When you consider the world of weather modeling combined with modeling from Geomatics applications, petaflops are what can take it to the next level. This recent BBC story that Intel is restricted from exporting a new generation of Xeon processors to China raises the international issues for large scale modeling hardware and software.
The US Commerce Department has rejected sending newer generation (number is not known of the 80,000 it uses from the story) Xeon processors to upgrade China’s Tianhe-2 supercomputer server farm. It would go from 33 to 110 petaflops with the upgrade.
While tossing petaflops at modeling will work to a degree, atmospheric randomness and model assumptions will restrict the usefulness until future research discovers other processes to improve them. It also restricts modeling at a huge petaflop level outside the United States due to the risk of other countries misusing the technology (in this story it brings up worries of nuclear weapon development).
In the story it does mention the new Aurora supercomputer at the Argonne National Laboratory in Illinois will have 180 petaflops of capacity, so the technology will continue to grow in the United States.
In summary, international relations will determine growth in modeling in the future as non-nuclear scientific and engineering research goes against risks of using the technology for nuclear weapon development.