NASA's Science Mission Directorate produced a cross-discipline report on medium term needs for computational modeling capabilities; ie what serious iron NASA might want to get to play with
Panels from Earth System Modeling and Assimilation; Solid Earth and Natural Hazards; Astrophysics; Heliophysics; Planetary Science and Mission Engineering, considered current and near future needs and desire, scalability and both state of the art of the codes and development issues.
It is an interesting read, if you like that sort of thing.
Just for fun I did a keyword search:
Fortran - 18 matches
GPU - 5 matches
FPGA - 4 matches
Java - 3 matches
NVIDIA - 2 matches
C++ - 2 matches
Cluster - 38 matches
Cray - 1 match
beowulf - 0 matches
Cloud computing - 9 matches
I think I see the future.
Cluster - 38 matches
Some of these may be false positives. "Cluster" happens to be the name of a currently operating heliophysics mission (mostly European but with some US participation).