These systems have important complexity layers that may not be immediately apparent: latency and edge hardware limits.

Yes, maybe, gpu cluster server can understand you quickly, but taking whatever model you have and getting it to work quickly enough for people not to be upset is a giant problem.