Data center servers have evolved quite since last xserve, to be welcome at datacenter an unit needs out of band management (aka IMPI) redundant PSU, ECC ram, no Mac got IMPI yet while not complicated it may require specific provisions at logic boards, not imposible but not what apple has deployed right Now (CentOS servers with Nvidia GPUs powers Siri), and unlikely what consumers (even pro and enterprise) would buy for in-house, and it's the way AI is deployed, AI training and inference is mostly (if not all) platform agnostics, and given the scale and costs, saving money is not something to argue against traditions or fanboy pride.I think an xserve like design would accommodate these features better
While I see apple deploying it's own (or purchased) AI cloud on arm servers (as Altra) loaded with custom TPU either developed by Apple or an contractor.