Hardware
Sovereign inference hardware, from the DGX Spark workstation to the GB300 NVL72 rack, across seven power tiers for every organisation size.
DGX Spark
The sovereign workstation of a small business puts a private model on the desk, with no reliance on the cloud.
DGX Station
The mid-size company gets a rack's worth of power in a tower case to develop and serve without a server room.
RTX PRO 6000 server
The sovereign workhorse runs air-cooled on standard servers, with a contained cost per token.
H200 SXM server
The HGX H200 board makes eight GPUs work as a single accelerator via NVSwitch, for training and inference of large models.
B200 SXM
According to NVIDIA, the move to Blackwell reaches up to fifteen times the Hopper generation in inference on large models.
B300 SXM
The top-of-range server adds sixty per cent more memory per GPU over the B200, for reasoning and long context.
GB300 NVL72
The large enterprise hosts its own frontier model in a rack, as a sovereign AI cloud.