The problem is that an ASIC can be custom designed to go orders of magnitude faster than a GPU for a specific task. As just a very rough estimate, one SHA256 engine with bitcoin hash difficulty detection would take something like 1 million transistors. The latest ASIC technology allows 2 billion transistors. So that would allow 2,000 of these engines per chip. If you layed them out by hand, you could probably get them up to 4 GHz. Assuming one hash per clock, this ends up being 8 TH/S per chip. In other words, this one chip could outperform the entire current BitCoin network. And once the multi-million up front tooling costs are covered, you can make each chip for $50 each. So for another $10,000 you could make a system with 200 chips that would be able to dominate the entire network at 200 times its current capacity.
If these numbers are true, the whole Bitcoin system is doomed.
If one entity possesses more than 50% of the processing power, it can dominate everything because it can get wrong blocks (in their favor of course) verified simply because it has more processing power.
For example verifying some self-created "everyone sent all his bitcoins to me"-Blocks, double-spending, scamming, all you can imagine suddenly comes possible if you have access to such an amount of huge Hashing power.
On the other hand, such an endaveour would have an initial cost of millions of dollars - why would they want to wreck their own system then?
It's just something you mustn't forget.
Given that the new "über-ASIC" would dominate everything else, it could get all the 2016 Blocks for itself, netting 1.5 million $ at 15$/BTC.
But I just realized that the difficulty would explode all over the roof, so actually it isn't that bad. Not likely that the owner of this ASIC would destroy his own business before he has paid off his hardware.