For the currently used architectures, it doesn't make sense to have 5 specialized, dedicated computers as the "AI" text processing and "AI" video processing and the others use very similar architectures and there's no benefit from specialization, the "video-specialized" hardware would be just as good at processing text and vice versa.