My original post was meant to say that aren't there other ASICs already designed by someone else that can do these same types of computations.
There are other "ASIC" Intellectual Property out there already that can do the same computation. However, those IPs are optimized for low area and low power, so that they can sell it to someone who wants to add hashing capability to their ASIC.
(How much sha256 crunching will a mobile phone need to do, really?)
This is the key point. IP out there are meant to do a few hashes. So the tech sheets I've seen from IP providers calculate a hash in 66 clock cycles. Furthermore, because they are minimizing area used, they reuse as much as possible during these 66 clock cycles, meaning you can only calculate one hash every 66 clock cycles. Assuming you are running it at 660MHz, then you only get 5MH/s per IP core (bitcoin hash is SHA256(SHA256(x))).
But to the mobile phone ASIC designers, they are happy to license it so that they can calculate a hash in 100ns, compared to however long it would take to put it through the MIPS or ARM core, or having to implement one themselves.
In comparison, the initial open source FPGA design unrolled the SHA256 cores (if the FPGA can fit it), which means it computes SHA256 in one clock cycle and a bitcoin hash in 2 clock cycles. It uses up a lot of area, but can roughly get MHz -> MH/s (because of pipelining, the first core can start calculating the next hash while the second one finishes)