The hardware being offered
Hetzner’s October 21 GEX430 announcement introduces a dedicated server with four NVIDIA RTX PRO 6000 Blackwell Max-Q GPUs. Each has 96 GB of graphics memory, giving the machine 384 GB in total. The product page lists an Intel Xeon Gold 5412U, 128 GB of system RAM and two 1.92 TB NVMe drives.
The listed location is HEL 1, and the operating-system options include Debian and Ubuntu. Hetzner positions the system for AI workloads, rendering and simulation; CGH has not tested those workloads on it.
Price and an important memory distinction
The announcement quotes $4,199 per month and a separate $4,199 setup charge, excluding VAT. That is issuer launch pricing, not a guaranteed checkout total or capacity reservation. The combined graphics-memory figure is spread across four devices. It should not be read as a promise that any application can automatically use one uninterrupted 384 GB memory space.
CGH perspective
The useful comparison is the cost of the actual deployment, including setup, administration and the time required to distribute a model across GPUs. A team should establish whether it needs one large model or several independent workloads before selecting hardware. A representative trial would measure concurrency, memory use and recovery from a failed process. Dedicated hardware gives configuration choices, while leaving responsibility for application operations with the customer.