The average cost of doing a full custom design with the latest technology is on average $14 million IIRC (read it on EETimes a while back).
At 0.13um the mask alone costs $2 million and you need some *very* expensive development tools and simulators because at that price you have to get it right first time.
Very few companies are now developing custom chips at this level and the ones left are dropping out like flies. Unless you know you are going to sell millions of them (think CPUs), or make a huge profit on each one (think high end graphics chips) there is simply no point.
A factory (FAB) to make these things in will run you about $2 Billion (Commodore used to have it's own FAB).
FPGAs are a much lower cost route and these days and if implemented I would guess every bit as fast - if not faster - than the real AAA.
If you want AAA get an FPGA, read up on VHDL or Verilog and get coding :-)