Nvidia's dream would be for everyone to buy a personal DGX H100 for private local inference. That's where open source could lead. Datacenters are much more efficient in their use of chips.