The gross margin doesn't matter, as long as profits are going up. Console APU sales have no marketing cost and the development cost is shared by the customer, so you can't treat the gross margin for that product the same as you would for the other stuff.
But that's SOP in the industry, no? Intel isn't spending today any significant amount of money in Haswell development. Haswell development was funded by Nehalen, Clarckdale, Sandy Bridge and Ivy Bridge. Intel is spending development money in Haswell-EX, Broadwell, Sky Lake, Knights Landing, Skymont and whatever comes after them. So you you must treat the gross margins of both products as the same.
What makes the embedded silicon different is that the development of those SoCs didn't stress AMD past cash flows as much as an independent product would, so when you factor this in a ROI analysis the smaller revenue stream will be mitigated by the smaller initial investment and you might end up with a descent ROI.
What is telling is the amount of cash this projects are going to generate. Factor a 24% gross margin, and I think this is an already exceptionaly higher estimate and even if they sell 1 billion per year they won't be making any exceptional amount of money. In fact, they will be generating less cash flows than they were generating with their big core line. To put things in perspective, despite 200 million extra revenues expected in Q3, AMD will be earning only 9 million more in gross profits.
In the end, everything points out for a different AMD, or at least one that won't be able to fund the kind of development they used to fund.