But, remember, all miners collectively are limited to about 7200 BTC per week. Building vast amounts of mining capacity can increase your share of that, but not total production.
Today's share of CPUs in mining is 0%. A product like Xilinx EasyPath ($75000 NRE + lots of orders) would bring current FPGA upfront costs down quite a lot. What's needed for someone to gamble on such a venture is the belief that ASIC (or if you want to call EasyPath a hybrid FPGA/ASIC solution) miners would grab a significant share of those
7200 BTC per week. (edit:
$2/BTC is still $750k/year)
If the BTC price is going down to levels where it's just not profitable for GPU miners to mine due to electricity costs, that will indeed happen.
edit: I went with 7200 BTC per week from Nagle without thinking. It's 7200 per day though which changes the calculations a bit. Even with $2/BTC that would still mean the value of bitcoin mining for a year to be $5.26M.