Critical raw materials (CRMs), such as tungsten and graphite, are vital to modern technologies but pose supply risks due to concentrated mining sources. For instance, around 75% of the global tungsten and graphite supply is mined in China. While upstream dependencies are well known, their downstream propagation along the value chain remains poorly understood.

Our goal is to quantify critical dependencies across entire CRM value chains—from raw materials to end-use applications—while accounting for granular intermediate production steps. In addition to revealing systemic risks from direct and indirect trade dependencies across multiple product categories, our approach helps assess the feasibility of self-sufficiency through recycling.

We present a method to automatically construct CRM value chains at the six-digit Harmonized System (HS6) level by combining AI-augmented analyses with international trade data. Using multiple seed nodes (e.g., tungsten ores), we identify production network links with large language models (LLMs) and validate them via trade correlation metrics. Applied to tungsten, our method yields a network of 50 products and 75 links across multiple sectors. Validation against a manually researched phosphate rock network shows precision and recall rates of 100% and 50–90%, respectively, for the core component of phosphatic fertilizer production, which accounts for 90% of phosphate rock use.