HP ZGX Nano G1n AI Station

The GB10 Superchip pairs Grace CPU cores with a Blackwell GPU across a coherent memory space, so 128GB is addressable by both rather than divided between them. That is what lets a machine this size hold models that would otherwise need a server.

The argument for it is access rather than throughput. Development, fine-tuning and evaluation that currently waits for a slot on a shared cluster runs at a desk instead, and the cluster keeps its time for training runs that genuinely need it.

It runs the same NVIDIA software stack as the rack systems, so work moves between the two without being rewritten. That matters more than any single specification: a deskside machine needing its own toolchain is a second thing to maintain rather than a smaller version of the first.