Systems of semantic categories vary across languages, but this variation appears to be constrained by pressure for optimizing a complexity-accuracy tradeoff known as the Information Bottleneck (IB) principle. This finding, however, has been based primarily on individual languages and it remains largely unknown how bilinguals navigate the category systems of two different languages, particularly when these languages’ category boundaries do not overlap. Here, we address this gap in the literature by combining theory-driven experiments with an extension of the IB framework to bilinguals. Specifically, we investigate bilingual vs. monolingual category boundaries in English and Mandarin via a two-alternative forced-choice (2AFC) labeling task on six continua that interpolate between two distinct everyday objects (e.g., plate and bowl). We find that: (1) bilinguals do not maintain two monolingual-like systems but rather exhibit a converged semantic system influenced equally by both languages; and (2) this departure from monolinguals is nonetheless constrained by the same pressure for efficiency that operates in monolinguals. These findings provide new insight into how bilinguals navigate cross-linguistic semantic variation and suggest that despite having to accommodate myriad sociolinguistic factors, a drive for efficiency is also a key factor that shapes bilingual category systems.