Artificial intelligence (AI) systems are increasingly perceived not merely as tools but as highly capable nonhuman agents, eliciting complex emotional responses from users. Little is known about how people emotionally respond when AI capability appears to exceed familiar human boundaries. Drawing on awe theory, this research investigates how perceived AI capability elicits awe and distinguishes between its positive and negative components. Across three experimental studies, we examine (a) whether AI capability evokes awe in a nonlinear manner, (b) how AI capability differentially shapes positive awe and negative awe, (c) the moderating role of AI human-likeness, and (d) whether AI and human agents are emotionally equivalent sources of awe. Results consistently show that higher AI capability elicits stronger awe, with effects becoming more pronounced at higher capability levels. Importantly, AI capability simultaneously amplifies positive awe and negative awe. AI human-likeness attenuates these effects, particularly by compensating for low AI capability in eliciting positive awe. Moreover, when capability is held constant, AI agents evoke stronger awe than human agents. These findings extend awe theory to human-AI interaction, reveal the ambivalent emotional consequences of highly capable AI. The study offers practical implications for the design and governance of advanced AI systems.