Because it's less likely to be 20nm at TSMC when clients have abandoned that process node and either went with 16nmFF or 28nm.
Of course TSMC refines their process nodes to improve leakage and power consumption.
http://www.tsmc.com/english/dedicatedFoundry/technology/28nm.htm
They've refined their 28nm and 16nm nodes but have hardly touched the 20nm nodes according to their foundry page.
So as Thraktor stated in a post a couple of pages ago, they have a 28nm HPC+ node which is very close in comparison to the 20nm node.
On the other hand there is 16nmFF+ and 16nmFFC which is obviously better in power consumption and density. The question is, which is cheaper and gives better yields?
Thraktor summarises the above in this post anyway so take a look:
http://m.neogaf.com/showpost.php?p=227547584