We consider the problem of adaptive sampling for global emulation (metamodeling) with a finite budget. Conventionally this problem is tackled through a greedy sampling strategy, which is optimal for taking either a single sample or a handful of samples at a single sampling stage but neglects the influence of future samples. This raises the question: “Can we optimize the number of sampling stages as well as the number of samples at each stage?” The proposed thrifty adaptive batch sampling (TABS) approach addresses this challenge by adopting a normative decision-making perspective to determine the total number of required samples and maximize a multistage reward function with respect to the total number of stages and the batch size at each stage. To amend TABS’ numerical complexity we propose two heuristic-based strategies that significantly reduce computational time with minimal reduction of reward optimality. Through numerical examples, TABS is shown to outperform or at least be comparable to conventional greedy sampling techniques. In this fashion, TABS provides modelers a flexible adaptive sampling tool for global emulation, effectively reducing computational cost while maintaining prediction accuracy.