The semiconductor industry is critical to modern electronics, but its supply chain is notoriously opaque. In this study, we use large-scale web data to map supply chain relationships across 1,857 semiconductor companies over ten years, collecting over 170 million unique web pages.

By analyzing inter-company links and applying natural language processing, including large language models, we identify and classify business relationships such as supplier-buyer ties, partnerships, and ownerships. Our models achieve high accuracy, and the resulting supply chain network exhibits a fat-tailed degree distribution, consistent with real-world production networks.

By studying the evolution of the network over time, we can show how the position of regions and companies within them in the semiconductor value chain changes (e.g., regions moving up or down the value chain, becoming more or less central) and how this, in turn, relates to changes in regional output and comparative strength.

While focused on semiconductors, this framework is broadly applicable and offers a new way to monitor industrial dynamics and supply chain risks.