Abstract: Against the backdrop of the in-depth implementation of the global "dual carbon" strategy and the comprehensive advancement of the national green power direct connection policy, green power direct connection has become the core path for the low-carbon transformation of intelligent computing centers. However, traditional optimal configuration methods for green power direct connection are generally based on the assumption of rigid loads and fail to leverage the load regulation potential of intelligent computing centers. To address this issue, this paper first investigates the load characteristics and reliability requirements of intelligent computing centers, classifies their loads into adjustable loads and non-adjustable loads, and quantitatively derives the overall load regulation boundary of intelligent computing centers. On this basis, a load-configuration two-layer iterative solution algorithm is proposed to realize the optimal configuration of green power direct connection for intelligent computing centers considering load elasticity. Finally, a case study is conducted on a 100MW intelligent computing center in a certain province for verification. The results show that compared with traditional methods, the method proposed in this paper can increase the self-consumption rate of new energy by 20.75%, reduce the curtailment rate by 9.2%, and lower the average electricity price over the entire life cycle by 0.0525 yuan/kWh, providing technical reference for the engineering practice of green power direct connection in intelligent computing centers.