Qapital Research

Vertical Thesis

Draft

AI

From compute to inference to deployment — where the durable moats are forming.

2

Companies in Coverage

5

Sub-themes

0

Trail signals (90d)

2026-05-20

Thesis updated

Our thesis

Draft. Aschenbrenner-style framing: compute → algorithms → unhobbling → systems. We track who owns the chokepoints (EUV at ASML, foundry at TSMC, GPU at NVIDIA) and who's positioned at the inference + deployment layer.

Sub-themes we track

EUV / lithography

ASML monopoly; multi-decade durable.

Leading-edge foundry

TSMC structural; Samsung + Intel as wildcards.

GPU / accelerators

NVIDIA platform lock-in vs custom silicon.

Inference economics

Where margins land when training arms-race normalises.

Vertical AI applications

AI in defense (cross-vertical), AI in biotech (cross-vertical), enterprise.

Indicators we monitor

  • Hyperscaler capex commentary
  • EUV tool shipments + ASML book/bill
  • Foundry capacity utilisation
  • Token cost trajectory (open + closed models)

Recent trail signals

See trail →
No smart-money signals yet. Trigger ingestion via /admin/run-13f.