The safest bet in the world is that computing power will continue to grow at an exponential rate while costs stay constant or fall. There is simply no technological barrier on the horizon to stop this process, and market forces are inexorably driving the evolution of technology in this direction. The best way for a software vendor to profit from this is to choose a problem domain that has demonstrated its worth on large computer systems, then repackage that tool for a much broader market, anticipating the arrival of low-cost, mass-market computers that will allow it to be used practically.
This is precisely what we did with AutoCAD, and we knew exactly what we were doing at the time. By choosing a product on the edge of currently available compute power, you avoid jumping into a heavily contested market. Instead, you can start the process of market development on the margin, among education and other users able to use the product on existing machines, putting the infrastructure in place so that when machines really suited to the product begin to arrive you're in an unassailable position of strength.
Actually, with AutoCAD, we thought the PC/XT was the machine that would make CAD real on the desktop but we were wrong. Looking back on what really happened, the whole XT era was precisely the time of market-building I described above. When the AT came out, serious practical work with desktop CAD was truly possible, and we simply rode the wave we were already perfectly positioned on.
In looking forward and choosing new products, we should be seeking products with the same properties; products which run on large computers which are useful enough that a small segment of the market able to afford such machines pays the price of admission, but which will be applicable to a much broader set of users when the price falls to a level they can pay. This is a property of many of Autodesk's current development directions: photorealistic rendering, solid modeling, finite element analysis, and 3D user interaction. Molecular modeling has the capacity, like CAD, to soak up all the additional computing power anticipated for the next decade, and in the process, expand the market for computational chemistry just as AutoCAD has done for CAD.
Editor: John Walker