A huge cost is R&D. You worry how much the wafers and heatsinks cost, but the real point to consider is R&D. The Fermi architecture, just like the G80 architecture, was a long time in development and expensive to produce.
But consider a few points:
- The architecture is being made into three seperate high-end products, the cheapest of which is the GeForce part. Quadro parts are often priced twice (or more) the price of GeForce parts, and Tesla parts even more.
- The architecture is extremely modular. Scaling up for the next generation after GTX 480 is almost as simple as CTRL+V. Scaling down is just as easy.
- Longevity. Fermi has it. It's an extremely advanced architecture with a ton of features built around the best part of DX11, tessellation.
Bookmarks