Intel's not lagging behind much; it just looks that way because Samsung and TSMC define their feature sizes differently.

Their 10nm process is comparable to Intel's 14nm process, etc.

Both will run out of steam at about the same point -- Intel's 5nm, and Samsung/TSMC's 3nm, will likely be the last process shrink.

There are libraries, and then there are libraries. Some libraries, like BLAS, are written by very good programmers who are expert in their domain, avoid common pitfalls (or evolved to avoid them after several development cycles) and have optimized the hell out of the implementation. Other libraries, like almost every javascript framework in existence, are poorly written, horribly slow, and full of bugs (but may provide hard-to-get capabilities). Code re-use is kind of like email implementation -- it seems like it should be straightforward, but it totally isn't. Doing it right requires care and experience.

This isn't precisely a pc on a chip (the core is MIPS-based), but Microchip's PIC32 offerings gives you a fully 32-bit processor with integrated RAM, ROM, and some peripherals for about $5 per unit. Perhaps not useful if you need x86, but plenty useful if you just want to compile and run ANSI-C applications (GCC has an appropriate MIPS target).