Although my roots before joining Microsoft were in supercomputing, I believe that "extreme computing" and adding gigaflops (billions of floating-point operations per second) are no longer the optimal solutions to most scientific and technical problems. Today, scientists and engineers can buy or build 10-gigaflop desktop computers for around $5,000, and within the next several years, we will see similar supercomputing power at the chip level.
Instead, the next breakthroughs in science and engineering will come from harnessing the power of software and data -- for example, using low-cost sensors to collect terabytes of real-world data and using data management tools to understand it.
Of course, combining computer models and real-world data presents new challenges, particularly in learning how to store, search, analyze, visualize, publish, and record the provenance of that data and the resulting conclusions. I believe the software industry can play a key role in developing tools that automate these data management tasks.
