The high computational cost of physics-based simulations is one of the main challenges in implementing Digital Twins for complex systems such as metal additive manufacturing (AM). Modelling the extremely rapid and multi-physical fields phenomena occurring during MAM processes such as laser powder bed fusion (LPBF) demands very fine time- and space-discretisation in classical simulation approaches, namely the finite element method (FEM). This study explores applying physics informed neural networks (PINNs) to thermal analysis of the LPBF process, through which reliable high-speed and real-time outcomes can be achieved. An unsupervised learning strategy was employed to (parametrically) solve the heat transfer equation for the LPBF process. The trained network calculates the temperature profiles and the melt-pool dimensions during the LPBF process for any given set of material's thermal properties and process conditions at practically zero computational cost. The reliability of the PINNs outcomes was verified based on several benchmark equivalent finite element simulations.