Rapid solidification and the complex thermal histories inherent to the laser powder bed fusion (LPBF) additive manufacturing (AM) present formidable challenges in modeling microstructural evolution, particularly at the submicron scale, where microsegregation plays an important role. This work introduces an open-source cellular automaton (CA) solidification model implementation, muMatScale, specifically developed to simulate the formation of grain and subgrain microstructures during LPBF. The model effectively captures essential rapid solidification phenomena, such as alloy segregation and solute diffusion, using externally imposed temperature profiles, which are crucial for modeling non-equilibrium solidification dynamics in LPBF. To address the substantial computational cost, muMatScale is parallelized utilizing MPI to distribute computations across nodes and leveraging OpenMP to offload computational kernels to GPUs, in combination with precomputation strategies for interface cells. This approach enables efficient, scalable simulations from local workstations to exascale high-performance computing platforms. To evaluate the model’s performance and accuracy, predictions of grain and subgrain microstructures within single LPBF melt tracks were performed. Nucleation behavior, subgrain characteristics, and microsegregation analysis were thoroughly investigated and compared against experimental benchmarks. Notably, full-scale single-track simulations encompassing 4.5 billion grid points at 0.1 μm resolution were completed in under 10 mins of computational time with 2000 CPUs plus 2000 GPUs on SUMMIT hosted at the Oak Ridge Leadership Facility (OLCF), underscoring the model’s computational efficiency. Additionally, further applications of the model to laser remelting scenarios are discussed, providing deeper insight into the mechanisms driving grain refinement. These findings demonstrate that CA-based solidification modeling, enhanced by GPU-accelerated computing, provides a robust and scalable computational framework. The presented methodology could further advance the fundamental understanding of complex solidification processes in AM and holds significant potential for guiding alloy design, optimizing LPBF process parameters, and achieving desired microstructural control in advanced materials.