Solutions

A GPU for every AI workload

Dedicated NVIDIA accelerators on bare metal. Pick the GPU that matches your workload and scale on demand.

The line-up

Choose your accelerator

Each GPU comes with the same sovereign, single-tenant infrastructure and 24/7 expert support.

Blackwell · Flagship

NVIDIA HGX B200

from $2.69 per GPU/hr

The most advanced accelerator for AI and HPC.

Eight NVIDIA B200 SXM GPUs (1,440 GB HBM3e, fully NVLink-connected) paired with dual Intel Xeon 6972P CPUs (192 cores) and 1.5 TB of DDR5 memory. Every GPU has a dedicated 400 Gb/s NVIDIA ConnectX-7 InfiniBand link for high-bandwidth multi-node scaling, plus 2x 100 GbE for data and storage and a separate management network. Includes 8x 3.84 TB local NVMe.

192GB HBM3e
Up to 15X inference vs Hopper

Generative AI
Training
HPC

View details Reserve GPUs

Blackwell Ultra · Next-gen

NVIDIA HGX B300

from $2.93 per GPU/hr

The most memory yet, for frontier-scale AI.

Eight NVIDIA B300 SXM GPUs (Blackwell Ultra) with 288GB of HBM3e each — 2,304 GB total, NVLink-connected — for the largest models, reasoning workloads and real-time inference. Configured as a fully integrated 8-GPU node with high-bandwidth InfiniBand for multi-node scaling.

288GB HBM3e
1.5X FP4 compute vs B200

Frontier models
Reasoning
Training

View details Reserve GPUs

Rubin platform

NVIDIA Vera Rubin coming soon

NVIDIA's next-generation Vera Rubin platform is on the way — the successor to Blackwell, purpose-built for the next wave of frontier AI. Specs and availability to be announced.

Not sure which GPU you need?

Tell us about your workload and our team will help you size the right configuration — from a single GPU server to multi-node bare-metal clusters.

Talk to our team