Variability and intermittency of green power is the main obstacle for its utilization. Different from other power consumption, due to the distributed nature, load balancing on geographical range can be used to dispatch computing tasks to the data centers with abundant renewable energy. The premise of this new strategy is: there is always abundant green power at some of the renewable power portfolio, yet this is not always the truth. The stable availability of renewable energy is built on the compensation of different power plants, but due to the constraint of constructed data centers and the on-site powering strategy, this compensation effect has not been fully explored. In this paper, we propose a solution for Renewable Energy Portfolio Optimization (REPO) problem, and take wind farm location selection as an example to stabilize the variable and intermittent wind power. The simulation is conducted based on the real-world climatic traces from 607 candidate wind farms. The optimal renewable energy portfolio can provide stable wind power supply at the price of 70 USD/MWh. When simulated with Google workload trace of May 2011, with installed capacity 4 times of average power demand, REPO can save 59.5% of energy while a combination (on Google data center locations) without consideration of mutual compensation could only save 30%.