Numerical simulation of large-scale fluidized beds presents significant challenges for both scaling up the simulations and evaluating their performance, as full-scale computational fluid dynamics (CFD) simulations incur substantial computational time and cost. To overcome these challenges, we propose a novel approach that combines CFD simulations with data-driven deep-learning models to predict complex hydrodynamics. In this study, we trained two machine learning (ML) algorithms, namely the Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), using spatiotemporal data obtained from CFD simulations. Impressively, the trained models accurately predicted future events for successive time steps, exhibiting excellent agreement for mean voidage, gas velocity, and particle velocity. Among the models utilized, LSTM outperformed the other. This demonstrated its superior ability to capture spatiotemporal CFD data while requiring less computational time and power. The trained models provide a reliable means to extrapolate desirable mean particle velocity and voidage values within fluidized beds. These findings hold promising opportunities for enhancing the design and operation of gas-solid fluidized bed reactors, potentially leading to a significant reduction of up to 57–68% in simulation time.