Core thesis
The investable AI chain is broadening. The scarce inputs are no longer only accelerators, but also power, cooling, network capacity and construction execution.
What the evidence says
Microsoft said calendar-year 2026 capital expenditure could reach about US$190 billion, while capacity remains constrained.
TSMC reported US$40.2 billion of Q2 2026 revenue and a 67.7% gross margin, with Q3 revenue guidance of US$44.6-45.8 billion.
The IEA identifies data centers, advanced manufacturing and electrification as renewed drivers of electricity demand in advanced economies.
Portfolio implication
Own the bottlenecks, but require visible order books, pricing power and funded expansion plans. Avoid treating all AI-adjacent capital spending as equivalent.
What would change our view
Faster efficiency gains reduce compute demand
Grid interconnection delays defer revenue
Capital intensity compresses free cash flow
Primary sources
Public facts are linked to primary sources. Portfolio weights and return assumptions are illustrative.