During Run 2, the simulation of physics events at LHCb has taken about 80% of the distributed computing resources available to the experiment. The large increase in luminosity and trigger rates with the upgraded detector in Run 3 will require much larger simulated samples to match the increase of collected data. About 50% of the overall CPU time in the simulation of physics events is spent in the calorimeter system. In this talk we describe the solution adopted in Gauss, the LHCb simulation software framework, to avoid the need to simulated the calorimeter response to particules with the Geant4 toolkit, instead inserting the corresponding hits in a with a fast simulation. Two paths are being pursued to simulate the hits, based on libraries of pre-simulated energy deposits, or using machine-learning techniques for their generation at runtime. We discuss the performance of both approaches and their readiness in view of the start of Run 3.