PCA SOF Investment Thesis
The single argument that justifies the entire portfolio. If a holding doesn't serve one of the pillars below, it shouldn't be in the book.
The Thesis In Three Layers#
Layer 1, The Macro Bet#
Artificial intelligence is a general-purpose technology in the early innings of a multi-year, multi-trillion-dollar capital cycle, comparable in scope to electrification, the internet, and mobile. The spend is real, contracted, and accelerating: hyperscalers are pouring record capex into compute, and that capex cascades down a long, identifiable supply chain. The fund's first conviction is simply: this build-out is bigger and longer than the consensus believes, and it is supply-constrained at multiple points (leading-edge wafers, HBM, advanced packaging, and, increasingly, electricity).
Layer 2, The Structural Bet#
The value of the AI build-out does not accrue evenly, it accrues to chokepoints and to those who control scarce inputs. Rather than guess which application-layer company wins, PCA SOF owns the bottlenecks:
- NVIDIA: the compute platform + CUDA software moat.
- TSMC: the only foundry that can volume-manufacture leading-edge AI silicon.
- Micron Technology: one of three HBM suppliers; HBM is the gating input for accelerators.
- Disco Corporation: near-monopoly in the dicing/grinding tools needed for HBM stacking & advanced packaging.
- Constellation Energy / Vistra: clean, dispatchable power, the newest chokepoint.
Owning chokepoints means the fund gets paid whether OpenAI, Anthropic, Google, or a startup wins the model race.
Layer 3, The Diversification Bet#
A fund that is only AI infrastructure is one capex cycle away from a 50% drawdown. So the thesis deliberately rings the core with adjacent secular themes that have their own multi-year drivers, lower correlation to AI capex, and, critically, also benefit from AI as a tailwind:
- Payments (Adyen, PayPal), secular cash-to-digital shift.
- EM internet (Tencent, MercadoLibre), non-US digital penetration.
- Electrification (BYD), the EV/battery S-curve.
- China consumer (Li Ning), premiumisation of domestic demand.
- Healthcare innovation (BioNTech, IQVIA), mRNA platform + data infrastructure.
The Six Pillars (every holding maps to ≥1)#
- Compute & Silicon Scarcity: NVIDIA, TSMC, Micron, Marvell, Disco.
- The Power Wall: Constellation, Vistra. AI is electricity in a trench coat.
- Hyperscale & Cloud Monetisation: Microsoft, Amazon, Alphabet, Meta.
- The Software Tax on AI: ServiceNow, Salesforce, Snowflake, Datadog, CrowdStrike, Zscaler.
- The Digital Consumer: Meta, Netflix, Roblox, MercadoLibre, Tencent.
- Global Secular Compounders: Adyen, PayPal, BYD, Li Ning, BioNTech, IQVIA.
What You Must Believe (the load-bearing assumptions)#
| # | Assumption | If wrong… | |, -|, -|, -| | 1 | AI capex compounds for 3-5+ more years before digesting | Core spine de-rates hard → AI Capex Cycle Risk | | 2 | Compute/memory/foundry stay supply-constrained → pricing power | Margins normalise; cyclicality dominates | | 3 | AI value accrues to infrastructure and to software (the fund owns both, so it wins either way) | Only one wins; the other layer disappoints | | 4 | TSMC keeps operating without catastrophic geopolitical disruption | Tail risk realised → Taiwan Strait Risk | | 5 | Power becomes a binding constraint, re-rating IPPs | CEG/Vistra thesis is early/wrong | | 6 | Non-US franchises deliver despite policy/FX risk | EM ring drags rather than diversifies |
Why The Whole Is Greater Than The Parts#
The portfolio is engineered so holdings reinforce each other:
- When NVIDIA ships more GPUs, that is direct evidence for the TSMC, Micron, Marvell, Disco, Constellation, and Vistra theses simultaneously.
- When Microsoft/Amazon/Alphabet/Meta raise capex guidance, it is a forward indicator for the entire silicon-and-power spine.
- When the software names show AI-driven net revenue retention, it validates that the capex is monetising, protecting the whole chain.
This is why the fund reads earnings across holdings, not in isolation. See Cross-Holding Read-Throughs and Knowledge Graph.
Sell Discipline (thesis-level)#
The fund exits a name when: (a) the role it played is structurally impaired (moat breaks, chokepoint dissolves), (b) the theme matures and growth is fully priced, (c) capital allocation turns value-destructive, or (d) a better expression of the same theme appears. Position-level sell triggers live in each company note under Why PCA SOF Owns This Company.
Linked Notes#
- Related: Fund Overview · The AI Value Chain · Fund Philosophy & Process · Portfolio Construction Overview · Knowledge Graph
- Risks: Risk Master Note · AI Capex Cycle Risk · Taiwan Strait Risk
- Macro: Macro Master Note