Intel can make a better performing CPU with x86. They have a 30 year history of doing that well. My guess is that It would take a very long time for the internal tools and validation suites to be mature enough to switch ISAs for a custom microarchitecture - not one of the off-the-shelf ones.
A bigger problem is that Intel is most likely still on a learning curve for all of the other pieces for a mobile SOC.
There's a tremendous amount of engineering required for the IP blocks like camera ISP, modem, audio and so on. Furthermore, there's a ton of systems engineering that is required when all of these things share a path to memory. For example, you need highly tuned QOS algorithms so you don't drop frames and provide a poor user experience in bandwidth starved environments such as a phone. It takes a long time to develop these learnings and methodologies (BKMs in Intel-speak).
Intel is even missing some key blocks like the modem and GPU. When dealing with 3rd party IP, there may be integration issues with Intel's MPU centric design flows. So there's also a learning curve there. Yes, I know they bought Infineon, but it's still a foreign IP with respect to the design flow. Same with Imagination GPUs.
That is just the hardware side. Perhaps even more daunting is the software side where Intel needs to provide drivers and low level code to make all of these integrated devices work well together in a system. Intel's software team isn't exactly the best in the industry...
On the customer side, there is also a certain amount of co-engineering where each handset maker will require a great deal more cooperation and transparency vs. a traditional PC ODM relationship. Is Samsung going to risk their bread and butter handset on an unproven partner? They even build their own Exynos SoC to back up Qualcomm.
On the process side, Intel may be more optimized for performance vs. cost. If you look at Atom die photos, they look very large vs. some ARM implementations.
Over time, they will get better. It's not clear how far along the learning curve they are.
Market Data 
Markets 
AI
