ST [Brand]

GPU Cloud Infrastructure

GPU Cloud Infrastructure

NVIDIA GB200 NVL72 and HGX H100 capacity

Purpose-built GPU infrastructure for AI training, inference and rendering, managed end to end on a Kubernetes-native platform. ST [Brand] operates from Malaysia and provides NVIDIA GB200 NVL72 and NVIDIA HGX H100 capacity, GPU virtual servers, managed Kubernetes, software-defined storage and private networking.

Every deployment is quoted against the workload you send us — on-demand GPU hours, reserved capacity, or a dedicated enterprise cluster. Email sales@stratustech.tech or use the form on this page to start that conversation.

What is GPU cloud infrastructure?

GPU cloud infrastructure is a way to run training, inference and rendering on hosted NVIDIA GPUs, instead of building and staffing a full facility for every project. The operator provides the racks, power, cooling, network fabric and the platform that schedules work. Your team keeps control of the software stack, data and job design.

ST [Brand] offers this model in the Asia-Pacific region, with a Kubernetes-native control plane and, where a VM is a better fit, GPU virtual servers based on KubeVirt with PCI pass-through.

Dedicated and customized GPU capacity

Capacity is not limited to a single instance size. You can start with on-demand NVIDIA HGX H100 billed per GPU-hour, reserve capacity for one to twelve months, or ask us to size an enterprise cluster that includes NVIDIA GB200 NVL72 rack-scale systems. GPU virtual servers are available from $0.24/hr with dedicated NVIDIA GPUs via PCI pass-through, on Linux or Windows.

Networking for the HGX H100 offering is described as NVIDIA Quantum-2 InfiniBand. Storage is Ceph-based and triple-replicated, from 1 GB to petabyte scale. Final rates are confirmed in a quote; published figures are indicative starting rates in USD and exclude taxes.

Common service models

On-Demand

From $3.49 /GPU-hr for NVIDIA HGX H100, billed per GPU-hour with no long-term commitment. GPU virtual servers from $0.24/hr.

Reserved

Commit to capacity for 1 to 12 months. Same infrastructure and platform as on-demand; discounted reserved rates are quoted on request.

Enterprise Cluster

NVIDIA GB200 NVL72 rack-scale systems and large NVIDIA HGX H100 training clusters. Starting rate for GB200 NVL72: Custom / Contact sales.

Placeholder — do not publish as a product Cabinet, cage or suite colocation is not documented on the current site. This section will stay hidden from sales use until it is confirmed.

Starting rates and NVIDIA GB200 specifications

Indicative starting rates (USD)
ProductStarting rate
NVIDIA GB200 NVL72Custom / Contact sales
NVIDIA HGX H100From $3.49 /GPU-hr
GPU Virtual ServersFrom $0.24 /hr
CPU Virtual ServersContact sales (no published figure on the current site)

Prices are indicative starting rates in USD, exclude taxes, and are subject to availability; final pricing is confirmed in a quote.

NVIDIA GB200 NVL72 and GB200 Grace Blackwell Superchip (public NVIDIA specifications, as published on the current product page)
GB200 NVL72GB200 Grace Blackwell Superchip
Configuration36 Grace CPU : 72 Blackwell GPUs1 Grace CPU : 2 Blackwell GPU
FP4 Tensor Core 21,440 PFLOPS40 PFLOPS
FP8/FP6 Tensor Core 2720 PFLOPS20 PFLOPS
INT8 Tensor Core 2720 POPS20 POPS
FP16/BF16 Tensor Core 2360 PFLOPS10 PFLOPS
TF32 Tensor Core 2180 PFLOPS5 PFLOPS
FP326,480 TFLOPS180 TFLOPS
FP643,240 TFLOPS90 TFLOPS
FP64 Tensor Core3,240 TFLOPS90 TFLOPS
GPU Memory | BandwidthUp to 13.5 TB HBM3e | 576 TB/sUp to 384 GB HBM3e | 16 TB/s
NVLink Bandwidth130 TB/s3.6 TB/s
CPU Core Count2,592 Arm Neoverse V2 cores72 Arm Neoverse V2 cores
LPDDR5X Memory | BandwidthUp to 17 TB LPDDR5X | Up to 18.4 TB/sUp to 480 GB LPDDR5X | Up to 512 GB/s

NVIDIA official relative-performance figures (including 30X / 4X / 25X / 18X versus H100 or CPU, with their footnotes) are kept on the GB200 product page and are not repeated as a homepage statistic bar.

