A machine should not be asking, “Can I run the whole intelligence locally?” but rather, “Which portion of this operation needs which substrate?”
The cloud could handle giant models or computationally expensive operations.
If your local machine contains your memory, context, indexes, preferences and working state, then the cloud becomes an instrument being queried by your local system.
Reverse that arrangement, and you get today's more common structure: you are effectively a remote terminal into someone else's indexed system.
A genuinely distributed intelligence would not necessarily be one monolithic entity occupying one machine. It could be a process whose components are distributed across substrates. What could live locally is the continuity of the process.
Does the person own the centre of the process, or does the infrastructure provider?
Analysis—Isshin