I am definitely not an infra engineer, but I see it as an attempt to fix problems that have accumulated over decades as standards calcified and no single vendor was able to improve due to interoperability issues. Some of these things can’t be tackled without a full end-to-end solution, hence the full server. It doesn’t mean components couldn’t be swappable in the future, but at the beginning it’s best to develop on a narrow set of hardware until a solid baseline of stability is reached. Sort of an Apple-like approach for large scale servers.