Ultimately Apple is going to have to make XServe's again with M-class chips, probably when Mac Pro is released.
I doubt that! I don't think that the server market is where Apple wants to play - Steve Jobs killed the XServe because it didn't make any money. We also don't know whether Apple's virtualization hypervisor is suitable for running multi-tenant workloads, and this is a core feature for Cloud services (almost no-one is running on physical machines now).
The M1 is also still a very small CPU by modern standards and the economies of scale lead to the use of much larger chips, whether Intel, AMD or the various ARM designs (like Amazon, Ampere, Marvell). The ARM Neoverse N1 design used by the above vendors goes up to 128 cores per socket. These make sense for Cloud servers where users are provisioning a subset of the available vCPUs for each virtual instance. These servers are often "oversubscribed" to support a larger number of VMs - i.e. the server will support more VMs than it has physical cores for because not all VMs are running 100% of the time. An M1 CPU could probably only run up to a dozen or so small VMs, and maybe only 1 or 2 larger ones. This is not cost-effective if you consider that each server needs its own memory, storage, power supply, interfaces etc. There could be multi-socket machines, but there are no indications that Apple Silicon has been designed to support multi-socket configurations.
Commodity Server hardware (and running cloud data centers) probably have low margins, which are recouped by the services offered on them and AFAIK, Apple does not want to become a generic Cloud Service provider like AWS, Microsoft, Google, Rackspace etc.
It's conceivable that Apple might use some variant of Apple Silicon in its own data centers for some specific Apple services...but it's probably not worth the investment in developing specialize high-core count server chips, unless this is an offshoot from their Mac Pro designs.
