One comparative advantage that comes to mind is the institutional knowledge of computational and analytics systems. A lot of computational science is still done with a FORTRAN mindset (if not actual code), whereas faster, lighter, dare I say more agile approaches to simulation and data mining could be much better.
Similarly, although I wouldn't guess is what YCR is thinking because it doesn't have the same cachet (though I really don't know what YCR thinks) is building tools for science. In my field (geoscience) there are boundless slopes of fresh, trackless powder in the mountains of computational geoscience just waiting to get skied. Building tools to do, say, probabilistic graphical modeling of stratigraphic systems requires approaching a traditionally intuitive and descriptive science with a much more formal and precise mindset and re-casting a lot of ideas and methods, while at the same time keeping everything accessible for those without the requisite background in probability or graphs... To me this gets closer to the real roots of science than a lot of what gets funded or makes headlines, but won't by itself get funded or make headlines (until you build it and sell it).