As GF found out the hard way, transitioning from a few product fab to a foundry is much easier said than done. Intel is moving at a glacial pace in that direction, and herein lies the conundrum. When they move slow they learn on the go and don't get into a GF style mess. But by moving that slow they don't force themselves to confront the problems head on, and may implement a patchwork of solutions to their current systems that do not scale well. Only time will tell.
Intel will always cater to the highest performance markets. ARM is divorced from semiconductor manufacturing process. This divorce is ARM's business model. Assuming that ARM can somehow eliminate Intel's process advantage and "take over" (which is not their goal either) the processor world flies against common sense. ARM can try to stay close to the process, but there is always the "fab secret sauce" that ARM will never have access.
The problem with Intel is that they still don't seem to have SoC chips that allow people to do embedded-class complete boards, in terms of size, power and cost. You can get Raspberry Pi, BeagleBoard/Bone, etc. ARM boards that run Linux and can be powered from a USB cable. Intel needs to invest in ecosystem enough so that such boards will be available for Atom.
it is indeed interesting if Intel is not investing in the Atom ecosystem enough...does this mean they have other alternatives to Atom? Maybe their high-end server processors frozen in architecture over next 5 years automatically becomes an Atom in a semiconductor process 5 years from now (such as a 7nm process)