Cascading supply shocks with a GPU-friendly input–output network model
Modern economies aren’t just “a bunch of industries.” They’re networks of dependencies.
A shock in one upstream node (energy, fertilizer, chips, shipping, critical metals) doesn’t stay local — it propagates through intermediate inputs, forcing downstream sectors to reduce output, which then feeds back upstream as reduced demand.
This post is a first-pass, matrix-first model of that cascade, designed to be fast (and GPU-friendly when PyTorch+CUDA is available).