New · For Saudi SMEs
SoverAIn One.
Enterprise AI in one box.
A sovereign AI system for businesses that cannot send their data abroad. Large open models, your keys and full governance, in a single unit in your own server room.
ذكاء اصطناعي مؤسسي في جهاز واحد
From
SAR 375,000
about US$100,000 · three years included
- 384 GB
- GPU memory across four Blackwell GPUs
- 235B
- parameters: the largest open models it is sized for
- 50 to 500
- staff it is sized to serve, by intensity of use
- 0 bytes
- that need to leave your building
Inside SoverAIn One
Every layer built for data that stays home.
- 01
Tamper-evident chassis
A sealed, locking lid. Opening it is detected, logged and sent to your administrator.
- 02
Four Blackwell GPUs
- 03
Secure mainboard
- 04
Encrypted storage, redundant power
- 05
Cooled for Saudi summers
Illustrative cutaway. Final layout follows the agreed configuration.
Industrial design
Infrastructure that looks like it belongs in the Kingdom.
Mashrabiya-inspired lattice
The bezel perforation borrows the eight-point star of the traditional mashrabiya screen, with its open area tuned for airflow.
Status at a glance
An emerald light bar shows health: amber on an alert, red if the chassis is opened. A front display shows temperature, GPU health and security state without logging in.
Locked, but serviceable
A keyed bezel protects a tool-less filter door, so filters can be swapped in dusty seasons without opening the sealed chassis.
Built like infrastructure
Bead-blasted anodised aluminium bezel over a steel 4U chassis, on sliding rails, with front handles and rear hot-swap power.
Design concept. Final finish, dimensions and markings are confirmed at order.
Cooling
Four Blackwell GPUs make real heat. Saudi summers make more.
- ~3.5 kW
- heat at full load: about 12,000 BTU/h for your room to remove
- 40°C
- design inlet temperature, for server rooms that run hot in summer
- 0 litres
- of facility water needed with the recommended option
Filtered air
High-static-pressure fans push filtered air front to back over the GPUs and CPUs.
For
- Simplest to service
- No liquid in the chassis
- Lowest upfront cost
Against
- Fans run hard in a warm room
- Loudest option
- Most sensitive to room temperature
Best fit: A well-cooled server room that stays below about 27°C all year.
Hybrid closed-loop liquid
Sealed liquid loops with cold plates on the GPUs and CPUs carry heat to a radiator at the rear; filtered air cools the rest. No connection to building water.
For
- Holds GPU clocks in a hot room
- Fans run slower and quieter
- No plumbing or facility water
- Factory-filled and pressure-tested
Against
- Higher upfront cost than air
- Loop serviced by SoverAIn, not on site
Best fit: Most Saudi SMEs: small server rooms, hot summers, dust, no chilled water.
Direct-to-chip, facility water
Cold plates connected through a coolant distribution unit to your building’s chilled-water loop.
For
- Removes the most heat
- Lowest room air-conditioning load
- Scales to many units and to SoverAIn Rack
Against
- Needs facility water and a CDU
- Plumbing work and leak management
- Overkill for a single unit
Best fit: Data centres and larger estates, or several units in one room.
Why the hybrid loop wins in Saudi Arabia
The question is not how cold a lab can be kept, but what happens in a real SME server room in August, during a dust storm, with the air conditioning already stretched.
Heat
Summer outdoor temperatures pass 45°C, and small server rooms drift well above their set point when the air conditioning is stretched.
Dust
Shamal winds and sandstorms clog filters and coat heatsinks; air-only designs lose cooling as filters load.
Water
Many SME buildings have no chilled-water loop, and evaporative cooling uses water the region does not have to spare.
Power
Every watt of heat the room must remove adds to the air-conditioning bill; liquid loops cut fan power and let the room run warmer.
Heat output and inlet temperature are design targets for the indicative configuration. We survey your room before delivery and confirm the cooling option in writing.
