Key specifications
Illustrations are for reference. Service scope is defined by the plan you choose.
Save 5% with a 2-year term or 10% with a 3-year term.
Dedicated NVIDIA GPU servers on annual terms for AI training, inference, rendering and simulation, with full root access.
GPU Server Rental gives your data science and engineering teams dedicated NVIDIA GPUs without a hardware purchase, long procurement cycles or unpredictable hourly cloud bills. Each server is single-tenant: the GPUs, CPUs, memory and NVMe storage are reserved for you for the whole term.
Servers are delivered with your choice of Ubuntu LTS or Rocky Linux, current NVIDIA drivers, the CUDA toolkit and a container runtime installed and tested. You get full root access; we handle hardware health monitoring, failed-component replacement and the network.
Choose a tier by GPU memory and interconnect: the L40S suits inference, fine-tuning and visual computing, while H100 SXM servers with NVLink are built for large-scale training.
All prices in USD, per server per year at the 1-year rate. Swipe the table sideways to see every plan.
| Feature | Business | Professional Recommended | Enterprise |
|---|---|---|---|
| Price | $16,800.00 / year per server | $49,000.00 / year per server | $189,000.00 / year per server |
| With a 3-year term | $15,120.00 / year Save 10% | $44,100.00 / year Save 10% | $170,100.00 / year Save 10% |
| Best for | One NVIDIA L40S for inference, fine-tuning and visual computing. | Four L40S GPUs for multi-GPU training and high-throughput inference. | Eight H100 SXM GPUs with NVLink for large-model training. |
| What’s included |
|
|
|
| Order | Customize term & quantity for the Business plan | Customize term & quantity for the Professional plan | Customize term & quantity for the Enterprise plan |
| SKU | TS-RNT-301 |
|---|---|
| Category | Server & GPU Rental |
| Pricing | Per server, per year |
| Contract | Annual subscription; 1-, 2- or 3-year term |
| GPU | 1× L40S 48 GB / 4× L40S 48 GB / 8× H100 80 GB SXM |
| CPU & memory | From 16 cores and 128 GB up to 2× 48 cores and 2 TB |
| Storage | NVMe SSD, 3.84 TB to 30.72 TB depending on tier |
| Network | 1, 10 or 25 Gbps port; 20 to 100 TB monthly transfer |
| OS | Ubuntu LTS or Rocky Linux; others on request |
| Access | Root SSH, IPMI/KVM console, optional VPN |
| Support | Hardware replacement and response times by tier |
| Lead time | Provisioned in 3–5 business days |
No. Every GPU in your server is dedicated to you for the full term. There is no time-slicing or virtual GPU partitioning unless you configure it yourself.
You can upgrade to a higher tier or add servers at any time; the change is prorated to your renewal date. Downgrades take effect at renewal.
We install the operating system, NVIDIA drivers, the CUDA toolkit and a container runtime. Frameworks such as PyTorch or TensorFlow can be preinstalled on request; support for your own application code is not included.
Your data stays on the server until the last day of the term, and we recommend copying it off during the final two weeks. Drives are securely wiped before the hardware is reassigned.
Have another question? Ask a solutions architect
