OVHcloud
VPS-1
- 2 vCPU
- 4 GB RAM
- 40 GB NVMe
from$4.54/mo
- Unlimited traffic
- Anti-DDoS included
Price checked September 28, 2026
Advertising disclosure: We may earn a commission when you buy through links on this page. It does not change the price you pay or the order of the list.
Guide
Ollama on a VPS: run your own AI model
Small models (3–8B at 4-bit) run on a CPU-only VPS with 8–16 GB RAM - slowly, but for pennies. For chat that feels instant, or for anything above ~14B, rent a GPU by the hour instead.
CPU-only: what to expect
On a modern VPS with 4–8 vCPUs a 3B model answers at a comfortable reading pace, and an 8B model is usable for background jobs - summaries, classification, n8n automations. Mixture-of-experts models are a good fit because only a small part works per token.
When a GPU is worth it
Interactive chat for several people, models from 14B up, long documents or image generation. A single 24 GB card covers most of it; check the exact number in the VRAM calculator.
Setup
Install Ollama with its one-line script, pull a model, and put Open WebUI in front if you want a ChatGPT-style interface. Keep the Ollama port closed to the internet and reach it through the web UI or a VPN.
VPS hosting
We may earn a commission when you buy through links on this page. It does not change the price you pay or the order of the list.
IONOS
VPS M+
- 2 vCPU
- 4 GB RAM
- 120 GB NVMe
from$5.00/mo
- Unlimited traffic
- Very cheap first months
Price checked September 28, 2026 · Renews at $14.00/mo
Contabo
Cloud VPS 4
- 4 vCPU
- 8 GB RAM
- 100 GB SSD
from$6.26/mo
- Very high RAM/storage per euro
- Unlimited traffic (fair use)
Price checked September 28, 2026
Questions people ask
Which model should I start with?
A 4-bit 7–8B model such as Llama 3.1 8B or Qwen3 8B is a good first test: it fits in 8 GB and handles most everyday tasks.
Is my data private with Ollama?
Yes - everything runs on your server. Just do not expose the API port publicly without authentication.