2026/07/19 by Mingxu Yang, Weiqi Zhang, Hui Geng +1
#math.OC
The exponential growth of AI is accelerating the deployment of data centers (DCs), placing unprecedented strain on power infrastructures. In response, major IT corporations are increasingly adopting on-site solar generation to reduce grid dependence and meet sustainability targets. However, the true impacts of this strategy remain ambiguous. While DCs are flexible assets capable of temporal load-shifting, anchoring them to self-generated power may inadvertently constrain their grid responsiveness. To evaluate these trade-offs, we propose a receding-horizon optimization (RHO) framework coordinating job scheduling, grid interactions, and on-site solar generation for a stand-alone DC. Our findings reveal a critical paradox: although solar integration increases energy self-sufficiency and reduces overall DC emissions, it inherently limits the facility's capacity to absorb low-cost, low-carbon electricity from the grid. This implies a fundamental tension between individual corporate sustainability goals and system-wide grid decarbonization.