Wireless networks are evolving from communication-centric systems to environment-aware intelligent infrastructures, characterized by dense deployments, dynamic operation, and tight integration with immersive applications such as sensing and robotics. This shift motivates a new paradigm—spatial intelligence for wireless (SI4Wireless)—where wireless systems learn, represent, and reason about their underlying 3D geometry to achieve scalable, adaptive, and generalizable functionalities. This talk will outline a pathway toward spatially intelligent wireless infrastructures for 6G and beyond. We will start by applying 3D Gaussian splatting, a prevalent technique for 3D scene reconstruction, for wireless field reconstruction (WRF-GS), and then develop a unified radio-optical radiation field for 3D radio map construction (URF-GS). We will also introduce the latest development in foundation models for cross-scene radio fields (Point2Radio), enabling generalization across diverse scenarios.