Extreme precipitation events (EPEs) are increasing in frequency and intensity in arid zones due to global climate change, profoundly altering soil hydrological conditions and nutrient availability, which in turn affect soil microbial community structure and ecological functions. However, the responses of soil microbial co-occurrence networks (an important indicator of community stability and interspecific interactions) to EPEs remain poorly understood. This study conducted a 3-year field manipulation experiment in the Gurbantunggut Desert, a typical arid zone in Central Asia, to explore the effects of different EPE magnitudes (light: 20 mm, moderate: 40 mm, heavy: 60 mm) on bacterial and fungal co-occurrence networks. High-throughput sequencing combined with network analysis showed that EPEs significantly restructured soil microbial networks: compared with the control (natural precipitation), moderate and heavy EPEs increased network complexity (node number, edge number, average degree) and modularity, while light EPEs had no significant effect. Bacterial networks were more sensitive to EPEs than fungal networks, with heavy EPEs increasing bacterial network connectivity by 42.3% but fungal network connectivity by only 18.7%. Soil moisture, available nitrogen, and organic carbon were the key drivers of network changes, explaining 35.6%, 22.1%, and 18.9% of the variation in microbial network structure, respectively. Keystone taxa in microbial networks shifted from oligotrophic groups (e.g., Actinobacteria) under natural precipitation to copiotrophic groups (e.g., Proteobacteria) under heavy EPEs. Additionally, microbial network complexity was significantly positively correlated with soil nutrient cycling potential (indicated by functional gene abundance). This study reveals the response patterns and driving mechanisms of soil microbial networks to EPEs in arid zones, providing a theoretical basis for predicting the stability of arid ecosystem functions under future climate change scenarios.