Hardware
AI servers that arrive ready to run.
خوادم ذكاء اصطناعي جاهزة للتشغيل، داخل المملكة
NVIDIA DGX Spark (GB10 Grace Blackwell)
Design partners wantedSoverAIn Node
A sovereign AI lab on a desk.
A compact AI computer with the SoverAIn stack preinstalled, for teams that want to prototype, evaluate models and run private assistants without a data centre.
- Compute
- NVIDIA GB10 Grace Blackwell superchip
- Memory
- 128 GB unified memory
- Form factor
- Desktop, standard power
- Connectivity
- Connected, or fully air-gapped
- Best for
- Pilots, model evaluation, small teams, classified labs
NVIDIA HGX B200, 8-GPU
Design partners wantedSoverAIn Server
Eight GPUs for production inference and fine-tuning.
A rack-mounted GPU server that runs your production models, retrieval and agents for a department or a whole organisation, with the SoverAIn stack and your keys from day one.
- Compute
- NVIDIA HGX B200: eight Blackwell GPUs with NVLink
- Form factor
- Rack-mounted server for your data centre or ours
- Cooling
- Air-cooled
- Scale
- One server to a cluster, added a server at a time
- Best for
- Production assistants, document pipelines, fine-tuning
NVIDIA GB200 NVL72
On the roadmapSoverAIn Rack
Rack-scale AI for national and enterprise workloads.
A liquid-cooled rack that acts as one large GPU, for the biggest models, long contexts and many concurrent users, delivered and operated in the Kingdom.
- Compute
- 72 NVIDIA Blackwell GPUs and 36 Grace CPUs in one NVLink domain
- Form factor
- Full rack, delivered integrated
- Cooling
- Liquid-cooled
- Facility
- Requires liquid-cooling-ready space; we assess your site
- Best for
- Large models, national-scale services, shared AI platforms
Compare
Pick the size, keep the software.
| SoverAIn Node | SoverAIn Server | SoverAIn Rack | |
|---|---|---|---|
| Platform | NVIDIA DGX Spark (GB10 Grace Blackwell) | NVIDIA HGX B200, 8-GPU | NVIDIA GB200 NVL72 |
| Compute | NVIDIA GB10 Grace Blackwell superchip | NVIDIA HGX B200: eight Blackwell GPUs with NVLink | 72 NVIDIA Blackwell GPUs and 36 Grace CPUs in one NVLink domain |
| Form factor | Desktop, standard power | Rack-mounted server for your data centre or ours | Full rack, delivered integrated |
| Best for | Teams starting out, labs and air-gapped offices. | Production workloads for a department or an organisation. | Ministries, large enterprises and shared national platforms. |
In the box
The hardware is half the product.
Models
Arabic-first and open-weights models, loaded and evaluated on your hardware
Inference gateway
An OpenAI-compatible API, quotas and usage metering
Guardrails
PII redaction, policy filters and prompt-injection defence
Retrieval
Connectors, Arabic and English search, citations
Workspaces
A secure assistant for staff, ready on day one
Audit
Every call logged, exportable to your SIEM
From order to first answer
Sized, built, installed and run in the Kingdom.
01
Size
We size the hardware to your models, users and data classification, and say what it will not do.
02
Build
Hardware configured and the SoverAIn stack installed and burned in before delivery.
03
Install
Delivered, racked and commissioned in your facility or an in-Kingdom data centre.
04
Operate
Run by your team or ours, with updates, model upgrades and support from Saudi Arabia.
Product 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.
Tell us your models and your users. We will size the hardware.
A short conversation about workloads, data classification and your facility is enough for an indicative configuration.