The challenge of exascale supercomputing has led to major changes in architectures of the top supercomputers. The Exascale supercomputers has become more parallel and heterogeneous than ever which brings new challenges to communication libraries such as MPI. The MPI library today not only need to find a way to provide high-performance communication at large scale, but also need to be able to interact with different programming models efficiently.

In this talk, I will present the recent improvement and progress on the MPICH project for addressing those challenges. I will also discuss what are the future directions for MPI to keep evolving to support emerging applications such as AI for Science.