Current modeling approaches for brain tumor dynamics, based on numerical solvers that simulate tumor growth using a given differential equation, are too time-consuming for clinical implementation. Recent data-driven approaches are able to emulate physical simulations, but typically fail to generalise across the variability of boundary conditions imposed by patient-specific anatomy. Here, the Menze group proposes a learnable surrogate for simulating tumor growth that maps biophysical model parameters directly to simulation results, while taking patient geometry into account. This neural solver is tested in a Bayesian model personalisation task for a cohort of glioma patients. Bayesian inference using the proposed surrogate leads to estimates analogous to those obtained by solving the forward model using a regular numerical solver. Owing to the near real-time computational cost, however, the method proposed here is suitable for clinical settings.