LLM Factory

Sovereign AI academy · United Arab Emirates

Train local.Fine-tune local.Own the model.

LLM Factory is a hands-on academy where engineers learn to fine-tune large language models on hardware that never leaves the country. Then they take the same stack back to their own organisations.

Loading the 3D center…

Concept design · drag to look around

Drag to orbit · scroll to zoom · right-drag to pan

One building, three businesses

A place to learn, a partner to deploy, and a studio for Arabic-first models

01

Academy

Hands-on LLM engineering courses on 34 aiDAPTIV+ workstations. Graduates leave certified and able to fine-tune a model on their own.

02

Enterprise deployment

We supply, install and support the same machines at the client's site, and run fine-tuning projects inside our Secure Core.

03

Own models

Arabic-first models for specific sectors, trained when the machines are idle and licensed with the hardware or through a private API.

The technology

How a desktop PC fine-tunes a 13-billion-parameter model

Full fine-tuning needs far more memory than a consumer GPU has. Phison aiDAPTIV+ adds a dedicated AI SSD that acts as an extension of GPU memory, so the training state that does not fit in 16 GB moves to a 320 GB cache.

GPU memory
16 GB
Needed for a 13B full fine-tune
≈ 200 GB
aiDAPTIVCache SSD
320 GB
1h 08mLlama-3.1-8B fine-tuned on one RTX 6000 Ada GPU
2h 53mLlama-3.1-70B fine-tuned on four GPUs
13BLargest model one RTX 5070 Ti desktop fine-tunes in full

Sources: Signal65 lab report for Phison, July 2025; ABS AI Training PC listing on Newegg. The memory figure is an approximation for full-parameter training with mixed precision.

The fleet

36 machines in three tiers, colour-coded in the building

24Trainer stations
GPU
RTX 5070 Ti 16 GB
CPU
Core Ultra 7 265KF
RAM
32 GB DDR5-6000
SSD
2 TB + AI100E 320 GB

One seat per learner in the Orbit Hall. Fine-tunes models up to 13B.

10Forge Pods
GPU
RTX 5090 32 GB
CPU
Core Ultra 7 265KF
RAM
32 GB DDR5-6000
SSD
2 TB + AI100E 320 GB

Semi-enclosed pods for advanced learners, capstone projects and client work.

2Secure Core servers
GPU
4 × RTX 6000 Ada 48 GB
CPU
Xeon W5-3535X
RAM
512 GB DDR5 ECC
SSD
2 TB aiDAPTIVCache + 4 × 2 TB

ABS Zaurion Pro in an air-gapped glass enclave. Client fine-tuning up to 70B and beyond.

Programs

From a first fine-tune to a certified team

3 days

Foundation bootcamp

From zero to your first fine-tuned 8B model, on your own seat.

AED 4,900
60 hours

Professional certification

Data preparation, full and LoRA fine-tuning, evaluation and on-premise deployment.

AED 14,500
24 seats · 5 days

Corporate cohort

A private cohort that works on your organisation's own data inside the Secure Core.

AED 150,000 per cohort
Semester

University partnership

Credit-bearing LLM engineering courses and overnight research compute.

On request

Indicative prices for the Dubai flagship.

Universities

Labs, courses and research compute for students and faculty.

Government

Sovereign fine-tuning and staff upskilling, with data kept on-premise.

Large companies

Banks, energy, telecom and healthcare teams building private models.

SMEs

The lowest-cost path to a private model: one PC, one course, monthly support.

24 hours

The center works three shifts

Classes by day, client work in the evening, and our own model training overnight. The hardware is never idle for long.

08:00 – 18:00

Academy classes

Bootcamps, certification tracks and university labs across all 34 learner seats.

18:00 – 22:00

Evening cohorts and client jobs

Part-time professionals, plus fine-tuning projects running in the Secure Core.

22:00 – 08:00

Overnight model training

Idle machines train LLM Factory's own models and serve compute passes for alumni and researchers.

Roadmap

From one flagship to six cities across the GCC and the Levant

  1. 2026–27

    Dubai flagship

    The Orbit Hall opens, with training-partner terms agreed with Phison and ABS.

  2. 2028

    Abu Dhabi

    A second site near government and university clusters. First own models released.

  3. 2028–29

    Riyadh

    The largest GCC market, with Arabic models tuned for Saudi use.

  4. 2029–30

    Doha · Kuwait City · Muscat · Amman

    Franchise or joint-venture sites built on the same blueprint and curriculum.

Partners, universities and investors

Build the region's models here

We are raising a seed round for the Dubai flagship and signing founding university and government partners. Get in touch to see the full plan.

Rami Shawwa
Phone · Oman+968 9943 8526
Rami Shaheen
Phone · Jordan+962 777 777 766
Phone · UAE+971 50 219 0444
Emailrami@osdoman.com