Or you could always focus on getting someone really good at image rendering and announce a PCI-E card that no one has anything close to and charge customers to upgrade at a chance of getting something meaningful.
Cointerra beats Monarch in every respect. Delivery date, price, and efficiency. Not saying one should buy it trying to project ROI% for Dec with some much hashpower being sold (consensus guestimate - >6 TH by end of year). However there is no reason anyone should pick Monarch over Cointerra.
Or become a chip company and focus on providing chips for others to use in their product.
In the long run I think the successful ones will. Look at Bitfury. They sell no boards to the public instead they wholesale chips and rely on three major OEM (US, EU, Russia). Think of how well that allows them to focus on the core task of chips, chips, chips. Given their 55nm design has higher efficiency (J/GH and GH/mm2) than some 28nm designs it seems to be working. If I was a ASIC company I would be worried about what a 28nm die shrink will look like.
Even AMD doesn't make graphics cards. They make GPUs for a dozen or so OEMs which make graphics cards.