Bay Trail is no competitive knockout and the situation with the 14nm successor will almost certainly be much worse.
Much worse ? A57 against 14nm Airmont and then 14nm Goldmont ? Unlikely. A15 already has a performance/power design/process mismatch that limits sustainable high clock-speeds well under its theoretical maximum and its fatter A57 derivation will suffer probably worse.
The big problem though is that far better ARMs will be around starting to compete in the server market now the software offerings are getting fairly complete. What's Intel going to do, offer a cut down Haswell for $20?
You are dreaming imaginary chips up. Even the AMD licensees are selling their server SoCs starting from 3 figure numbers and an A57 or X-Gene is barely in Avoton's class never mind Haswell Xeon's.
And about not paying a foundry margin - of course Intel pays for its foundries and has to do R&D and get profit from them. Costs and the need for profit don't suddenly disappear just because two functions are done within the same company.
The proof is in the pudding, none of these strictly ARM SoC only companies make enough profit to pay for a continuous FAB and silicon process treadmill like Intel has had for decades and the mobile Atoms can take advantage of this PC/Server bought process infrastructure as a free foundry.
As to foundries being clueless after 16nm - that is just wishful thinking given the published plans and talks about 10nm.
Talks, plans ? Morris Chang has not sounded confident in years and that guy has trouble with masking the truth when he should. If everything was hunky dory in foundry land they would not be thinking of introducing these hybrid FinFET transistors on larger back-ends which they are unable to scale further at the moment which is why it is going to remain 20nm over two generations. They are scrabbling to keep up already and it's only going to get worse with quadruple patterning which is their only concrete idea for getting under 20nm now. They are all hoping for Intel technical infrastructure crumbs to cascade down to them to help them advance further.
As for MIPS it is conspicuous by its huge absence in mobile application processors and there is no-one pushing hard for that to change. MIPS and its licensees seem happy enough to stay in their current niches. So whatever challenge they could provide is purely theoretical.
Market Data 
Markets 
AI
