Green AI Technologies: Patent Analysis of 63,000 Innovations Drives Climate Change Mitigation

Artificial intelligence increasingly offers solutions to the pressing challenge of climate change, and a new study by Lorenzo Emer, Andrea Mina, and Andrea Vandin, from the Institute of Economics and L’EMbeDS at Scuola Superiore Sant’Anna and DTU Technical University of Denmark, investigates this intersection through detailed analysis of over 63,000 patents related to ‘Green AI’. The researchers examine how innovation in this field evolves, identifying a clear transition from established areas like combustion engine technology towards emerging domains such as data processing, smart microgrids, and sustainable agricultural practices. Their work reveals a growing concentration of patent ownership among corporations, alongside a surge in the number of companies actively involved in Green AI innovation, and importantly, highlights a disparity between the potential climate benefits of certain technologies and the current economic incentives driving their development, suggesting a need for targeted policy interventions to accelerate progress. This comprehensive analysis provides crucial insights into the structure and evolution of Green AI, and its potential to deliver meaningful climate solutions.

Patent Analysis, Topic Modeling, and Sustainability

This research explores the convergence of patent analysis, natural language processing, and sustainability, revealing key trends in technological innovation. Scientists are increasingly using computational methods to analyze large collections of text, such as patents and research papers, to identify emerging technologies and understand research landscapes. This interdisciplinary approach combines expertise in data science, patent law, and specific scientific fields to drive innovation and address global challenges. The growing focus on sustainability highlights the increasing importance of environmental impact in scientific and technological advancements.

Central to this work is topic modeling, a technique used to discover the underlying themes within a collection of documents. Researchers employ advanced algorithms, including BERTopic, to identify key areas of innovation and track their evolution over time. These methods utilize sentence embeddings, created with models like Sentence-BERT, to capture the contextual meaning of patent abstracts and group them into coherent topics. Techniques like UMAP and HDBSCAN are used to visualize and refine these topic clusters, providing a nuanced understanding of the technological landscape. Researchers are applying these techniques to diverse areas, including healthcare, manufacturing, and neuropsychiatric diseases, to uncover innovation patterns and potential breakthroughs. They are also utilizing specialized tagging schemes, such as the EPO’s Y02-Y04S system for sustainable technologies, to refine their analysis and focus on climate-related inventions. A key goal is to ensure the reproducibility and optimization of topic modeling processes, allowing for reliable and consistent analysis.

AI and Climate Patent Topic Modeling

This study pioneers a novel approach to understanding innovation at the intersection of artificial intelligence and climate technologies through the analysis of approximately 63,000 patents spanning from 1976 to 2023. Researchers constructed this dataset by identifying U. S. patents classified as both artificial intelligence inventions and climate-related technologies, utilizing established classification systems and comprehensive patent datasets. The work leverages this comprehensive dataset to uncover key technological domains and assess their impact on subsequent inventions and market value.

A central methodological innovation involves the application of topic modeling, specifically using the BERTopic algorithm, to identify sixteen distinct thematic clusters within the Green AI patent landscape. This research harnesses the power of transformer-based models, employing BERT to generate dense sentence embeddings that capture the contextual meaning of each patent abstract. These semantic representations are then clustered to create coherent topical groups, revealing underlying technological domains. To interpret these clusters, researchers applied Term Frequency, Inverse Document Frequency, highlighting the most informative words within each group and ensuring interpretable topic extraction.

The study further assesses the impact of these identified domains by analyzing both forward citations, a widely used proxy for technological influence, and market value, estimated through indicators of stock market reaction to patent publication. By leveraging BERTopic’s built-in UMAP dimensionality reduction, scientists explored the semantic similarity between domains, identifying four broader macro-domains and providing a nuanced understanding of the relationships between different technological areas. This detailed analysis of temporal trends, assignee concentration, and technological content provides a comprehensive picture of innovation in the rapidly evolving field of Green AI.

Green AI Patents Show Sustained Growth

This work presents a comprehensive analysis of approximately 63,000 Green AI patents, revealing key trends in climate-focused innovation. Researchers documented a modest but steady growth in Green AI patent filings from the early 1970s, increasing from a few dozen patents annually to surpassing one hundred by the late 1990s. A significant surge occurred in 1998, with filings nearly doubling, coinciding with a period of optimism surrounding technological progress. Following a slight dip in 2000, patent activity rebounded, peaking in 2006 and again in 2010, before experiencing continuous and significant growth since 2010.

Despite economic challenges, including the 2008 financial crisis and legal rulings impacting patent eligibility, patenting continued to rise, reaching a new all-time high in 2023. Analysis of patent ownership reveals distinct waves of innovation over time. From the late 1980s to the 1990s, industrial control firms like Westinghouse and Fanuc, and then Bosch and Intel, dominated early Green AI activity. The 2000s saw a transition towards electrification and data-center optimization, with Honda, Ford, and Intel leading patenting efforts. More recently, the focus shifted to AI hardware and edge computing, with Intel and IBM remaining central actors, and Samsung and Qualcomm significantly expanding their presence. Researchers quantified market concentration using both the share of the top ten assignees and the Gini coefficient. The top-10 assignee share decreased, indicating broadening participation at the top, while the Gini coefficient increased, demonstrating a growing disparity in patent ownership across the full distribution of assignees, suggesting an increasingly long-tail structure within the innovation ecosystem.

Green AI Innovation, Patent Landscape and Trends

This research presents a comprehensive analysis of Green Artificial Intelligence, achieved through the examination of a substantial patent dataset encompassing approximately 63,000 patents. The team successfully identified sixteen distinct technological domains within this corpus, revealing a clear shift in innovation away from traditional combustion technologies towards areas such as data processing, microgrids, and agricultural water management. This detailed mapping of innovation trends provides valuable insight into the evolving landscape of climate-focused AI. The study also demonstrates a growing concentration of patent ownership among corporations, alongside an increasing number of patenting firms overall. Analysis of both green technology classifications and AI functional categories reveals that AI-Assisted Planning and Control currently dominates, alongside significant activity in AI Hardware development, suggesting a strong focus on industrial applications and energy efficiency. While some domains exhibit both substantial impact and market value, the research highlights instances where private incentives for innovation appear weaker, despite the relevance of these technologies for climate adaptation and mitigation.