Accurate prediction of the part-level thermal history in the Laser Powder Bed Fusion LPBF process is critical for anticipating resulting material properties and defects, and is essential for effective process parameter and scanning strategy optimization. Existing numerical methods for simulation provide valuable insights, but at a computational cost for optimization. To address this limitation, data-driven approaches have recently emerged as a promising alternative for constructing efficient thermal surrogate models in additive manufacturing. E.g. [1] developed a PINN framework along complex scanning paths, while [2] integrates PINNs and RNNs to enable part-scale prediction. Despite advancement, the generalization across arbitrary scan paths and a natural extensions to 3D domains remains challenging. In this work, a DeepONet framework is introduced allowing predictions of the (latent) thermal state (LTS) of a part with arbitrary geometry with a given scan strategy. This framework combines a geometry aware network, allowing for interpreting boundary conditions and relevant geometrical features, and a Gated Recurrent Network (GRU), evolving an initial latent representation of the thermal state by considering the independent scan vectors as sequential events, acting as a learned latent time integrator. This formulation yields a neural operator capable of directly inferring the thermal field at arbitrary points in space and time on complex geometries given some provided scan strategy. It can be directly made applicable in 3D and its mesh-invariant design enables mesh-free predictions with essentially boundless resolution. The primary goal is to train the machine learning model using simulation data and initially validate and benchmark its predictions within the simulation environment. Subsequently, the model will be compared against experimental data. [1] Faegh, M., Sanvelly, R. R., Arabpoor, R., Rao, P., Mukherjee, T., & Haghighi, A., Physics-Informed Neural Networks for Thermal Modeling Transferable across Paths, Print Parameters and Beam Profiles, Additive Manufacturing, 105060, 2025. [2] Bostan, B., Vulimiri, P., Hinnebusch, S., Aryal, D., & To, A. C., STEAM: A Scalable Data-Driven Surrogate Modeling Framework for Part-Scale Scanwise Thermal Process Simulation of Laser Powder Bed Fusion, Additive Manufacturing, 105090, 2026.