The ATLAS physics program relies on very large samples of simulated events. Most of these samples are produced with GEANT4, which provides a detailed simulation of the ATLAS detector. However, this simulation is very time consuming, especially in the calorimeters. As a result, many physics analyses are limited by the available Monte Carlo statistics and will be even more so in the future. To solve this problem, sophisticated fast simulation tools have been developed and they will become the default tools in ATLAS production in LHC Run-3 and beyond. Fast calorimeter simulation tools employ machine learning algorithms at various levels, either in conjunction with classical parametrization approaches, or on their own (e.g., Generative Adversarial Networks). In this talk, I will describe such tools and demonstrate their importance for future applications in ATLAS. Further powerful machine learning applications used in recent physics analyses will also be discussed.