AI & GPU Servers

Mineshop.eu supplies enterprise AI servers built on NVIDIA HGX – the eight-GPU Blackwell platform behind NVIDIA's DGX systems – engineered by Supermicro. Choose HGX B300 (Blackwell Ultra, 288 GB HBM3e per GPU) for large-scale inference and reasoning, or HGX B200 for training, fine-tuning and HPC. Every system is configured to order, priced on request and delivered to data centres across the EU.

Compare NVIDIA HGX AI servers

Choosing the right GPU server

• HGX B300 vs B200: B300 doubles attention performance and adds 60 % more memory per GPU – best for inference at scale. B200 keeps 296 TFLOPS of FP64 – best for training plus HPC.
• Intel or AMD: Intel Xeon 6 (6900P / 6700P) or AMD EPYC 9005 hosts. Choose by fleet standard, memory capacity (up to 8 TB with 32 DIMMs on the SYS-822GS-NB3RT) and software licensing per core.
• 8U or 10U: the HGX B300 systems use a compact 8U chassis with eight 800G ConnectX-8 ports built in; the HGX B200 systems use 10U with PCIe slots for 400G NICs.
• Gold Series or custom: Gold Series presets are pre-configured and pre-tested for shorter lead times; custom builds let you choose every component.
• Power and cooling: six Titanium PSUs in 3+3 redundancy (5250 W or 6600 W each), 200–240 V, inlet air 10–35 °C – plan rack power and airflow before ordering.
• Networking: one 400G (B200) or 800G (B300) port per GPU for multi-node training and inference clusters, plus storage and management networks.

Ordering process

1. Open a server and click Request a quote.
2. A hardware specialist confirms the configuration, quantity and delivery location with you.
3. You receive a written quote within 1 business day with lead time, service plan and payment terms.
4. We build, test and deliver – with optional rack integration and onsite installation.

Frequently asked questions

How much does an 8-GPU NVIDIA HGX server cost?

An HGX B200 or HGX B300 system is a six-figure investment, and the final price depends on GPU allocation, CPUs, memory, storage, networking and service plan. Because GPU pricing changes frequently, every system is quoted individually – you receive a written B2B quote, excluding VAT, within 1 business day.

What is the difference between NVIDIA HGX and NVIDIA DGX?

HGX is NVIDIA's eight-GPU baseboard that server makers such as Supermicro build into their own systems. DGX is NVIDIA's branded system built on the same HGX design, with a fixed configuration and NVIDIA's software and support bundle. An HGX server gives you the same GPUs and NVLink fabric with your choice of CPUs, memory, storage and networking.

Which is better for me, HGX B300 or HGX B200?

Pick HGX B300 (Blackwell Ultra, 288 GB per GPU) for large-scale LLM inference, reasoning and long-context models. Pick HGX B200 (180 GB per GPU) for training, fine-tuning and mixed AI + HPC work: it keeps strong FP64 performance (296 TFLOPS per system versus 10 TFLOPS on B300).

How long does delivery take?

Supermicro Gold Series HGX B300 presets currently reach Europe in about 6–12 weeks. Custom HGX configurations can take up to ~30 weeks depending on GPU allocation. Every quote states the confirmed lead time.

Can a bitcoin mining site host AI servers?

Often yes – mining sites already have the power capacity, but AI servers also need redundant (2N) power feeds, 200–240 V distribution, clean inlet air at 10–35 °C, fibre connectivity and higher uptime than ASICs. Our team can review your site and propose a realistic first deployment.

Do you deliver outside the EU?

We deliver across all 27 EU countries. Other destinations are possible on request after export-control and end-user screening; NVIDIA data-centre GPUs are export-controlled and cannot be supplied to embargoed countries.