Supplied and installed on your premises
We supply the hardware, install it in your server room or data centre, and bring up the model and the assistant on it. The hardware belongs to you.
AI servers and infrastructure for local models
We supply and install the hardware, run the open-weight model that suits your work on it, and connect a Persian assistant to your systems: from one Mac mini for a small team to racks of multi-GPU servers.
What AI infrastructure means here
A language model on your own servers needs three things: hardware with enough memory for the model, software that serves it to several people at once, and an application your staff actually use. ZharfAI delivers all three together. Data stays inside your network and nothing depends on a foreign service.
Three ways to work with us
We do not rent GPUs by the hour. The hardware is either yours, or a machine that works only for you.
A dedicated machine we maintain
Hardware that serves only you, with maintenance, updates and monitoring handled by us; your team just uses the system.
Set up on hardware you own
If you already have servers, graphics cards or Macs, we choose and set up a model that runs well on them, and if they fall short we say exactly what to add.
Hardware and models we provide
Each hardware family has its own page: configurations, memory, which models run on it and what it suits.
Mac Studio
Mac Studio and Mac mini with large unified memory: one machine, or a cluster of Macs for very large models.
ViewNVIDIA
From DGX Spark and RTX cards on a desk to eight-GPU servers and racks, with the widest software support.
ViewAMD
High-memory Instinct cards, Radeon cards for workstations and small Ryzen AI Max computers.
ViewChinese hardware
Huawei Ascend and other Chinese AI accelerators, and high-memory Chinese mini PCs.
ViewLocal models
Chinese, American and European open models that run on your hardware, from 2 billion to several trillion parameters.
View
Which model on which hardware
A quick guide by model size. We make the exact choice by testing on your own work, because the number of users and the length of texts take memory too.
| Model size | Examples | Weights at 4-bit | Suitable hardware |
|---|---|---|---|
| Up to 15B parameters | Gemma 4 12B, Qwen3.5-9B | Up to about 10 GB | A Mac mini, one RTX card or an Atlas 300I Pro |
| 20B to 40B | Gemma 4 31B, Qwen3.8-27B | About 15 to 25 GB | One RTX 5090, a Mac mini M5 Pro or a Radeon AI PRO R9700 |
| 70B to 130B, often MoE | gpt-oss-120b, Qwen3.5-122B-A10B | About 45 to 80 GB | RTX PRO 6000, DGX Spark, Ryzen AI Max or a Mac Studio M5 Max |
| 200B to 400B, MoE | DeepSeek-V4-Flash, Tencent Hy3, Qwen3.5-397B | About 120 to 250 GB | A Mac Studio M5 Ultra, a four-GPU H200 server or an Atlas 800I A2 |
| 500B to 1T | GLM-5.2, DeepSeek-V4.1-Flash, Kimi K2.7 Code | About 300 to 620 GB | An eight-GPU H200, B200 or MI355X server, or a 512 GB Mac Studio |
| Over 1T | DeepSeek-V4-Pro, Kimi K3, Qwen3.8-2.4T | About 1 to 1.7 TB | An eight-GPU B300 or MI355X server, a Mac Studio cluster or Atlas 950 |
Memory covers the model weights only and is our estimate; long context and concurrent users need more. MoE models use only part of their parameters for each word, so they answer faster than their size suggests.
From the first call to a working system
Hardware is chosen after a model has been measured on your own work, not from a table. A quote is issued only once it is clear which model has to serve how many people, and how fast.
- 01
First conversation
We go through the task, the number of users, the data, where the hardware goes and what you already own.
- 02
Measured on your work
We test several open-weight models on samples of your real work and size memory and speed from those results.
- 03
Written quote
You get the exact configuration, the proposed model, the delivery time and the support terms in writing. Prices appear only in the quote.
- 04
Installation
We install the hardware, bring up the model and connect the assistant to your systems, then test it once with international internet cut off.
- 05
Handover and support
You get documentation and training for your team; afterwards we maintain the system and update the models under contract.
How to get in touch
Fill in the form or message us on WhatsApp. The clearer these are, the more exact the quote:
- What the model has to do (answer from documents, code, summarise, process images or speech)
- How many people use it at the same time, and what response time is acceptable
- The model or hardware family you have in mind, if you already know
- Where it goes: your server room, a data centre, or hardware you already own
- How sensitive the data is, and whether it must run with no internet at all
We reply within one business day.
Zharf
Try the ZharfAI assistant right now
Ask about Persian assistants, document automation, or how to start a pilot.
Frequently asked questions about AI servers
Which hardware should we start with?
Start from the work, not the hardware. We usually suggest starting with one machine, such as a Mac Studio, one RTX PRO 6000 card or a DGX Spark, and a mid-size model, then sizing hardware for the whole organisation after a pilot on real work.
Which models do you set up?
Any model whose weights are published, from Gemma and Qwen to DeepSeek, Kimi, GLM and MiMo. The full list, with size, commercial licence, Persian test score and suitable hardware for each, is on the local models page.
Do you also supply Chinese hardware such as Huawei Ascend?
Yes. We supply and set up Huawei Ascend cards and servers, Moore Threads, Cambricon, MetaX and Hygon accelerators, and high-memory Chinese mini PCs. Because the software for this hardware is less mature than NVIDIA's, we test your model on it before you buy.
Do you rent GPU servers by the hour or month?
No. ZharfAI does not rent GPUs by the hour or month. There are three ways to work with us: we supply the hardware and install it on your premises, we run a dedicated machine for you and maintain it ourselves, or we set the model up on hardware you already own. If you only need a few hours of compute for a test, say so in the first conversation and we will suggest a suitable route.
How much does it cost?
We do not publish prices, because they depend on the configuration, quantity, the day's market and how it is delivered. After the first conversation and a test on your work you receive a written quote that also names the exact hardware model, the delivery time and the support terms.
Does it work without international internet?
Yes, if the model, the serving software, the assistant, user sign-in and monitoring all sit inside your network. Anything that depends on a foreign service stops during a cut, so before handover we test the system with international internet deliberately cut off.
Start with the work that has to stay on your own servers
Tell us what the model has to do and how many people will use it. In the first conversation we say what hardware it needs, and if what you already have is enough, we say so plainly.