Climate change is already significantly affecting the frequency of wildfires in most regions of the world, and the risk of wildfires is expected to amplify further with global warming. Accordingly, there is growing concern about the mechanisms and impacts of extreme fires. In this study, a coupling of the Weather Research and Forecasting model and the Rothermel Fire model (WRF-Fire) is employed to reproduce the spread of fire within the national boundary of inner Mongolia from 21 to 27 May 2009. Simulations were run with or without feedback from fire-to-atmosphere models, and the study focused on how the energy flux of simulated fires changes the local meteorological environment. The coupled simulation could reproduce the burned area well, and the wind speed was the dominant factor in the fire spread, with a maximum value no more than 6.4 m/s, when the terrain height changes little and the proportion of grassland is low. After the feedback, the propagation speed of the fire accelerated, accompanying the release of latent and sensible heat, and local circulation formed near the front of the fire, leading to a convergence and divergence zone in the downwind area. It is worth noting that during a period of more than 140 h of simulation, the area of the fire field increased by 17% from ignition time. Therefore, considering the fire–atmosphere interaction is necessary for accurately predicting fire behavior.