Intel Corp (INTC)

[This Causes an Error]

Interestingly, for once I agree with ibc (on perhaps his only non-pessimistic post on Intel), that big-core (core M) may be on the way for some high-end smartphones.

Core M in its current form is wholly unsuited for smartphones. It's a two chip solution (one on leading edge node, one on n-1/n-2 node), so right off the bat it is at a power consumption/cost disadvantage.

Additionally, the merchant market has shifted 100% to integrated modem + AP, with basically only Apple using a discrete modem, premium tier included. Intel typically builds core M on its CPU-targeted processes which do not support the high-voltage I/Os that its SoC-targeted processes do.

These processes also use leakier, higher performance transistors which means that the CPUs can clock higher, but you pay for it in standby power consumption. If you think about how people use phones (i.e. often in standby mode), this is a non-starter from a total power consumption perspective.

Beyond that, a high-end mobile SoC needs much more than a fast CPU and GPU. If you pay attention to the vectors along which companies like MediaTek and Qualcomm innovate along, you will see particular focus in things such as the Image Signal Processor (phones are used often as cameras). These are IP blocks that Intel has not demonstrated leadership in.

Also, as far as the CPU goes, the Core CPUs are very bloated for mobile workloads. Things like AVX-256 units (which are helpful for HPC and high-end desktop applications but worthless for mobile applications) take up unnecessary die area that could either be removed (to save cost) or allocated to improve legacy x86 performance.

So, put bluntly, the "converged core" approach just isn't optimal against the approach of building targeted IP for a given application. Core m in phones, so long as Core m even remotely resembles today's Core m, is a no-go and will probably tank worse in the market than the Atom lines if Intel were to try to seriously push it.

To summarize: if Intel wants to compete in the smartphone market, it needs to build (or license) intellectual properties that are targeted at the kinds of products that it aims to power. Atom should have been a more power efficient version of Core specialized for mobile, not the neglected, red-headed step-child that it was ultimately treated as.

FWIW, if Intel is serious about mobile, and if it can get its modems in shape, then there is no reason why Intel can't license off-the-shelf IP, integrate with a modem, build it at TSMC, and try to make some money.