Local strong winds could be hazardous to street structures and pedestrians, and they may affect outdoor activities and traffic safety. The formation of local strong winds is associated with both mesoscale weather and underlying surface condition, particularly over urban areas where a large number of tall buildings exist. The building-resolving simulation of strong winds over megacities at given atmospheric conditions is challenging due to the limitations in numerical models and computational resources. Here we investigate a typical event of strong winds induced by a cold surge in Guangzhou using an advanced local weather prediction system, in which a mesoscale model is downscaled to a Computational Fluid Dynamics (CFD) model with Large-Eddy Simulation. Both CFD simulations and observations suggest that during the cold surge, mean wind speed and high-frequency disturbances enhanced significantly in the urban area, particularly at the top of the urban canopy. There are also evident differences between the old and new districts with different buildings characteristics. Strong winds and disturbances occurred over a large area of the old district with dense low-rise buildings and its downstream open areas. Over the new district with tall buildings, although the regional mean wind speed reduced, local strong winds occurred in the main streets parallel to the wind direction. In particular, high-rise buildings induced obvious veridical circulations that could lead to downward movements of wind disturbances and caused strong near-surface winds. Such eddies of building-induced strong winds could propagate up to several kilometers downstream. They were reinforced when passing other high-rise buildings such as the Canton Tower. Local strong winds also occurred at the flanks of tall buildings aligned along the Pearl River that is perpendicular to the ambient wind direction. The blocking effect of those buildings resulted in an along-river air flow, which then flowed out via the gaps between the buildings, leading to local strong winds. These findings have important implications for understanding the fine-scale structures and formation of local winds in megacities, which will be helpful to improve the prediction of urban winds.