Contabo just launched GPU VPS, and the pitch is refreshingly narrow for a market that usually runs on vague promises: one dedicated Nvidia RTX PRO 6000 GPU, wired into a standard VPS, for a flat €999 a month with no metered billing and no full-server commitment.
That specificity matters, because it targets a gap the GPU rental market has mostly ignored. Hyperscalers offer managed access, but billing by the token or the hour can turn a growing workload into an unpredictable bill fast. Spot and marketplace GPU providers run cheaper, but availability and pricing shift constantly, which makes planning a real project difficult. Dedicated GPU servers offer stable capacity, but they demand a higher upfront cost and force customers to commit to an entire machine even for a single-GPU job. Contabo positions GPU VPS squarely between all three.
The card itself explains why this configuration works for the workloads Contabo is targeting. The RTX PRO 6000 carries 96 GB of GDDR7 memory, enough to hold a 70-billion-parameter model on a single card, alongside 24,064 CUDA cores and roughly 1.8 TB/s of memory bandwidth.
For anyone running local LLM inference, RAG pipelines or AI agents, that memory ceiling is the difference between staying on one GPU and fragmenting a workload across several instances just to fit it. The card also carries ISV certification for tools like Maya, Houdini and Cinema 4D, which matters directly for rendering and VFX studios juggling deadline spikes a single workstation can’t absorb.
The instance pairs that GPU with 18 vCPUs, 96 GB of RAM and 900 GB of NVMe storage. It runs Ubuntu 24.04 with CUDA preinstalled, so there’s no driver setup standing between signup and actual use. Contabo is launching the product in Europe and US Central only for now, offering 15 percent off on 12-month commitments and 20 percent off on 24-month terms for anyone willing to lock in longer.
Contabo says plainly that this product does not target large-scale, multi-node model training, where H100 or A100 clusters with NVLink remain the better fit. GPU VPS targets single-node work instead, inference, rendering, simulation, where reserving one serious GPU without buying or leasing an entire dedicated server actually solves a real problem instead of just adding another pricing tier to choose from.
