The snap-through phenomenon occurs in many elastic structures and is characterized by a rapid change in deformation (a ‘snap’) between two equilibrium states. For a thin shell held by two opposing hedges and subjected to an adequate load, this type of buckling leads to a significant deformation of the structure. Despite the success of Finite Element Methods (FEM) in modelling such phenomena, there are clear limitations regarding real-time applications - a requirement for immersive Virtual/Augmented/Mixed Reality (VR/AR/MR) applications involving haptic feedback in industrial envirnments. To overcome FEM speed constraints, we propose the use of Thermodynamics Informed Graph Neural Networks (TIGNN). TIGNN is a physics informed graph neural network, that guarantees the fulfilment of 1st and 2nd laws of thermodynamics via a Hamiltonian approach, the GENERIC framework, able to describe a physical system in terms of its dissipative (entropy dependant) and conservative (Hamiltonian dependant) components. Using a FEM database of bending thin plates, a TIGNN model was successfully trained and subsequently applied to unseen geometries. This approach allowed significant speed gains over the FEM simulations and was able to produce realistic outputs, adequate for VR/AR/MR industrial applications.