Abstract: This study employs the Monte Carlo simulation method to compare the power supply reliability of two configurations for backup power in Artificial Intelligence Data Centers (AIDC): diesel generators and a third utility power source. By constructing a failure probability model incorporating random events such as utility power interruptions and generator startup failures, large-scale random sampling is conducted to quantify the annual power supply availability metrics of each configuration. The simulation results indicate that the third utility power source solution achieves comparable reliability to the diesel generator approach while effectively mitigating engineering challenges such as high land footprint, complex daily maintenance, and low operational utilization rates. This provides quantitative decision-making support for the green and intensive transformation of data center power architectures.