Parallel computers with more than a million processing cores will be available in the next three years. Although MPI is the dominant programming interface today for large-scale systems that already exceed 100,000 cores, many people wonder whether MPI will scale to systems with millions of cores. This talk will examine the issue of scaling MPI to such large core counts, including what needs to be fixed in the MPI standard and what needs to be fixed by MPI implementations. We will present some results of improving MPI's memory consumption at scale on Argonne's Blue Gene/P.