Source: Hetzner Pressroom
Hetzner has expanded its portfolio with the GEX430, the German hosting provider’s first multi-GPU dedicated server. The machine comes with four NVIDIA RTX PRO 6000 Blackwell Max-Q GPUs and a total of 384 GB of GDDR7 ECC graphics memory, aimed at businesses that need serious computing power for AI training, inference of large language models, fine-tuning, and demanding rendering and simulation work.
Why graphics memory matters for AI workloads
According to Hetzner, memory is just as crucial as raw computing power when it comes to AI inference. The more requests a model handles at the same time, the more memory it generally needs. With four GPUs and a combined 384 GB of VRAM, the GEX430 offers headroom for companies that want to run multiple language models in parallel or serve a large number of concurrent users, for example AI chatbots in customer support or internal knowledge assistants for employees.
Each of the four GPUs carries 96 GB of GDDR7 ECC VRAM and supports fifth-generation Tensor Cores, fourth-generation RT Cores, and FP4. The rest of the configuration includes an Intel Xeon Gold 5412U processor with 24 cores and 48 threads, 128 GB of DDR5 ECC registered memory, and two 1.92 TB Datacenter Edition (Gen4) NVMe SSDs.
European AI infrastructure without external dependencies
The interesting angle for European businesses is sovereignty. Hetzner points out that companies can run open-weight AI models such as GLM, Mistral, or Qwen on dedicated hardware within the EU without relying on external AI platforms. With the right configuration, sensitive information such as internal documents, customer data, or source code stays inside the company’s own server environment, which helps when implementing data protection and compliance requirements.
The GEX430 is available for EUR 3,599.00 per month plus a one-time setup fee of the same amount, excluding VAT. Hetzner operates its data center parks in Nuremberg and Falkenstein in Germany and Helsinki in Finland, with additional locations in Singapore and the USA.
Running AI workloads on European hardware is becoming a practical compliance decision, not just a preference. If you want help planning AI infrastructure that keeps your data under your control, reach out to us at our contact page for professional advice.