Runs serious models
Big enough for frontier-class open models. Small enough for one rack slot.
| Model | Role | What it is good for |
|---|---|---|
| Qwen3 235B (mixture of experts) | Flagship | Strong bilingual reasoning, drafting and analysis; quantised to fit |
| Llama 3.3 70B | Workhorse | Fast, dependable general assistant and document work |
| gpt-oss 120B | Reasoning | Open-weights reasoning model for analysis and agents |
| ALLaM 7B | Arabic-first | Arabic drafting, correspondence and customer replies at high speed |
| Embedding and reranking models | Search | Arabic and English retrieval over your documents |
Examples of open models the unit is sized for. Availability depends on each model's licence and on evaluation against your own Arabic and English content. Large models run quantised where needed.
Sovereignty and governance, built in
Controls an auditor can check, not promises in a brochure.
Your keys, in the box
A built-in HSM generates and holds encryption keys. They never leave the unit, not even to us.
Tamper-evident
Locking lid with intrusion detection; any opening is logged and alerted.
Secure boot, signed updates
Only signed firmware and software run. Updates arrive by signed media or an opt-in channel you control.
Never phones home
No telemetry by default. The unit works fully air-gapped.
Policy packs
Ready-made controls mapped to PDPL and NCA ECC: PII redaction, retention, data-classification labels.
Who can do what
Single sign-on with your directory, role-based access to models, data sources and workspaces.
Every answer on the record
An append-only log of every prompt, model, source and user, exportable to your auditors or SIEM.
People approve actions
Agents draft and prepare; a named person approves anything that leaves the system.
Made for SMEs
The data of a large enterprise, the budget of a growing one.
Law firms
Contract review and research on client files that cannot leave the office.
Clinic groups
Clinical summaries and patient letters with records kept on site.
Finance and family offices
Credit memos, KYC documents and reporting, privately.
Engineering and contracting
Search decades of drawings, specifications and tenders.
Schools and training providers
Arabic and English assistants for staff and students.
Government suppliers
Work on restricted material under your client’s data rules.
Specifications
What is in the box
| Component | Specification |
|---|---|
| GPUs | 4 × NVIDIA RTX PRO 6000 Blackwell Server Edition, 96 GB each (384 GB total) |
| CPUs | Dual-socket server-class x86 processors |
| System memory | 1 TB DDR5 ECC |
| Storage | 30 TB NVMe, self-encrypting, plus a mirrored boot pair |
| Security | Built-in HSM, TPM 2.0, secure boot, tamper-evident locking chassis |
| Networking | Dual 25/100 GbE, separate out-of-band management port |
| Power | Redundant hot-swappable power supplies; site power check included |
| Form factor | 4U rack-mount with rails; choice of three finishes |
| Cooling | Hybrid closed-loop liquid on GPUs and CPUs with filtered air (recommended); air or facility-water options |
| Software | SoverAIn stack: models, gateway, guardrails, retrieval, workspaces, audit |
| Support | Three years, from Saudi Arabia, with on-site hardware replacement |
Price
From
SAR 375,000
about US$100,000
- Hardware. The SoverAIn One unit, configured and burned in before delivery
- Software. Three years of the SoverAIn stack, model updates and policy packs
- Installation. Delivery, racking, network and identity integration on your site
- Onboarding. Admin training and a first use case live with your team
- Support. Three years of support from Saudi Arabia, including on-site hardware replacement
Indicative, before VAT. Includes the hardware, three years of software and support, delivery and installation in the Kingdom. Final price depends on configuration and currency at order.
Reserve a unitProduct photos are illustrative NVIDIA hardware, not SoverAIn units. NVIDIA, DGX, HGX, Grace and Blackwell are trademarks of NVIDIA Corporation; their use here does not imply endorsement. Configurations are indicative and agreed per order.
Private AI for your business, delivered to your server room.
Tell us your team size and the documents you want AI to work on. We will confirm the configuration and a delivery date.