When do we expect to see Microsoft’s server chip? Intel is really screwed if even their former partner in crime is abandoning them. Microsoft is #2 in cloud, and Amazon is #1 and all in on custom silicon already.
Depends upon what Microsoft is trying to do. Short term they are going to buy. ( have already started doing that with Ampere Altras). [ see post above this one ]
Not just Intel . AMD is on slippery slope too. That's is in part way AMD is doing a 'Bergamo' , high core , lower power/core variant of the Eypc Zen 4 server chips. Selling cloud services is primiarily about lowering cost of computational services for customers. If Intel can't deliver lower operating costs then they will loose out in cloud services placements. Customers will drive their virtual machines and/or containers to the shop that provides the lowest cost at scale. If costs are higher... just get dropped. Nothing personal.
For the hyperscalers with huge dynamic growth Intel is in deep doo-doo . Very deep doo-doo. Intel is largely depending upon server deployments to be relatively slow moving and heavy on inertia ( "ain't broke don't fix it" ) to hold folks in place until they can get their act together. stable small-to-mid/large shops that will probably have much longer transition times which gives Intel a shot to 'right the ship'.
The core issue on timing is when did Microsoft start. If they are just getting serious about doing it by themselves then 3-4 years. If the project has been in flight for 1-2 years and it is a mild dervirative of Neoverse N2 it could be about 10-18 months. ( This Apple person could/would be working on one of the follow on generations. )
Amazon also has a custom DPU/Infrastructure card that is ARM based ( Data Processing Unit / Infrastucture Interface card). [ basically offload the network storage and "back end" work associated with a virtual machine is unloaded off the application CPU. ] Microsoft could be doing a SoC to do that also.
Microsoft probably does more Windows hosting than Amazon (and the other major competitors). A Server SoC skewed to multi-tenant Windows instances could be what they are working on. That probably would have started a couple of years ago because there has been a "server version" of ARM for a while ( that hasn't gotten much traction before. But would have been something could tune a specialized server implementation to. )
Microsolf has been vocally cheerleading some other folks to do a Arm Server implementaion for a couple of years. I don't think they were gaslighting the competition for all of that time. There is probably a mix of Arm server implementations they want for Azure.
I wonder if Apple will make a server chip, even if it’s only for their internal data centers.
More than probably not. Most of what Apple needs to run for their own data center services is Linux. They could either buy "off the shelf" Ampere Altras for that or at worse pull a reference Neoverse implementation with very minimal "touch ups" of their own. Ampere says they are looking to do an even more "cloud" optimized core in 2023-2024. Apple really isn't in the lead here for cloud workloads. ( Neoverse N2 scaled at 60-80 cores under high load from dozens of clients. Apple hasn't demonstrated can do anything near that. )
For XCode cloud Apple can just just stuff M-series Minis and "Mac Pro" (although probably not a classic Mac Pro form factor) into racks. Same chip as shipping to non data center clients. macOS licensing for cloud services is oriented toward more Macs being sold (and racked). Apple quite likely will just do the same thing as they are asking the other MacOS service providers to do. ( Really doesn't build good partnerships to engage is huge hypocracy. Even less so with several antitrusst regulators are itching the throw the "book" at you. )
Relatively though the macOS specific percentage of their cloud service is likely to remove small.
