Guide · Local AI
Run your own LLM in the cloud without buying a GPU
By Tiny · how I pick
As an Amazon Associate I earn from qualifying purchases. The Amazon hardware links below are affiliate links. Full disclosure
If you want your own LLM without buying a graphics card, pick based on how much you'll use it:
- A small private chatbot that's always on: a cheap VPS running Ollama is the easiest start.
- Bigger models or occasional heavy use: rent a GPU by the hour on RunPod or Vast.ai.
- Hours of use every day, or privacy is the whole point: owning hardware wins over time.
I run Ollama on my own hardware at home and reach it from anywhere through a Cloudflare Tunnel. Not everyone has a spare machine, though, and renting is a sensible way to try this before spending money on a graphics card.
Words you'll see in this guide
- LLM: large language model, the kind of AI behind ChatGPT and Claude. A local LLM runs on your own computer.
- VPS: a small rented server in a data center that runs around the clock. You control it like your own computer.
- vCPU: a share of a processor on a rented server. More vCPUs means more processing power.
- GPU: the graphics card. It does the heavy math for AI models much faster than a processor.
- VRAM: the memory on your graphics card. AI models load into it, so more VRAM means you can run bigger models.
- Cloudflare Tunnel: a free service that makes an app running at home reachable from anywhere, without opening ports on your router.
The three ways to run your own LLM
| Option | Best for | The catch |
|---|---|---|
| Ollama on a VPS | A private chatbot that's always on, with small models | No graphics card, so it's slow and limited to small models |
| Rented GPU by the hour | Big models, image generation, occasional heavy use | You pay while it runs, and you have to remember to shut it down |
| Your own hardware | Daily use and full privacy | Upfront cost, plus noise and power |
Option 1: Ollama on a VPS (easiest start)
Hostinger has a one-click Ollama template that installs Ollama, a Llama 3 model, and Open WebUI, a ChatGPT-style chat page in your browser (Hostinger's guide). You get your own private chat app without touching a terminal.
The catch: these plans have no graphics card. Hostinger's own guide says reply speed depends on the model and the number of processor cores. In practice, that means small models only, and slower replies than a graphics card would give you.
| Hostinger plan | Processor / RAM | What fits |
|---|---|---|
| KVM 2 | 2 vCPU / 8 GB | Small models, roughly 3B to 8B at 4-bit |
| KVM 4 | 4 vCPU / 16 GB | Up to about 14B at 4-bit, still slow |
| KVM 8 | 8 vCPU / 32 GB | Mid-size models, more comfortably |
Pick this if: you want a private chatbot that's always available, you're fine with small models, and you don't want to manage hardware.
Skip it if: you want big models, fast replies, or image generation.
Option 2: Rent a GPU by the hour
RunPod and Vast.ai rent you a real graphics card by the hour. You start a machine, run Ollama or another tool on it, and shut it down when you're done.
- RunPod bills by the second for on-demand machines (RunPod pricing).
- Vast.ai is a marketplace of other people's graphics cards, so prices move with supply (Vast.ai).
| GPU on RunPod | VRAM | Listed price |
|---|---|---|
| RTX A5000 | 24 GB | About $0.27/hr |
| RTX 3090 | 24 GB | About $0.50/hr |
| RTX 4090 | 24 GB | About $0.74/hr |
| RTX A6000 | 48 GB | About $0.53/hr |
Watch out: you pay for every hour the machine is on, even when you're not using it, so always shut it down when you finish. Keeping your models stored between sessions also costs extra.
Pick this if: you want big models or image generation a few hours a week, or you want to test before buying a graphics card.
Skip it if: you want AI running all day, every day. The hours add up fast.
Option 3: Run it on your own hardware
Once you use AI for more than a couple of hours a day, owning a graphics card usually beats renting one. I walk through the break-even math in rent a GPU vs build your own.
For the hardware, a used RTX 3060 12 GB is the cheapest real start, and a used RTX 3090 is the step up, though used 3090 prices have climbed a lot this year. More picks in the best GPU for local AI under $500, or check what your current PC can handle with the Server Calculator.
Check RTX 3060 price on Amazon
A word on privacy
A VPS or a rented graphics card is still someone else's computer. It's far more private than typing into a big AI company's chatbot, because the model and your chats live on a server only you control. But if privacy is your main reason, your own hardware at home is the only option where your data never leaves the house. More in is local AI private?
What I'd tell most beginners
- Start with a small VPS to learn how Ollama and Open WebUI work.
- If you want bigger models, rent a graphics card for a few evenings.
- If you're still using it every day a month later, buy hardware.
That way you only spend real money once you know you'll use it.
Quick answers
Can I run my own LLM in the cloud without a GPU?
Yes. A VPS without a graphics card can run small models with Ollama, like 3B to 8B models on 8 GB of RAM. Replies are slower than on a graphics card.
What's the cheapest way to run your own LLM?
For light use, a small VPS or a graphics card rented by the hour. For heavy daily use, owning a used graphics card usually works out cheaper over time.
Is running an LLM on a VPS private?
It's much more private than a public chatbot, since the model and chats sit on your own server. But that server is still in someone else's data center. For full privacy, run it at home.
Does Hostinger's Ollama VPS have a GPU?
No. Its Ollama plans run on the processor only, so they suit small models rather than large ones.
Free email updates
Cloud GPU prices drop every few months. Subscribe and I'll tell you when renting gets cheaper than it is today.
No spam. Unsubscribe anytime. Privacy policy.