AI Lab Ownership Map
The frontier-AI-model labs are mostly private, but the fund owns them indirectly through its hyperscaler holdings. This maps that exposure. Companion: Ownership Network Map.
The Lab → Backer → Holding Chain#
FRONTIER LAB PRIMARY BACKER(S) PCA SOF HOLDING (indirect exposure)
──────────── ──────────────── ──────────────────────────────────
OpenAI ───▶ Microsoft (~49% econ.) ───▶ Microsoft
Anthropic ───▶ Amazon + Google ──▶ Amazon + Alphabet
Google DeepMind ───▶ Alphabet (wholly owned) ──▶ Alphabet
Meta AI / Llama ───▶ Meta (in-house) ──▶ Meta Platforms
(Scale AI) ───▶ Meta (stake) ──▶ Meta Platforms
xAI / others ───▶ not held ──▶ (none)
What This Means For The Fund#
- The fund owns the model layer without picking a model. Whether OpenAI, Anthropic, Gemini, or Llama "wins," PCA SOF has exposure through Microsoft, Amazon, Alphabet, + Meta Platforms, mirroring its "own the theme, not the name" doctrine (Competitive Landscape Map).
- The labs drive the capex that flows to the silicon spine (NVIDIA, TSMC, Micron Technology, Marvell Technology) + power (Constellation Energy, Vistra). Lab progress = spine demand.
- The labs are also a risk: if model progress stalls or a winner-take-all dynamic emerges, the value distribution across the fund's hyperscalers shifts.
The Compute Counter-Dependency#
The labs depend on the fund's holdings as much as the holdings benefit from them:
- OpenAI/Anthropic train on NVIDIA GPUs in Microsoft/Amazon clouds.
- Those GPUs are made by TSMC, fed by Micron Technology HBM, networked by Marvell Technology, powered by Constellation Energy/Vistra.
- Tencent's Hunyuan + China AI are constrained by US export controls limiting NVIDIA access, a key China-AI nuance.
So the AI-lab layer sits above The AI Value Chain, pulling demand down through it.
Linked Notes#
- Ownership Network Map · The AI Value Chain · Knowledge Graph · AI Ecosystem Map
- Holdings: Microsoft · Amazon · Alphabet · Meta Platforms · NVIDIA · Tencent
Linked from 5 notes