NVIDIA AI servers and rack-scale systems from leading OEMs, across current and upcoming generations, specified for your workload and the facility it will run in.
We source OEM-built NVIDIA systems for GPU cloud operators and AI companies building or expanding their fleets. Bring a preferred platform, a full specification or just the workload you need to support.
NVIDIA HGX platforms in a single chassis, for inference and fine-tuning, or as the building block of a larger cluster.
An entire rack built as one machine, for the largest training and inference workloads.
Rack power from NVIDIA’s GB300 NVL72 reference architecture. Configurations, cooling and supply terms are confirmed in each proposal.
Discuss systems across NVIDIA’s data-centre platforms, from Blackwell in service today to the Rubin generations that follow. You choose the platform for the workload; we are not tied to one generation.
Each quotation names the exact model and configuration, whether it is an NVIDIA-Certified System, and the support and supply terms.
NVIDIA
| Generation | 8-GPU server | Rack-scale system | NVLink scale-up configuration | GPU memory type | NVIDIA platform status |
|---|---|---|---|---|---|
| Blackwell | HGX B200 | GB200 NVL72 | NVL72: 72 GPUs per rack | HBM3e | Shipping |
| Blackwell Ultra | HGX B300 | GB300 NVL72 | NVL72: 72 GPUs per rack | HBM3e | Shipping |
| Vera Rubin | HGX Rubin NVL8 | Vera Rubin NVL72 | NVL72: 72 GPUs per rack | HBM4 | In production |
| Rubin CPX | — | Announced for context processing | Configuration to be confirmed | — | Announced; availability to be confirmed |
| Rubin Ultra | To be announced | Rubin Ultra NVL144 (Kyber rack) | NVL144: one Kyber rack. NVL576: eight 72-GPU racks combined | HBM4E (announced) | Announced |
| Later generations | As announced | As announced | — | — | Followed as NVIDIA announces them |
From NVIDIA public product and roadmap material, checked November 2026. Roadmap details and timing are NVIDIA’s and may change.
Networking, storage, rack space, power and cooling belong alongside the hardware specification. If you need somewhere to operate the systems, colocation can be part of the same discussion.
Explore colocation
GPU servers and rack-scale systems, chosen for the workload and the generation you want to run.
The fabric that turns systems into a cluster, and the data paths that keep it fed.
High-density racks move the constraint to power delivery and heat rejection, often liquid to the chip.
Racks, a dedicated suite or a hall, in the right location, with room to grow.
A full specification or a first outline both work. Share what is decided and what is still open.
You can start with a defined system or a broader infrastructure requirement. Share the information you have and the decisions still to be made.
Yes. Include them in the brief; the quotation sets out what we propose to supply and how it connects to your wider infrastructure.
Each quotation specifies warranty and support for the systems quoted. Tell us the coverage you need in your enquiry.
Current Blackwell and Blackwell Ultra systems, Vera Rubin as it reaches the market, and later generations as NVIDIA announces them. Availability is confirmed per proposal.
