Local AI workstations and servers
PCs and servers for running AI models on your own hardware: chat assistants, coding models, document search, image generation and fine-tuning, with your data staying on your premises.
We spec them around the one number that decides which models fit: the graphics card’s memory (VRAM). Every machine is assembled and stress-tested in our Hyderabad workshop and shipped across India.
01
How much VRAM you need
A model has to fit in graphics memory to run fast. Its size is roughly its parameter count times the bytes each weight takes: about half a byte at 4-bit, one byte at 8-bit and two bytes at 16-bit, plus room for the context. Most people run 4-bit or 8-bit versions, which lose little quality.
| Model size | 4-bit | 8-bit | 16-bit |
|---|---|---|---|
| 7B to 8B | 6 GB (8 GB card) | 10 GB (12 GB card) | 18 GB (24 GB card) |
| 12B to 14B | 10 GB (12 to 16 GB card) | 17 GB (24 GB card) | 31 GB (32 GB card or more) |
| 27B to 32B | 21 GB (24 GB card) | 36 GB (48 GB card) | 68 GB (96 GB card) |
| 70B | 42 GB (48 GB card, or two 24 GB cards) | 78 GB (96 GB card) | 150 GB (two or more large cards) |
Long contexts, many users at once and fine-tuning need more. Mixture-of-experts models (a 30B model with 3B active, for example) need memory for all their weights but answer much faster than a dense model of the same size.
02
Our AI-ready builds by graphics memory
Pre-built workstations and servers we sell now, by how much graphics memory they carry. Any of them can be changed in the builder.
16 GB to 23 GB of VRAM
Runs 12B to 14B models at 4-bit, 8B models at 8-bit. 30 builds, from ₹2,24,537 incl. GST.
24 GB to 31 GB of VRAM
Runs 27B to 32B models at 4-bit, 14B models at 8-bit. 9 builds, from ₹3,64,172 incl. GST.
32 GB to 47 GB of VRAM
Runs 32B models at 4-bit with long context, 14B models at 16-bit. 5 builds, from ₹4,23,059 incl. GST.
- Kuro DaVinci Workstation · Flagship: Ryzen 7 5700X · Radeon AI PRO R9700 32GB · 64GB · 1TB · ₹4,23,059
- Kuro Turing Workstation · Performance: Ryzen Threadripper 9970X · Dual RTX 5070 Ti 16GB · 64GB · 2TB · ₹11,10,816
- Kuro DaVinci Workstation · Flagship: Ryzen 9 9950X · RTX 5090 32GB · 64GB · 1TB · ₹12,07,328
48 GB to 95 GB of VRAM
Runs 70B models at 4-bit, 32B models at 8-bit. 2 builds, from ₹10,21,892 incl. GST.
03
What goes into an AI machine
Graphics card
The VRAM decides the models; NVIDIA GeForce RTX, RTX PRO and Radeon AI PRO cards from 16 GB to 96 GB.
System memory
At least as much RAM as VRAM, ideally twice as much, so models load and spill over without stalling: 64 GB to 128 GB is common.
Storage
Fast NVMe SSDs: model files run from a few GB to well over 100 GB each.
Power and cooling
A model keeps the card at full load for hours. Power supply and airflow are sized for that, and every build is stress-tested before it ships.
Software
Tell us the tools you plan to use (Ollama, llama.cpp, vLLM, ComfyUI and so on) and the operating system, and we will spec and quote the machine for them.
04
Questions people ask
- Can I run ChatGPT-like models without the cloud?
- Yes. Open-weight models such as Llama, Qwen, Mistral, Gemma and DeepSeek run on your own PC or server, so prompts and documents never leave your premises. How capable the model is depends on its size, which the graphics memory limits.
- NVIDIA or AMD for local AI?
- NVIDIA cards have the widest software support (CUDA), so most tools work out of the box. AMD Radeon AI PRO cards give more memory for the money and work well for running models with tools like llama.cpp; check that the software you need supports them.
- One large graphics card or two smaller ones?
- One card with enough memory is the simplest and fastest. Two cards can split a model between them, which lets you run larger models, but needs a bigger power supply, a board with room for both and more cooling.
- Is a workstation or a server better?
- A workstation sits under a desk and suits one person. A server suits a team sharing models over the office network, and can take more cards, more memory and redundant storage.
- Do you ship outside Hyderabad?
- Yes. Every machine is built and tested in our Hyderabad workshop and shipped across India.