AI/ML VPS Rental — PyTorch and TensorFlow

Order a server for your ML environment and build your own stack — a 5-hour test period before payment, so you risk nothing

Order an ML server
Ordering a VPS for machine learning

An AI/ML VPS is a server with root access and NVMe drives where you build the environment yourself: CUDA libraries, PyTorch, TensorFlow, Jupyter and your own conda environments. The plans in the price list come without a graphics card: they are sized for data preparation, classical ML and inference of small models. If you need a GPU, there are ready GPU servers with a Tesla P4 from €249/mo, and configurations with more powerful cards are built to spec. The server is issued in a 5-hour test mode — verify your environment is compatible before paying. Not sure what to pick — see the guide on choosing a server for machine learning.

Plans

VPS without a graphics card

These plans are sized for workloads the CPU handles: data preparation, classical ML, inference of small models and development. A configuration with a graphics card is not part of the plan and is built per task — write to us. 5-hour test period before payment.

Compare all plans
5.49 mo

Standard 1

4 vCPU · AMD EPYC
4 GB RAM
50 GB NVMe
Unlimited traffic
IPv4 · KVM
24/7 support
6.99 mo

Standard 2

4 vCPU · AMD EPYC
6 GB RAM
60 GB NVMe
Unlimited traffic
IPv4 · KVM
24/7 support
1

What the plan includes

Dedicated vCPU on AMD EPYC, RAM, an NVMe drive, unmetered traffic, one IPv4, a KVM console and 24/7 support. A graphics card is not included.

2

Your own ML environment

Install any versions of Python, conda, CUDA libraries and frameworks — PyTorch, TensorFlow, scikit-learn — with no restrictions.

3

GPU as a separate service

For neural networks there are ready GPU servers with a Tesla P4 from €249/mo; configurations with more powerful cards are built individually.

How to order and set up the environment

Three steps — from placing the order to your first trained model.

1

Place the order

Pick a plan for the RAM and disk space you need, set the billing period and choose an OS image — usually Ubuntu or Debian, the standard choices for an ML stack.

2

Install the OS yourself

Through the panel and the KVM console you deploy the system and get hardware-level access to the server.

3

Build the stack

Over SSH install Python, the frameworks and Jupyter, upload your datasets to NVMe and start training or inference.

Not sure which plan to take

Not sure which plan to take

Start with memory: the model weight in your chosen quantisation plus context and the system. A table of configurations for inference, ETL and classical ML is on the page about choosing a server for machine learning.

If the task needs a graphics card, see the GPU server comparison: it covers what matters beyond the card model and how a configuration is built for the task.

How to size the configuration

Frequently asked questions

Didn't find your answer?
Write to us via the contact page.

No, the VPS plans come without a GPU. A server with a graphics card is a separate service: ready configurations with a Tesla P4 from €249/mo, and builds with more powerful cards are quoted individually.

Pick a plan, set the billing period and OS image, register — the server is created automatically. Payment is made after the test, not before the order.

5 hours. That is enough to install the OS, build the environment and run inference or a short training pass on the production configuration.

Yes. Root access lets you install any versions of Python, CUDA libraries, PyTorch and TensorFlow — you build the environment entirely for your task.

You do: choose the image at checkout and deploy the system through the panel or the KVM console. Most people take Ubuntu or Debian.

Monthly, six months with a 10% discount, and a year with an 18% discount. The discount also applies to renewals, not just the first period.

Yes, you move to a higher plan with a short restart: more vCPU, RAM and disk are added. Your environment and datasets stay in place.

Count by memory: the model weight in its quantisation plus context and the system. A task-by-task table is on the page about choosing a server for machine learning.