Not always. Small and medium quantised models run on CPU servers with enough memory, which suits private assistants, document search and batch jobs. Answers come more slowly than on a GPU. For real-time use with large models or many users, ask our team about GPU options.

How much RAM do I need?

Roughly the number of parameters times the bytes per parameter, plus about 20% overhead and room for context. An 8B model at 4-bit needs around 7GB, while a 70B model at 4-bit needs around 45GB. Use the sizing guide above and leave headroom for your application.

Is my data used to train anyone’s model?

No. The model runs on your server, and your prompts and documents are not sent to an AI provider. What you store and log on the server is under your control.

Can you set up the model for me?

Yes. Tell us the model and how you plan to use it, and our engineers can help install a runtime such as Ollama with a web interface or an API endpoint.

Private AI and LLM hosting

Run open AI models on a server that only you control.

Host Llama, Mistral, Qwen and other open models for chat, search and automation. Your prompts and documents stay on your own server instead of a third-party API.

Size my server See AI plans

AI servers from RM40.53 / month

Private assistantexample: an 8B open model

Question: What notice period do our supplier contracts require for renewal?

Illustrative. Speed depends on the model, its quantisation and the server you run it on.

Open models you can run

Popular open-weight model families that run with tools such as Ollama, llama.cpp or vLLM. Check each model’s licence for commercial use.

Llama
Meta
1B, 3B, 8B, 70B
General chat, summaries and agents
Mistral and Mixtral
Mistral AI
7B, 8x7B, 24B
Fast assistants and European languages
Qwen
Alibaba
0.5B to 72B
Multilingual work and coding
Gemma
Google
2B, 9B, 27B
Compact models for focused tasks
Phi
Microsoft
3.8B, 14B
Small models that reason well
DeepSeek R1 distills
DeepSeek
1.5B to 70B
Step-by-step reasoning
Embedding models
Various
Under 1B
Search and retrieval over your documents
Whisper
OpenAI
Up to 1.5B
Speech to text, kept private

Your data never leaves your server

With a public AI API, every prompt and every document you send is processed on someone else’s infrastructure. On your own server, the model, the data and the logs all live in one place you control.

Contracts, records and customer data stay in-house
No per-token bills: a fixed monthly server cost
Choose, fine-tune and swap models freely
Connect your own apps, n8n workflows and databases

Public AI API

Your app

Provider’s serversprompts, files, logs

Your own server

Your app

Open modelyour documents, your logs

How much memory does a model need?

Pick a model size and precision. We estimate the memory it needs and list servers from our catalog that can hold it.

Model size8B parameters

Precision (quantisation)

Rule of thumb: parameters × bytes per parameter, plus about 20% and a little room for the conversation context. Leave extra memory for your application and operating system.

Memory needed, roughly

7 GB

Servers in our catalog with enough memory, with headroom where possible:

H-LINUX-VPS-0216GB RAM, 6 vCPURM73.76 / month
H-LINUX-VPS-0324GB RAM, 8 vCPURM103.25 / month
Pro - Europe12GB RAM, 6 vCPURM118.95 / month

Servers listed are from our current catalog and run models on CPU. For GPU servers, ask our team.

AI hosting plans

Servers prepared for running open models.

Starter

4 vCPU Cores, 6GB RAM DDR4

RM1313.17/ 3 years

Billed every 3 years, about RM36.48 a month, save 10%

4 vCPU Cores
6GB RAM DDR4
100GB NVMe SSD Storage
Llama 3.2 3B (Primary Model)
Phi-3 Mini 3.8B (Fast Alternative)
Llama 3.2 1B (Ultra-fast)
20-40 tokens/sec Speed

Show all 14 features

3-5 Concurrent Users
Free SSL Certificate
API Key Authentication
Nginx Reverse Proxy
Auto-restart on Reboot
30 req/min Rate Limit
7-day Log Retention

Order AI server View details

Recommended

PROFESSIONAL AI

6 vCPU Cores, 16GB RAM DDR4

RM2496.19/ 3 years

Billed every 3 years, about RM69.34 a month, save 10%

6 vCPU Cores
16GB RAM DDR4
200GB NVMe SSD Storage
Llama 3.2 8B (Primary - Best Quality)
Mistral 7B (Fast Alternative)
CodeLlama 13B (Coding Specialist)
Llama 3.2 3B (Speed Backup)

Show all 17 features

15-25 tokens/sec Speed
10-15 Concurrent Users
Free SSL Certificate
Multiple API Keys Support
Nginx Proxy + Caching
Health Monitoring
100 req/min Rate Limit
30-day Log Retention
Multi-model API Routing
Priority Support

Order AI server View details

ENTERPRISE AI

8 vCPU Cores, 24GB RAM DDR4

RM3416.32/ 3 years

Billed every 3 years, about RM94.90 a month, save 10%

8 vCPU Cores
24GB RAM DDR4
300GB NVMe SSD Storage
Mixtral 8x7B (47B - GPT-3.5 Level)
Llama 3.1 13B (High Quality)
CodeLlama 13B Instruct
Mistral 7B (Fast Queries)

Show all 19 features

Llama 3.2 8B + Gemma 2 9B
8-15 tokens/sec Speed (Mixtral)
20-30 Concurrent Users
Free SSL Certificate
Unlimited API Keys
Nginx + Load Balancing
Advanced Monitoring
500 req/min Rate Limit
90-day Log Retention
Multi-model Load Balancing
Automatic Model Switching
Priority Support + SLA

Order AI server View details

Prices shown live from our billing system, in MYR.Browse dedicated servers

What you get with private AI hosting

Private by design

Prompts, documents and outputs are processed and stored on your server only.

Memory for real models

Servers in our catalog go up to 256GB of RAM, enough for large quantised models.

Full root access

Install Ollama, llama.cpp, vLLM, Open WebUI or your own stack.

OpenAI-compatible APIs

Tools like Ollama and vLLM expose an API your existing code can call with a URL change.

Plug into automation

Connect models to n8n workflows, internal tools and databases on the same server.

Predictable cost

A monthly server price instead of a bill that grows with every token.

Search your own documents

Run embeddings and a vector database next to the model for retrieval over your files.

Engineers on call

Help choosing a model and server size, by phone, email, WhatsApp and ticket, around the clock.

AI hosting questions

Planning something specific? Our team can size a server with you.

Any open-weight model your server has the memory for, including Llama, Mistral, Qwen, Gemma, Phi and DeepSeek distills, as well as embedding and speech models. Check each model’s licence before commercial use.

Bring your AI in-house

Tell us the model you want to run and we will recommend a server sized for it.

Talk to our AI team Size my server

Run open AI models on a server that only you control.

Open models you can run

Llama

Mistral and Mixtral

Qwen

Gemma

Phi

DeepSeek R1 distills

Embedding models

Whisper

Your data never leaves your server

How much memory does a model need?

AI hosting plans

Starter

PROFESSIONAL AI

ENTERPRISE AI

What you get with private AI hosting

Private by design

Memory for real models

Full root access

OpenAI-compatible APIs

Plug into automation

Predictable cost

Search your own documents

Engineers on call

AI hosting questions

Which models can I run?

Do I need a GPU?

How much RAM do I need?

Is my data used to train anyone’s model?

Can you set up the model for me?

Bring your AI in-house