GPU clusters
Dedicated NVIDIA HGX pods on multi-year terms, for enterprise AI workloads.
Platform
We build on NVIDIA HGX and NVL72-class systems, air- or liquid-cooled, and successor platforms as they ship. HGX B300 is our reference configuration. We spec the platform to the workload.
| Reference configuration | NVIDIA HGX B300 |
| GPUs per node | 8 |
| NVLink domain | 8 GPUs |
| Form factor | 8U, air-cooled |
What we deliver
Bare-metal nodes, a dedicated fabric, and a storage tier. You run your own scheduler, images, and models. Nothing is shared with another tenant.
Workload fit
Domain size follows the workload. HGX B300 gives an eight-GPU NVLink domain. NVL72-class systems give a 72-GPU domain. We size it to the job.