AIPNET — AI-Generated Production Network

AIPNET uses a two-step "build-prune" approach with an ensemble of prompt-tuned generative AI classifications to construct a production network over 5,000+ product nodes.

In the "build" step, generative AI provides an initial distribution of edge predictions representing input-output relationships between products. The "prune" step then re-evaluates all edges to improve precision. The resulting network captures directed relationships — which products serve as inputs to which other products in the global production process.

The dataset enables research on production network spillovers, global trade structure, on-shoring dynamics, industrial policy, and other shifts in the global economy. We document shifts in the network position of products and countries during the 21st century, and validate the network using the natural experiment presented by the 2017 blockade of Qatar.