I think Nvidia should try to create an compute-analogue of wifi routers where computation will be offloaded to smart-home gpu-servers. This strategy will cement their future for eternity.
Because you are pooling computer resources across many more users you have better amortized cost than if you had a workstation on every home that was idle 99% of the time. Of course, latency and bandwidth to the cell tower is worse than wifi, but better than if the compute is done on a remote server.
I have no idea of which model will succeed for which use cases, but the idea is sound.
Again: which computation paradigm works best depends on the use case. It is not an all or nothing situation.
As much as there are market drivers to make the cloud attractive to businesses and legitimate reasons why cloud solutions are better than alternatives there is a real desire in a growing number of people to have a solution that they can tinker with but that also has a polished UI so they don't have to tinker with it when they don't want to.