They could give up on design and just try to be an American TSMC.

There's a lot of interest in that from a national security perspective anyway.

With Intel's cashflow, they can do both without any difficulty. But yes, they should be the american TSMC.

I'm not sure they really can do both (because doing the TSMC model well intentionally excludes the other).

Part of TSMC's offer is that they won't do design, they won't compete with you on design etc.

They specialize in the best possible fab and service for your design and that's it.

> Part of TSMC's offer is that they won't do design

That's kind of a hoax. GUC (Global Uni Chip) is huge, and a nearly-wholly-owned subsidiary of TSMC.

https://en.wikipedia.org/wiki/Global_Unichip_Corporation

TSMC has a large design operation. It simply is part of a subsidiary rather than the parent company, that's all.

There are a lot of shady links between TSMC and GUC. I was once pressured to "collaborate" with GUC as an explicit condition of getting access to the then-bleeding-edge 16nm PDK. I turned it down. A competitor of mine (with whose chief engineer I am friends) had the same screws put to them.

Interesting - thanks, I didn’t know that.

Exactly right.

Someone else mentioned it, but that's really not true about TSMC.

Doesn’t Samsung make a whole bunch of components and use the components to create products laptops tv phones storage

Actually I'd be curious to see how a Korean Chaebol works in this regard. At various points, Apple was competing w/ Samsung (mobile phones) to get fab space on Samsung (chip fab), and I think at various times different parts of Samsung were suing Samsung...

Fly Wheel, a term people used to describe Amazon.

Imagine 100x of that, and that is Samsung.

I would LOVE them to be an American TSMC. Would save so much pain.

That makes no sense. Intel's designs are fine, industry leading actually. Intel's biggest problem by far is the repeated inability to launch new process nodes. It makes no sense for Intel to give up on what it's good at to focus on the part where it's struggling.

I would love a RISC-V laptop, but only if the vector and matrix extensions can be used for training and inferring neural networks. Having separated vectorized processing for CPU (SSE) and a neural engine doesn't make sense to me.

I want to be able to train NNs with a laptop that can normally last 20 hours.

Imagine what Apple does to x64 to work on ARM, Intel having Altera, could do x86 translation to FPGA gates... It wouldn't take whole programs, but it could reconfigure itself to run most commonly executed routines on FPGA.