A discussion brief on the Taiwan AI infrastructure wedge, backed by four research exhibits mapping the physical and economic layers behind AI compute.
Build the Taiwan AI Inference Factory
7-minute synthesis of the build thesis: 10 MW, ~$300M, 5,000 GPUs in a building, selling tokens. Where the unit economics need an anchor customer, how the Macquarie relationship sequences across funding stages, where the Taiwan pitch is strongest, and the five questions today has to answer.
Read the brief →Supporting Exhibits · The Receipts
NVIDIA GPU Deployment Stack — Taiwan Constraint Map
14 layers from land permitting to wafer fabrication, with the binding constraints at each level.
AI Data Center Energy Stack — Plain English Map
How a watt at the substation becomes a token at the GPU, explained without jargon.
NVIDIA Server Types — DGX, HGX, MGX, EGX
What each chassis family is for, who buys it, and how the SKUs actually differ.
Real Estate, Land, Oil & Gas → AI Energy
Step by step: how legacy land, mineral, and power assets get repositioned into AI-ready capacity.