Origami-inspired structures are increasingly critical for developing lightweight, deployable components in autonomous robotics and aerospace engineering. However, simulating their non-linear mechanical behavior and complex folding patterns remains a major computational bottleneck. This research presents a high-performance simulation framework built on JAX, designed to accelerate the design and optimization of three fundamental origami geometries: Miura-ori, Yoshimura, and Kresling. By leveraging JAX’s just-in-time (JIT) compilation and automatic differentiation, we developed a physics-informed model that incorporates kinematic constraints with spring and bending energies. To further enhance efficiency, we trained a supervised graph neural network (GNN) on simulated deformation data to serve as a fast surrogate model. Results indicate that the JAX-based framework achieves over 100x speedups compared to traditional NumPy-based baselines, enabling real-time simulation of large-scale patterns. Statistical validation confirms that the GNN surrogate accurately predicts complex deformations in milliseconds with minimal loss. This work provides a scalable foundation for the automated design and control of deployable robotic systems, where rapid iteration and differentiable physics are essential for performance optimization. By bridging the gap between high-fidelity physics and real-time execution, this framework facilitates the discovery of next-generation autonomous structures with mathematically guaranteed structural integrity.