Hollow concrete structures with customized cavities containing infill, such as clay-filled morphed honeycombs, not only retain structural stiffness but also contribute to decarbonization efforts due to clay’s low embodied carbon and recyclability. However, evaluating these designs requires laborious simulations, like Finite Element Methods (FEMs), which impedes efficiency. To tackle this issue, we propose a hybrid model that integrates a Graph Neural Network (GNN) and a physics-informed loss function for swift and precise structural analysis of these structures. In this hybridization, the GNN excels in learning the complex geometries of the clay-filled morphed honeycombs, while the PINN offers interpretability by aligning with physical principles. Specifically, we simplified the simulation by modeling the structures using beam elements instead of shell elements, serving a dual purpose. First, beam elements offer computational efficiency in simulations. Second, using traditional physics-informed loss functions based on Partial Differential Equations (PDEs) for shell elements is often computationally heavy. Instead, we chose a simpler approach by using Hooke's law, which better matches the behavior of beam elements and is less demanding in terms of computation. This simplification of simulation is validated by mechanical test results for regular concrete honeycomb components with and without clay infill, each with porosities of 33%, 45%, and 55%, cast using 3D-printed molds. The hybrid model proposed in this study significantly outperforms the benchmark GNN in accuracy, reducing the Mean Relative Error Rate (MRER) of nodal displacement from 12.4% to 7.5% and structural stiffness from 6.2% to 5.5%.