No Priors · Coverage Brief
The AI infrastructure cascade —
from capital deployed to the materials that bottleneck it.
~$1.1T of AI infrastructure capital is being committed in 2026. It moves through seven CapEx categories, narrows into eighteen semiconductor value-chain stages, and ultimately depends on eight critical materials concentrated in two or three countries. Each layer is more concentrated than the one above it.
- Total CapEx ’26
- ~$1.05T
- Hyperscaler share
- ~60%
- Cohort names
- 13
- Theses tracked
- 7
- As of
- Mar 2026
click any element for detail · ★ marks coverage cohort
01
Capital Sources — who is writing the checks
Six pools of capital, sized to FY26 commitment scale. Hyperscalers alone account for ~60% of total deployment.
~$965Bannual commitments
02
Where the capital lands — the seven CapEx categories, proportional
The flowbar below is sized to each category's % share of total spend. Click a segment to expand its sub-categories.
$1.05Ttotal FY26 budget
03
The semi value chain — eighteen stages, three of them bottlenecks
Design → Fabrication → Packaging & Beyond. Stages flagged with ▼ are the binding constraints (TSMC CoWoS, ASML EUV, HBM).
$350–450Bof total CapEx flows here
A · Design → Fabrication → Equipment →
B · Materials → Packaging → Memory →
C · Test → Specialty silicon → Substrates →
00
Critical materials — what AI ultimately runs on
Each bar shows the direct AI-infra share of total end-use demand. Most materials are dilute. Hafnium and germanium are the exceptions.
8 inputsof which 3 have acute concentration
Risk concentration:Acute (>90% single country)ElevatedStrategicBulk / diversifiedEnd-use mix:Direct AI infraIndirect / adjacentUnrelated demand
Working theses · what the visualization argues
Seven observations the cascade makes legible
click a thesis to expand