This study develops an orientation-aware interaction-based deep material network (ODMN)- based surrogate model for carbon-reinforced concrete (CRC) to overcome the computational challenges inherent in direct multiscale simulations of nonlinear solid mechanics. CRC exhibits an inherently heterogeneous microstructure—characterized by a concrete with intricate microlayer behavior combined with carbon fibers arranged in a woven architecture—which leads to complex microstructural interactions and pronounced nonlinear damage evolution under extreme loads. Such complexity makes traditional direct multiscale simulations computationally prohibitive. Consequently, our ODMN surrogate model efficiently captures these detailed behaviors while significantly reducing computational costs. Recognizing the need for high fidelity and efficiency, our approach leverages a microlayer technique to model the concrete —accurately capturing its detailed material response—and utilizes a woven structure to represent carbon fiber reinforcements’ anisotropic distribution and orientation. High-fidelity CRC Representative Volume Elements (RVEs) are generated to accurately reflect the complex microstructure precisely and serve as the training foundation for the ODMN. This network is specifically designed to encapsulate the intricate microstructural interactions and nonlinear responses, including damage evolution, at the microscale. By integrating the ODMN with a finite element (FE) framework within ANSYS, we establish an FE-ODMN model that robustly bridges the gap between the detailed microstructural behavior and the overall structural response. Experimental tests under uniaxial tension and compression confirm that the FE-ODMN model replicates the reactions observed in direct numerical simulations while substantially reducing computation time. Moreover, the model has been successfully applied to seismic load simulations and damage evaluation in CRC structures, providing critical insights for designing safer, lighter, and more sustainable construction systems. This work demonstrates the promising potential of combining advanced material modeling with machine learning techniques for practical engineering applications.