(2026). Analyzing the Dynamics of Innovation Networks in Climate Technologies Using Large Language Models 2026.
Innovation networks are essential for advancing climate technologies, yet their structure and dynamics remain poorly understood. To address this gap, we use large language models (LLMs) to analyze 26 million LinkedIn posts, mapping a global network of 166,459 organizations and 442,250 collaborations (> 2 million direct partnerships) across 189 countries. Our dataset spans 27 climate technologies and 17 collaboration types, including demonstration projects, product launches, adoption, and equity investments. We find that, between 2020 and 2024, the structure of climate-tech innovation networks has changed substantially. Following the recent wave of industrial policy, many governmental organizations shifted from peripheral supporters to central orchestrators of global innovation networks, specifically for technologies with a lack of incumbent industry acting as system integrators (\eg geothermal energy, direct air capture, green hydrogen). Increases in the centrality of governmental organizations are associated with substantial expansions in domestic partnerships, ranging from 2.7 (concentrated solar) to 9.2 (batteries) new domestic partnerships per additional governmental partnership. At the same time, 62% of all global partnerships now arise from collaborations with government participation (\eg via public procurement and financing), revealing a substantial structural dependence which is highest for concentrated solar (87%) and lowest for electric vehicles (55%). Our LLM-based approach provides a scalable method for continuously monitoring the structure and dynamics of innovation networks beyond climate technologies.
