Evidence for the semantic hub hypothesis shows language models develop a shared representation space, improving their multilingual and cross-modal capabilities.