Comment by AntiUSAbah

Nvidia has so much money, it would be a waste if they wouldn't attack current problems on multiply points at once.

People, Researcher, Investor etc. probably also want to see what would be possible and someone has to do it.

I can also imagine, that an inferencing optimized system like this could split the context for different requests if it doesn't need to use the full context.

Could also be that they have internal use cases which require this amount of context.