Multiple-input multiple-output (MIMO) communications will continue to be a cornerstone of modern wireless networks, thanks to the spatial multiplexing and beamforming gains enabled by large antenna arrays. However, acquiring precise downlink channel state information (CSI)—essential for effective precoding for large-scale antenna systems—involves substantial signaling overhead. In this paper, we propose a CSI acquisition framework that leverages digital twin to generate user-specific channel prior at the base-station, and reduces the CSI acquisition overhead. To achieve a low-complexity operation, we propose a multi-tiered digital twin architecture and complexity reduction methods, which reduce the ray-tracing runtime with minimal performance loss. Through comprehensive evaluations in an urban scenario, we demonstrate that the proposed framework can effectively reduce CSI acquisition overhead, and decrease the digital twin compute requirement by 56% while maintaining healthy cell coverage and spectral efficiency at the user.