Key features

AI-ready GPU platforms

NVIDIA GB200 NVL72 (72 Blackwell GPUs in the NVL72 rack-scale configuration) and NVIDIA HGX H100 are offered today. Public NVIDIA performance figures, when used, stay on the product pages with their footnotes.

Kubernetes-native platform

Managed Kubernetes is the orchestration layer; GPU virtual servers run on KubeVirt when a VM is required.

Direct liquid cooling

The Direct Liquid Cooling page describes a CS1200 CDU used with GB200 deployments, plus a consulting / design / production / construction / maintenance sequence.

Facility connectivity

The data-center page notes land-cable and submarine-cable connectivity for DC1, and for DC2 records generator configuration N+1, a 2N / N+1 / three-layer steel-structure description, IPM 3.0 and magnetic-levitation phase-change cooling.

Software-defined storage and networking

Ceph, triple replication, VPC, and the networking features already documented on those product pages.

Placeholder — certifications and unconfirmed services ISO, PCI DSS, Uptime Institute Tier ratings, 24×7 on-site security patrol, carrier-neutral meet-me-room, and data-center relocation are not claimed here. They will be listed only after ST [Brand] confirms each one. DC2’s city or country will replace the current “East South Asia” wording once confirmed. CDU efficiency numbers stay off this page until engineering confirms them.

Key benefits

Flexibility in scaling up

Move between on-demand hours, reserved terms and an enterprise cluster as the workload changes, without a claim of unlimited scale.

Infrastructure and connectivity

Use the facility facts already on the data-center pages, plus Quantum-2 InfiniBand where it is already stated for HGX H100. No invented backbone maps.

A high degree of control

PCI pass-through GPUs, a managed Kubernetes control plane, and your own images and job scheduling. We do not claim we operate your model or your data.

Security and support

Sales and technical follow-up via sales@stratustech.tech and +60 3-7859 6283. Formal SLA or 24×7 NOC language, and “Enterprise-Grade Security and Compliance,” will be added only when a certificate or policy page exists.

Why choose ST [Brand]

ST [Brand] is based in Malaysia. The office address published today is A-G-3, Block A, Jalan PJU 1A/3J, Taipan 1, Ara Damansara, 47301 Petaling Jaya, Selangor Darul Ehsan. The public product set is NVIDIA GB200 NVL72, NVIDIA HGX H100, GPU virtual servers, managed Kubernetes, storage and networking, with solution pages for machine learning, inference, VFX rendering and pixel streaming.

We will not describe the company as “a leading provider,” and we will not list customer logos or “Trusted by…” lines until a real roster and permission exist. Partner logos stay off this homepage until the partnership type and artwork rights are confirmed.

FAQ

What service models are available?

On-demand GPU hours, reserved capacity for 1 to 12 months, and enterprise clusters. Indicative starting rates are published on the Pricing page; GB200 NVL72 is quoted as custom.

How do we start a deployment?

Send the workload, required capacity and timeline through the inquiry form or to sales@stratustech.tech. The sales team replies by email. A site visit can be requested once that option is enabled on the form.

Where are the data centers?

Two facilities, DC1 and DC2, are described on the Data Centers page. A city-level location for DC2 will be published only after confirmation.

Facility facts

DC1 connectivity
Land cable; submarine cable; strategic connectivity partnerships, as on the current data-center page.
DC2 power / structure
Generator N+1; 2N / N+1; three-layer steel structure (as currently published; pending engineering review).
DC2 cooling / systems
IPM 3.0; magnetic-levitation phase-change cooling (as currently published; pending engineering review).
DC2 location
To be confirmed (city / country). The wording “East South Asia” is not used on this staging site.
Office
A-G-3, Block A, Jalan PJU 1A/3J, Taipan 1, Ara Damansara, 47301 Petaling Jaya, Selangor Darul Ehsan.
Certifications
None listed until confirmed.

Contact us

Share a few details about your workload and our sales team will reply by email.

[[PLACEHOLDER: “Book a site visit” can be added as an inquiry type after confirmation.]]

Scroll to Top