Beastian (noemail.delete@this.aol.com) on March 23, 2021 5:39 pm wrote:

> 1) They can kiss those jealously guarded, fat x86 IP margins goodbye as most customers will be
> fabbing non-Intel IP at the foundries. It will be something like cellular modems, GPUS, or ARM
> based server SoCs for hyper-scalers. Not only does Intel lose the easy IP margins, they empower
> their competition to compete with their existing server and laptop market share. There's little
> chance they'll ever license x86 as they'll try to ride this dwindling moat of software compatibility
> for as long as they can, but ARM designs and emulation will eventually bridge this moat.

Meanwhile, Google has just hired Uri Frank from Intel to develop SoCs for Google's future cloud infrastructure.

Designing and making these will be serious money even by Google and Intel standards; enough that they will be talking, and enough to overcome any resistance to licensing Intel's crown jewels. And one of the options on the table will be using Xeon x86 tiles from Intel fabs.

And I don't believe there will be much internal resistance from Intel: it's becoming highly obvious that if Intel isn't prepared to license their core IP, they can wave goodbye to Google's billions. Intel has often been prepared to do unusual things to get a really major contract -- witness the "custom" chips for Apple, or licensing the Pentium III bus to Microsoft for use in the original Xbox (when they wouldn't license to Nvidia without a cross-license and charged VIA a lot, but Microsoft got the CPU and bus cheaply enough that AMD said they couldn't make a profit at the prices Intel was charging).

And quite apart from ARM competition, I'd be surprised if AMD haven't regularly offered Google Epycs with standard 7nm compute chiplets and a custom I/O die. That's the standard Intel will have to match to keep Google's business.