You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
Right now Arm is making custom CPU chips for customers to use internally, it is a collaboration versus making a commodity chip to be resold on the open market.
Cloud companies are looking for every advantage possible. Amazon and Google are making their own chips so other competitors have to step up to the silicon plate and working with Arm makes sense.
But the reason why hyperscalers are making their own chips is...
I think it's probably because of the cost and the ability to customize it to suit your needs.
Under the same conditions, there is a possibility that you may choose an option other than ARM.
Personally, I can't imagine a big company like a hyperscaler buying a finished chip...
Very interesting. So Arm is competing with Broadcom. But a significant factor in choosing Broadcom is their very extensive IP portfolio. I wonder how Arm compares?
I wonder among the top tier AI factory build out, what is the leading edge SERDES market share? Marvel/Broadcom will use their own SERDES for sure in their custom silicon. nVidia the same. what about Alphawave and Synopsys?
Good info. Broadcom's attractions though include PCIe blocks, Ethernet PHY and NIC blocks, encryption blocks, among other things that may not be critical to big cloud companies (like wireless IP). I'm still thinking I'd bet on Broadcom. Admittedly, Arm cores & memory controllers on chiplets could be interesting, as Graviton 4, for example, uses chiplets. I wonder if Ampere's custom cores are part of the chiplet furture?
