In this paper, we propose a collaborative comparison between the online/offline algorithms for energy demand power saving purposes. Based on the Gaia Power model, resource-buffering algorithms are considered a practical peak-shaving model to effectively minimize the excessive power request. Although the algorithmic infrastructure is focused on a battery, this energy demand power saving problem is analogous to traditional demand and supply problem. In light of the similarity, we implement various machine-learning techniques, including Multiple-Layer Perceptron(MLP), Radial Basis Functions(RBF), Recurrent Neural Networks(RNN) to the identical peak-shaving model problem. In addition, the traditional naïve forecasting model and linear regression will also be discussed. Our findings suggest that the neural networks not only show faster demand smoothing in power saving algorithms, but being a nature of online algorithms is also theoretically and statistically more efficient than resource buffering algorithm and DCEC technology.