Background: Data centers are major energy consumers, with cooling systems accounting for a significant portion of operational costs and carbon emissions. Recent advances in artificial intelligence (AI) offer promising strategies for optimizing thermal management. However, a comprehensive overview of publication trends, thematic evolution, and collaborative networks in AI-enabled data-center cooling is lacking. Methods: We conducted a bibliometric analysis of 311 peer-reviewed articles retrieved from the Web of Science database, covering the period 2015–2024. Using VOSviewer and CiteSpace, we mapped annual publication trends, country-level contributions, co-authorship networks, co-citation structures, and keyword co-occurrence clusters. Burst-detection analysis identified emerging topics and their temporal dynamics. Results: Annual publications grew from 2 in 2015 to 80 in 2024, reflecting accelerating research interest. China (28 %) and the United States (25 %) led author-country instances, with strong bilateral collaborations. Co-occurrence analysis revealed 12 thematic clusters (modularity Q = 0.72, silhouette = 0.91), the largest of which focused on “reinforcement learning,” followed by “energy efficiency” and “free cooling.” The ten most frequent keywords—e.g., “data center cooling,” “machine learning,” “thermal management”—underscored the dominance of AI-driven control strategies. Burst analysis showed an early focus (2015–2018) on “neural network” and “genetic algorithm,” transitioning (2019–2024) to “reinforcement learning,” “deep reinforcement learning,” and, more recently (2022–2024), “digital twin” and “meta-learning.” Discussion: The field has matured from initial explorations of classical machine-learning algorithms to advanced, adaptive reinforcement-learning frameworks and simulation-based optimization. Real-world implementations—such as RL-based controllers deployed by major technology firms—demonstrate tangible energy savings. Future research should pursue multiobjective optimization, enhance digital-twin fidelity to bridge the sim-to-real gap, develop interpretable AI policies, and incorporate non-English literature and industry reports to broaden the knowledge base.
Wei et al. (Wed,) studied this question.