Get Started
Platform/Compute/GPU Cloud Servers

GPU power for AI, in-country.

Data-centre-class GPUs for training, inference and rendering — provisioned in hours, priced flat, and sovereign like everything else on the platform.

Data-centre GPUsHours to provisionSovereign AI
Talk to usAll Compute
Capabilities

From experiment to production inference

Training & inference
Single-GPU experiments to multi-GPU training nodes and always-on inference.
One platform, both phases
AI-ready images
CUDA, PyTorch and TensorFlow images maintained and patched.
CUDA current, always
Fast storage attachedNVMe Performance Storage feeds data-hungry jobs without bottlenecks.Data keeps up with compute
Kubernetes GPU poolsAttach GPU node pools to Managed Kubernetes for scheduled workloads.Schedule, don’t babysit
Whole-card allocationTraining tiers give you entire accelerators — no noisy neighbour steals throughput.Your card, your throughput
Partitioned inferenceSlice cards for inference so small models don’t pay for big hardware.Right-size the serving tier
Sovereign AIModels and training data stay in-jurisdiction on sovereign capacity.Compliance for AI too
GPU telemetryUtilisation, memory and thermals per card, streamed to dashboards.Idle GPUs get noticed
Hours to provisionGPU capacity provisions in hours, not procurement quarters.Experiments start this week
Snapshot the labSnapshot training environments and clone them per researcher.Reproducible experiments
Compare the tiers

Inference, training, or a cluster

Coming soon: same certified images, same sovereign placement — the difference is how much accelerator you hold and how it is wired together. Register interest and we will confirm your placement at launch.

ProductionEdgeComing soon
TIER 01
Inference
Serving, not training

A single accelerator per instance for production inference, embeddings and vision endpoints — sized for latency and steady throughput rather than batch power.

Best for
Inference endpointsEmbeddings & searchVision & OCR
Where it runs
Dubai AZ1RiyadhBahrain
Plans
Inference1 × 48 GB GPU · 16 vCPU · 128 GB RAMRegister →Inference Plus2 × 48 GB GPU · 32 vCPU · 256 GB RAMRegister →CustomAny accelerator and instance shapeRegister →
TIER 02
Training
Whole cards, reserved

Whole 80 GB accelerators with NVMe scratch and performance storage behind them, for fine-tuning and training runs that must finish in-country.

Best for
Fine-tuningModel trainingSimulation
Where it runs
Dubai AZ1RiyadhBahrain
Plans
Train2 × 80 GB GPU · 32 vCPU · 512 GB RAMRegister →Train Plus4 × 80 GB GPU · 64 vCPU · 1 TB RAMRegister →CustomAny card count and scratch capacityRegister →
TIER 03
Cluster
Multi-node interconnect

Several GPU nodes joined by a low-latency fabric for distributed training, schedulable through Managed Kubernetes when your team prefers it.

Best for
Distributed trainingHPC & researchRender farms
Where it runs
Dubai AZ1RiyadhLagosNairobi
Plans
Cluster4 nodes · 16 accelerators · RDMA fabricRegister →Cluster Plus8 nodes · 32 accelerators · dedicated fabricRegister →CustomAny node count, fabric and storage tierRegister →
Two ways to buy

Reserved cards, or a project window

Both are flat-fee. The difference is whether the accelerators are yours all year or for a programme.

Reserved AcceleratorsNamed cards
Specific accelerators held for your estate, available the moment your team needs them — no queue, no spot market.
✓Cards reserved, not competed for
✓Flat monthly fee per accelerator
✓Images and drivers maintained for you
Best when GPU work is continuous.
Sign in to explore plans
Campaign CapacityProject window
A block of accelerators for a defined programme — stood up for the window, returned when the programme closes.
✓Sized for the programme, not the year
✓Same certified images and fast storage
✓Extend or hand back at the review point
Best for funded AI pilots and bursts.
Sign in to explore plans
Who decides, who runs it

What each group gets from GPU

Four groups shape a cloud decision. Pick the one you belong to — the platform reads differently from each seat.

“Sovereign AI without a build programme”

Accelerated compute arrives in hours instead of a procurement cycle, and the data it trains on never leaves your jurisdiction.

GovernedSovereignPredictable

FAQs

Which accelerators do you run?
Data-centre-class 48 GB and 80 GB accelerators. We confirm the exact model available in your chosen zone during sizing.
Can we share a card between workloads?
Yes — inference tiers can be partitioned, while training tiers give you whole cards so a neighbour never steals your throughput.
How quickly can capacity be live?
Hours, not weeks, where the cards are in stock in your zone. Cluster tiers with dedicated fabric take longer to wire.
Does our training data leave the country?
No. Storage, compute and networking all sit inside the zone you choose, under your jurisdiction.
Can Kubernetes schedule the GPUs?
Yes. GPU node pools attach to Managed Kubernetes so jobs queue and schedule alongside your app workloads.
How is the flat fee different from public cloud pricing?
Everything inside the plan you opt for is included — no additional charge for the components that make it up. Public clouds meter each component on consumption, so the bill moves with usage. Here the plan is reserved for you and the figure you agreed is the figure you pay.
Can we drive it from Terraform and CI?
Yes. Every operation is exposed through the REST API, CLI and a maintained Terraform provider, with scoped API keys per team. If the console can do it, your pipeline can do it.
How do limits and quotas work?
Quotas are yours to shape — carve capacity per team or project and delegate inside it. Hard ceilings are a conversation, not a wall; the estate re-baselines at thresholds you approve.
Can security services be included in my plan?
Yes. Endpoint security, firewalls and WAF are scoped into the same flat fee when you add them — one figure, no separate security invoice.
Better together

The start of a solution

GPU Cloud Servers rarely works alone. Pair it with the platform around it — the outcome lives in the combination.

Store on itBlock StorageSSD and NVMe volumes that expand live under the workload.Explore storage →Run data on itManaged DatabasesPostgreSQL, MySQL, MS SQL, HANA and Oracle, managed.Explore DBaaS →Orchestrate itManaged KubernetesConformant clusters — the control plane is our pager.Explore Kubernetes →Work from itWorkspaceApplite streamed apps or full VDI desktops.Explore Workspace →Shield itFirewallsSegmentation as code — every path explicit.Explore firewalls →Guard the appsManaged WAFSOC-tuned OWASP rules at Layer 7.Explore WAF →Watch itObersva MonitoringTelemetry and alerts before users notice.Explore monitoring →Evidence itOditrLive controls mapped to frameworks — evidence on demand.Explore Oditr →
Noura
Noura — AI Cloud Experience Guide

Two ways Noura gets you moving

Choose the way that fits you best — Noura makes every step simple.

Not sure where to start?
Start the conversation

Describe the workload in plain language — Noura works out the tiers, sizes and protection with you.

Plain language input
Smart sizing & protection
No commitment. Just clarity.
Talk to Noura→
Want to see it working?
Experience it in the marketplace

Explore CloudSouq experiences built on these services — see GPU Cloud Servers working inside real solutions before you commit to anything.

Try real solutions
See it work in action
Commit with confidence
Explore CloudSouq→
Noura is your guide at every step—from first question to final success.Intelligent
GuidanceSecure &
TrustedHuman
Simplicity