I helped build a POC of this precise idea at a university datacenter ~6-7 years ago. I broadly agree that it's not super scalable if you try to use it as "all

I helped build a POC of this precise idea at a university datacenter ~6-7 years ago. I broadly agree that it's not super scalable if you try to use it as "all up monitoring" but there were certain instances in which I found it to be very useful. - We used it among other things to visualize datacenter temperature as volumentric clouds. It let us better diagnose airflow and cooling issues. - We previously relied on flashing lights on disk controllers to identify the bad slot of a disk that needs to be swapped. But when you have heterogenous SKUs/chassis/disks/backplanes, maintaining that infra is a PITA. AR offered avenues to do this that slightly reduced the complexity needed to provide this feedback. Basically, even if you "piped your data into a machine" a human still had to eventually interface with systems (the datacenter, the machines within them, networking topologies) that are highly opaque without augmented guidance. To your final point as well: I see the lighting solutions vs. portable AR as different points on the "AR spectrum" and they offer obvious tradeoffs in portability, extensibility, etc.