I doubt that any company would want to just "enter the CPU/SoC market". ARM is the way a company can build a CPU/SoC part to solve a specific application problem that is not solvable by an off the shelf part.
Transmeta opened that box 10 year ago, didn't they? The ARM method is just a repeat of the Tranmeta business model isn't it?
If I wanted to enter a market and grow my own CPU/SoC with my own innovation, what would I have to do?
ARM license. ARM models. Design tools. Employees and employee costs to design, implement and validate design. Company infrastructure to support those designers. Silicon fab to build parts.
Any of those development related costs incurred would have to be amortized over the units purchased, on top of the silicon fab costs. They are not free.
The alternative is to simply buy a part from someone who is selling something that will solve my problem. Work with a CPU provider to get my part built by them. Buy off the shelf from Intel, Nvidia, TI, ...
more relevant end product differentiation can be accomplished via [OS] software, than at the processor level. If the processing core basically meets your design objectives (power and performance), it is foolish to reinvent your 'own' wheel. That is not to say that the Drug companies don't do it all the time with various 'me too' offerings... they are just not game changers. A better example may be video transcribing. It is certainly MORE effective with hardware support, but even that is basically reduced to some rudimentary operation. Taking on the complexity of supporting your own CPU offering BETTER BE WORTH IT, 'cause it is gonna keep costing ya!