In the context of the upcoming 4th generation industrial revolution (industry 4.0), mechanical failures in the cyber-physical systems have huge financial impacts. The IT industry like Google, Facebook, Microsoft, etc. mostly depends on the Datacenters (DCs) to assure the quality of services. The equipment of the DC including the power supply system and the computational resources are sensitive to supplied power quality, thus predictive maintenance is needed to prevent failures and limit financial losses. The predictive maintenance assures operational security based on the monitored data that can characterize the failures of the physical machines, and also ensures the maximum return of the capital investment by prolonging the useful life of the equipment. The size of the monitored data typically occupies large memory space that can compare with “big-data” nowadays. Thus, the big-data-sized monitored data analysis is an additional computational challenge to characterize the failures of physical machines, hence, schedule the predictive maintenance. However, characterizing the failure and repair time of the major components based on the measured data is still a challenge that is the goal of this paper. Meanwhile, the revenue of the business also largely depends on the accuracy of predictive maintenance in general. In this paper, a predictive maintenance approach is presented based on the stochastic failure time of the major components of the DC. Additionally, the business challenges for predictive maintenance considering industry 4.0 are also analyzed in this paper.