01
H100 or L40S
Choose the GPU family that matches the workload: H100 for large models, L40S for inference and visual pipelines.
Platform · GPU Cloud
GPU cloud, on Malaysian soil
On-demand NVIDIA H100 and L40S instances, billed per second and hibernated when the work stops.
REF · SC-2026-GPUC
Mode · On demand Billing · Per second GPU · H100 / L40S Residency · Malaysia
Platform · GPU Cloud
Elastic capacity without an elastic jurisdiction
A public GPU cloud is convenient until the control plane, the logs or the support path crosses a border. Bijak Cloud keeps the compute, storage and audit trail in-country while preserving the on-demand workflow engineering teams expect.
Start a workload without a capacity negotiation. Provision a dedicated GPU instance, attach sovereign storage, and keep every byte of the data plane inside Malaysia.
- Dedicated GPU instances — no oversubscription of the GPU you requested.
- Per-second billing with instant hibernation when a workload goes idle.
- S3-compatible object storage and persistent volumes with encryption at rest.
- A jurisdiction-pinned data plane designed for PDPA 2010 obligations.
Capabilities
What ships with every instance
The baseline needed to run a real workload, not just a benchmark script.
02
Per-second billing
Metering starts when the instance runs and stops when it hibernates. Storage is billed separately and itemised.
03
Instant hibernation
Preserve instance state without paying for idle GPU time, then resume when the next job is ready.
04
SSH and API access
Provision through the console, CLI or API. Connect over SSH or call the inference API from your own services.
05
Persistent storage
Keep datasets and checkpoints on encrypted volumes that survive an instance lifecycle.
06
Audit logging
Provisioning, access and deletion events are written to immutable logs you can export.
Hardware
Available GPU shapes
Public specifications for the two GPU families. Exact capacity and shape availability are confirmed on your quote.
NVIDIA H100
SXM · 80GB- Memory
- 80GB HBM3
- Memory bandwidth
- 3.35 TB/s
- Interconnect
- NVLink / NVSwitch
- Cloud shapes
- 1× · 2× · 4× · 8× GPU
Large-model fine-tuning, training, high-throughput inference and HPC.
NVIDIA L40S
PCIe · 48GB- Memory
- 48GB GDDR6 ECC
- Memory bandwidth
- 864 GB/s
- Interconnect
- PCIe Gen4
- Cloud shapes
- 1× · 2× · 4× · 8× GPU
Inference, RAG, document AI, rendering, simulation and video workloads.
Availability is confirmed per order.
Process
From zero to a running instance
The operational path is short; the capacity commitment is confirmed before provisioning begins.
- 01
Choose
Select a GPU family, shape and storage profile, then submit the workload details for capacity confirmation.
- 02
Provision
The instance is placed in-country, attached to your private network and connected to the storage you designate.
- 03
Connect
Use SSH, the CLI or the OpenAI-compatible inference endpoint from your application.
- 04
Scale or stop
Add instances for a burst, hibernate when idle, and review usage and audit logs in the console.
Use cases
What teams run on GPU Cloud
Representative workloads; each can be validated in a bounded pilot before a longer commitment.
Model fine-tuning
Adapt an open model to a Malaysian corpus without exporting the training data.
Inference bursts
Absorb demand peaks with on-demand capacity and hibernate when the queue clears.
Document pipelines
Run OCR, extraction and embedding jobs next to the documents they process.
Experiments and teaching
Give research and training teams temporary GPU capacity with per-second cost control.
Controls
Sovereignty controls
The same control baseline used across the platform, applied to on-demand capacity.
In-country data plane
Compute, storage and audit data are designed to remain in Malaysian jurisdiction, and support is delivered from Malaysia.
Workspace keys
Storage encryption keys are managed per workspace with rotation support.
Immutable audit logs
Every provisioning and access event is recorded and exportable for review.
PDPA 2010 alignment
Retention and access controls are mapped to Akta 709 obligations.
FAQ
GPU Cloud questions
How much does a GPU instance cost?
Pricing is quote-based because the rate depends on the GPU shape, commitment length, storage profile and support model. We do not publish a list price that would not survive the first conversation. Request a quote with your workload and we will itemise it.
Is the GPU shared with other customers?
No. The instance you request is dedicated to your workspace and is not oversubscribed. Multiple tenants may share the same facility, but not the GPU you are billed for.
What happens when I hibernate an instance?
GPU billing stops. The instance state is preserved for a fast resume, and attached storage remains billable. The console shows both compute and storage lines separately.
Can I move an instance to another region?
The platform is designed around Malaysian jurisdiction, and support is delivered from Malaysia. Cross-border placement or replication is not offered by default and would require explicit written authorisation.
Platform · GPU Cloud
Request GPU capacity
Tell us the model, dataset and expected run pattern. We will confirm the shape, availability and an itemised quote.
Quotes are itemised by configuration, term and support. No fixed list price is published.