2. Concentration with role-clarity. ~27 names, each with a defined role. No holding exists without a job.
  3. Barbell the risk. Capital-intensive cyclical compounders (memory, foundry, power) on one end; capital-light secular software + a non-AI diversifier ring on the other.

The Three Construction Rings#

   RING 1: THE AI INFRASTRUCTURE SPINE  (the return engine, highest weight)
   NVIDIA · TSMC · Micron · Marvell · Disco · Constellation · Vistra
   + the hyperscalers: Microsoft · Amazon · Alphabet · Meta
        → highly correlated; share a RISK BUDGET (see Overlap & Diversification Map)

   RING 2: THE AI SOFTWARE / MONETISATION LAYER  (quality compounders)
   ServiceNow · Salesforce · Snowflake · Datadog · CrowdStrike · Zscaler
        → monetise AI in the enterprise; lower capex-cyclicality than Ring 1

   RING 3: THE DIVERSIFYING SECULAR RING  (the shock absorber, lower correlation)
   Payments: Adyen · PayPal
   EM internet: Tencent · MercadoLibre
   China consumer: Li Ning   |   EV: BYD
   Media: Netflix · Roblox    |   Healthcare: BioNTech · IQVIA
        → independent drivers; dampen AI-capex beta

The genius of the construction: Ring 1 is the torque, Ring 3 is the ballast, Ring 2 bridges them.


How Weighting Works (inferred)#

Weights follow the conviction-and-role logic of Position Sizing Framework:

  • Largest weights → Ring 1 chokepoints with the clearest AI leverage + deepest moats: Micron, NVIDIA, TSMC at/near the top (per the stated holdings order), plus the mega-cap hyperscalers.
  • Mid weights → Ring 2 software compounders + the power names (Constellation, Vistra) + Marvell.
  • Smaller weights → Ring 3 diversifiers + higher-variance satellites (Roblox, BioNTech, Li Ning, Disco).

The stated holdings order (Micron → NVIDIA → TSMC → Alphabet → Datadog → … → BYD) broadly tracks descending weight, confirming an AI-infrastructure-led book with a long diversifying tail.


Why The Portfolio Is Built This Way, Answering The Key Questions#

| Question | Answer | |, -|, -| | Which are core holdings? | The AI infrastructure spine + hyperscalers (Ring 1). → Core vs Satellite Holdings | | Which are satellites? | Ring 3 diversifiers + higher-variance names (Roblox, BioNTech, Li Ning, Disco, Snowflake). | | Which drive returns? | NVIDIA, TSMC, Micron, the hyperscalers, the power names, Ring 1. → Return Drivers vs Hedges | | Which hedge risks? | The Ring 3 diversifiers (payments, EM, healthcare, EV, media). | | Which overlap? | The entire AI spine is correlated, a managed risk budget. → Overlap & Diversification Map | | Which diversify? | Geography (TSMC, Tencent, BYD, MELI, Adyen, BioNTech, Disco, Li Ning) + theme (payments, healthcare, EV). |


The Barbell, Explicitly#


What Could Go Wrong With The Construction#

The honest critique: Ring 1 + much of Ring 2 share one factor, AI capex. If hyperscaler spend pauses, ~15+ holdings fall together regardless of their layer. The fund's only true diversification is Ring 3 + the geographic spread. This is the central risk the construction manages but cannot eliminate, see Overlap & Diversification Map + AI Capex Cycle Risk + Risk Master Note.


Linked Notes#

Back to PCA SOF