Instantly available in Oslo – sovereignly hosted in Europe
8-GPU server platform based on NVIDIA Blackwell Ultra. Available as bare metal or GPUaaS in Europe - with the performance of a hyperscaler and the control of your own infrastructure. Without months of waiting.
The NVIDIA HGX B300 is an 8-GPU server platform built on the NVIDIA Blackwell Ultra architecture. It is designed for the most demanding generative AI workloads: large language model training, fine-tuning, high-throughput inference, and multimodal AI.
Each HGX B300 node combines eight NVIDIA Blackwell Ultra GPUs with fifth-generation NVLink and high-bandwidth interconnects. Compared with the previous generation, it delivers substantially more memory per GPU and stronger performance on attention-heavy models — the type of workloads that dominate modern AI reasoning and long-context applications.
For teams running AI at scale, the B300 means:
The extra memory and compute density of the B300 make it practical to train and fine-tune larger models with longer sequences — without the engineering workarounds that slower hardware demands.
Serve more concurrent users from the same hardware. The B300’s memory capacity and bandwidth help reduce latency and increase throughput for production AI APIs.
Build AI systems that seamlessly integrate text, image, audio, and video. The balanced compute-and-memory architecture of the HGX B300 is well suited for next-generation multimodal and long-context reasoning workloads.
| HGX B200 | HGX B300 |
|---|---|
| 8x NVIDIA Blackwell SXM | ✓ 8x NVIDIA Blackwell Ultra SXM |
| 144 PFLOPS | 72 PFLOPS | ✓ 144 PFLOPS | 108 PFLOPS |
| 72 PFLOPS | ✓ 72 PFLOPS |
| 72 POPS | ✓ 3 POPS |
| 36 PFLOPS | ✓ 36 PFLOPS |
| 18 PFLOPS | ✓ 18 PFLOPS |
| 600 TFLOPS | ✓ 600 TFLOPS |
| 30 TFLOPS | ✓ 10 TFLOPS |
| 1.4 TB | ✓ 2.1 TB |
| Fifth generation | ✓ Fifth generation |
| NVLink 5 Switch | ✓ NVLink 5 Switch |
| 1.8 TB/s | ✓ 1.8 TB/s |
| 14.4 TB/s | ✓ 14.4 TB/s |
| 0.8 TB/s | ✓ 1.6 TB/s |
| 1x | ✓ 2x |
Performance metrics are based on NVIDIA published specifications for HGX B300. Real-world performance depends on workload characteristics, model architecture and software optimization.
Polarise operates AI-ready and sovereign cloud infrastructure in Germany and Europe, helping teams meet data residency, latency and compliance requirements without routing workloads to foreign regions. If your users, regulators or customers expect data to stay in Europe, Polarise keeps it there.
Polarise is an official NVIDIA Cloud Service Provider and NVIDIA Preferred Partner. Our platform is streamlined and designed against NVIDIA architecture across multiple GPU generations for best and guaranteed performance.
Polarise has achieved NVIDIA Preferred Partner status and is listed as official NVIDIA Cloud Service Provider (CSP), solidifying our position as a trusted leader in cloud innovation. This designation is reserved for select partners who operate large clusters built in coordination with NVIDIA, adhering to a tested and optimized reference architecture.
Launch your first HGX B300 instance and tell us about your workload — we will respond within one business day.
