Imagine the latency on a data center in space. Uplink/downlink every time your server gets an inferencing request? Lol.

I could see it being fine for longer running asynchronous requests, but that would be if the cost/benefit made any sense at all, and if the servers had any resources worth talking about.