Advances in computer and communication technologies have stimulated the integration of digital video and audio with computing, leading to the development of various computer‐assisted collaborations. In this article, we propose a multilevel conferencing paradigm calledsuper conferencefor supporting collaborative interactions between geographically separated groups of users, with each group belonging to possibly a different organization. In a super conference, each participant must receive and display the composite media stream obtained by mixing media streams transmitted by all the other participants. Hierarchical communication architectures are naturally suited for media mixing in super conferences. We present algorithms for designing hierarchical mixing architectures that optimize real‐time end‐to‐end delays of media. In order to improve their real‐time performance further, we proposemultistage mixingtechniques by which mixers can carry out mixing concurrently with communication. Surprisingly, the optimal architectures for multistage mixing are widely different from those ofmonostage mixing(in which, mixing and media communication sequential as opposed to concurrent). Based on real‐time delay constraints of multimedia, we obtain interesting limits on the sizes of both super conferences and groups within super conferences in optimal hierarchical architectures, which go to show their high scalability in terms of both the maximum number of participants and the geographical separation between them.