Deep learning has been widely recognized in the field of CO₂ geological utilization and storage applications. With the development of deep learning algorithms, intelligent models are gradually able to improve multi-source, multi-scale and multi-physicochemical mechanism barriers with high-fidelity solutions in practical applications. In this perspective, an overview of the traditional and state-of-the-art deep learning architectures involved in CO₂ geological utilization and storage is outlined in terms of evolutionary trajectories. Meanwhile, the favorable directions and application scenarios of different deep learning algorithms for geo-energy intelligence modeling are summarized. Moreover, further insights into the future direction of deep learning burgeoning architectures in this field are provided. The physics-guided deep learning, explainable artificial intelligence, and generative artificial intelligence are expected to deliver more accurate solutions for information extraction and decision support within the CO₂ geological utilization and storage communities.