Motivated by the global energy transition and subsurface energy resource (oil, gas, coal-bed-methane, geothermal, etc.) development, subsurface hydraulic fracturing technology is undergoing a paradigm shift from traditional experience-driven approaches to data- or intelligence-driven techniques. This work systematically elaborates on the connotation, recent practices, and future trends of artificial intelligence (AI)-driven subsurface hydraulic fracturing technology. This work proposes a three-step technical evolution framework centered on data-driven→dynamic optimization→autonomous decision-making. Recent key practices in the framework are also introduced, including smart characterization and optimization of hydraulic fracturing, smart forecast of production operation after fracturing, and real-time regulation of entire fracturing-to-production lifecycle. The smart characterization of three-dimensional fracture propagation is achieved by constructing the Dy-Fracture-Net model. A dual-model collaborative architecture is developed to enable real-time warning and smart optimization during the fracturing process. Furthermore, the innovative Dy-Production-Net network is designed to predict the dynamics of post-fracturing reservoir parameters and production. Through integrating with intelligent optimization algorithms, a real-time regulation system encompassing the entire fracturing-to-production workflow is formed. To address the bottlenecks such as the lack of downhole monitoring data and insufficient model interpretability, future efforts are recommended as follows: miniaturization of multimodal perception agents, self-interpretability of mechanism-data fusion modeling, and autonomous closed-loop control. The findings of this work provide theoretical support and practical pathways for realizing the future AI-driven subsurface fracturing technology, holding significant strategic importance for advancing the digital transformation of the oil and gas industry.