CXL memory vs. CXL potential

IMO if vendors want to see this in prod fast enough, it's best to go for a tiny hypervisor and the good old Adaptive Replacement Cache, if it looks like these discusions will stay theoretical while use-cases are waiting to be accelerated... It would seem to be all about that - are there actual use-cases waiting to be accelerated? Or is this motivated "just" by something like managerial regret of having to buy the whole prod amount of RAM every few years (as new generations come along) from 0 and add the required upgrade amount on top of that?

Reducing the whole CXL potential/debate to "memory recycling" technology, is also heavily underwhelming. It's also for CPU-to-CPU cache coherent comms, so I was expecting to get fiber bridges between systems which fit in PCIe slots, so we can finally build out rackscale NUMA machines, as mere mortals, without own R&D...

Please don't kill the technology with such a weak momentum by coming at it from a wrong angle, it can be far more than glorified swap - there's also nothing novel for us by thinking in that direction, I think.