Assuming 22nm is largely depreciated at this point, I'd peg wafer cost at about $3000-$3500. Assuming 90% yield and assuming a 102mm^2 die size, I estimate we can get 540 good die per wafer, putting chip cost at about $5.50 - $6.50. Add a buck for packaging and test, and we're at $6.50 - $7.50.
If they're selling these for even $20 a pop, then Intel is making =60% GMs on these puppies. Even if we say wafer cost is $4000, we're at 58% GM.
Intel is going to make a lot of money in mobile going forward, I think. We just need the n-1 products to be competitive and the n products to be clear leadership. I think we get this at 14nm/22nm.
Those are the sort of numbers and analysis I get too. However like backbay said it is the overall margin/profit dollars and product mix that ultimately count. However I don't believe Bay Trail will cannibalize Haswell as there is still a clear performance difference between the two and Haswell now gives much better mobile battery life than previous Core processors. Then the only thing is how much financial inducements is Intel currently giving to advance Haswell and Bay Trail quickly over the incumbent inventory/opposition as that could also mask the great margins these chips actually have. Many variables to consider. However it is clear that basically Intel has no trouble achieving raw profit at standard mobile SoC prices, just needs the volume going forward now.
p.s. I suspect test/packaging may be closer to $0.5 as Intel has its own packaging facility. Bay Trail die cost could be as low as $6 especially when reused defect dies are included (e.g dual-cores). $8 tops.