"Given systemic risks and geopolitical sensitivities, accurately mapping the multi-relational semi-conductor network at the firm-level is essential for anticipating choke points, assessing economic inter-dependencies and informing industrial or trade policy." Christian Diem, University of Oxford
- Complexity economics groups in Vienna and Oxford find a new way to map the global semiconductor network;
- 78% of newly identified supply chain links for the semiconductor industry were not mapped in comparison data;
- Methodology offers a new tool for policymakers and researchers aiming to assess resilience and navigate global shocks, and a blueprint for a new approach to economic modelling.
A new methodology for mapping the supply chain network for the critical worldwide semi-conductor industry has found connections that traditional approaches could not - with 78% of newly identified links not mapped in comparison data - paving the way for more effective industrial and trade policy.
The semiconductor industry has been at the centre of a 'triple whammy' of disruption since the turn of the decade. The global chip shortage of 2022-3 (triggered by the pandemic-related lockdowns and rapid shift in consumer demand) was quickly followed by geopolitical tensions in 2024-5 (impacting supply); and then the emergence of AI and resulting demand surges of GPUs by NVIDIA (a revolution in the sector).
During these shocks, Governments have struggled to manage the risk and vulnerabilities of the supply chain using their traditional models and approaches.
However, using a new Large-Language Model-based approach (LLMs) to extract relations from the open web archive common crawl, a team of researchers including Seyda Köse, Assoc. Prof. Peter Klimek, Dr Christian Diem and others from complexity economics groups in Vienna and Oxford have managed to construct a dataset of 1,300 linked firms at a finer grained level than typically considered. The resulting data-set can substantially enrich traditional data sources relied upon by governments, with 78% of the links identified not present in the comparison commercial database.
The work is the first comprehensive approach to reconstructing multi-relational firm networks at scale. The study has established that the open-web, processed at scale, provides a vital repository of economic intelligence.
Results ‘challenge previous assumptions’ about the shape of the network
The dataset built by researchers captured rapid structural adjustments in the supply chain and challenged a number previous assumptions.
Insights include:
- Semiconductor shortages of 2021-2022: One might have expected these crises to be associated with an increased dependency on a small number of key firms, or the emergence of bottlenecks. However the opposite is true, suggesting firms formed or reported relationships with firms in a different way before and after the chip crisis.
- Shifting geopolitical landscape of 2024-25: Our results capture rapid structural adjustments in response to geopolitical shocks. Following an escalation of trade restrictions and export controls targeting the semiconductor sector, we observe a sharp collapse in links connecting Chinese firms to US partners, and distinctive trajectories for European, US and Japanese firms.
- Structural shifts associated with the generative AI boom (2022-): The widespread adoption of ChatGPT in 2022 generated an abrupt and highly specific surge in demand for parallel processing hardware. Our novel supply chain reconstruction method confirms NVIDIA's ascent to one of the nexus suppliers in the semiconductor industry.
Method is 'generalisable and sector agnostic'
Research lead Şeyda Köse of the Supply Chain Intelligence Institute Austria (ASCII), said that a key benefit of the research was that it's methodology was generalisable, making it a useful tool for mapping the supply chains in other other sectors.
"The semiconductor industry was our proving ground, but the methodology is sector-agnostic.
"Any industry where firms maintain a public web presence and disclose business relationships online is, in principle, mappable with this approach, and given how little firm-level network data is openly available globally, that scope is significant."
LLM approach helps construct 'vital repository of economic knowledge'
Commenting on the results, which are published in an INET Oxford Working Paper, Christian Diem of the University of Oxford said:
"This study establishes that the open web, when processed at scale, serves as a vital repository of economic intelligence.
"By applying Large Language Models to millions of archived webpages, we have constructed a validated, dynamic multi-relational map of the global semiconductor supply chain that overcomes the latency and opacity of traditional administrative and business intelligence data, and importantly adds information that is complementary to these traditional data sources.
"Our results confirm that this approach captures essential industry dynamics and tracks the impacts of changes in the geopolitical landscape and the ascent of AI. The reconstructed network accurately traces the contraction of supply links during the 2021 chip shortage and the rapid ascent of AI specialised firms such as NVIDIA, to central positions in the global ecosystem.
"These findings demonstrate that firm-level web data contains a robust economic signal that complements existing financial disclosures.
"Ultimately this framework offers a blueprint for a new approach to economic modelling that can easily be applied to other sectors.
"As supply chains become increasingly complex and geopolitically sensitive, the ability to generate real-time bottom-up visibility into inter-firm dependencies offers a crucial tool for policymakers and researchers aiming to assess resilience and navigate global shocks."
Key Links
