AI Infrastructure · Global Trade · Technical Services
AI infrastructure, sourced and delivered worldwide.
Noventraxis Global Limited supplies AI compute and data centre hardware to operators, integrators and enterprises across Asia, the Middle East, Europe and the Americas — backed by compliant cross-border logistics and hands-on technical support.
Headquartered in Hong Kong · Serving 20+ markets · English / 中文 / 日本語 support
Capabilities
Four connected lines of business
One counterparty for equipment, sourcing, movement and support — so your build schedule does not depend on stitching together four different vendors.
AI Infrastructure Equipment
Accelerated compute nodes, AI servers, head nodes and full rack-level configurations specified against your workload and power envelope.
Technology Hardware
Networking, storage, memory, power distribution and thermal equipment — new, factory-sealed and traceable to authorised channels.
Global Supply Chain
Sourcing, allocation management, export/import documentation, freight, bonded handling and last-mile delivery to data centre floor.
Technical Services
Rack-and-stack, burn-in and validation, firmware baselines, spares pooling, RMA coordination and capacity planning.
Portfolio
Equipment we source and integrate
Configurations are built to the customer's specification. Availability, lead times and eligible destinations are confirmed per enquiry.
Compute
- GPU and accelerator servers
- Multi-node AI training platforms
- Inference and edge appliances
- CPU compute and management nodes
Network
- High-speed Ethernet switching (100G–800G)
- InfiniBand fabrics, adapters and cabling
- Optical transceivers, DAC/AOC
- Structured cabling and patching
Storage & Memory
- NVMe all-flash arrays and JBOF
- Parallel and object storage platforms
- Enterprise SSD / HDD, server DRAM
- Backup and archive tiers
Power & Cooling
- Rack PDUs, busway and UPS
- Air-cooled and liquid-cooled solutions
- CDUs, manifolds and rear-door heat exchangers
- Containment and airflow management
Racks & Facility
- Standard and high-density racks
- Pre-integrated, factory-built rack systems
- KVM, console and out-of-band management
- Environmental monitoring
Software & Licensing
- Cluster management and scheduling stacks
- Virtualisation and container platforms
- Monitoring and observability tooling
- OEM support contracts and renewals
AI Rack Platforms
NVL72 rack-scale systems, air and liquid
We supply and deploy rack-scale NVIDIA platforms end to end — GB300 NVL72 and GB200 NVL72 at L11/L12, plus B300 and B200 systems in both air-cooled and direct-to-chip liquid-cooled configurations.
GB300 NVL72 — rack specification
- Form factor48RU rack, 1,150 × 2,547 × 1,719 mm, ≈1,472 kg
- Compute18 × 1RU compute tray, 4 GPU + 2 Grace CPU each
- GPUBlackwell Ultra, 279 GB on-die, up to 1,400 W TGP
- Fabric9 × NVLink switch tray, 57.6 Tbps full duplex
- Scale-outConnectX-8 SuperNIC, 800 Gb/s per GPU
- Power in380/415 Vac to 8 × 33 kW power shelf, 6 × 5.5 kW PSU N+1
- Power out50 Vdc over a 1,400 A busbar, 336 kW rack draw
- CoolingHybrid: ≥118 kW liquid, ≥18 kW air, front-to-rear
- CoolantPG25, 117–160 L/min, ≤45 °C in / ≤65 °C out, ΔP ≤19.4 psi
- CDU optionsIn-row L2L, in-rack L2L, sidecar L2A
- SafetyUQD blind-mate, leak detection, emergency shut-off
- ManagementRedfish/RESTful, PMC power telemetry, U-level tracking
Against the GB200 NVL72 generation the same rack envelope delivers roughly 1.5× LLM inference and 1.4× training throughput. GB200 NVL72 remains available where a proven platform is preferred.
B300 / B200 systems
HGX-class 8-GPU servers for teams that scale by node rather than by rack.
- Air-cooled chassis for existing halls at conventional rack density
- Direct-to-chip cold plate variants for high-density rows
- Mixed air/liquid rows where the facility is part-converted
- Head, storage and management nodes specified alongside
Cooling, supplied with the rack
- CDU range300–1,500 kW
- In-row CDUup to 300 kW
- Centralised CDUup to 1.2 MW
- Cold plate>20 kW per rack
- Air cooling4–20 kW per rack
Power, supplied with the rack
- Modular UPSup to 1.2 MW per cabinet, 0.8 m wide
- TopologiesUPS, HVDC, integrated power module
- Distributionsmart busway, PTU, metered PDU
- Storagehot-swap Li-ion, supercapacitor buffering
Platform names and specifications are those of the respective manufacturers, quoted from their published documentation and supplied to Noventraxis customers under OEM white-label agreements. Figures are design values; delivered configuration, availability and eligible destinations are confirmed per project and subject to export control.
Reference Deployment
A 6 MW modular liquid-cooled AI site
Two groups, sixteen 40ft containers, factory-built and commissioned on site — available through our OEM white-label programme and scalable a group at a time.
Supplied under an OEM white-label agreement. The layout is built to a Tier III‑equivalent design and is not certified by the Uptime Institute. All figures are manufacturer design values; delivered performance depends on site, climate, water quality and utility conditions, and is confirmed per project.
Rack & Thermal Design
Density is a power and cooling problem first
Accelerated racks now land well beyond the envelope most halls were designed for. We specify the rack, the power path and the thermal loop as one package, so the equipment we ship can actually run at the density it was bought for.
- Rack power budget, A/B feed and PDU selection
- Direct-to-chip liquid, rear-door exchangers or high-density air
- CDU sizing, manifolds, quick disconnects and leak detection
- Weight, floor loading and access-route survey
- Factory pre-integration to cut on-site build time
Supply Chain
From enquiry to installed capacity
A documented process, one project owner, and full visibility on lead time, landed cost and compliance status at every stage.
Requirement & specification
Workload, power, space and timeline are captured, then translated into a buildable bill of materials with alternates.
Sourcing & allocation
Quotation across authorised channels and vetted partners, with allocation secured against your committed schedule.
Compliance screening
End-user and end-use review, restricted-party screening, classification and licensing checks before any order is placed.
Freight & customs
Air, sea and consolidated shipping with full documentation, insurance, bonded handling and customs clearance.
Delivery & installation
Staging, inbound inspection, rack-and-stack, cabling, power-on and acceptance testing on site.
Lifecycle support
Spares pooling, RMA handling, firmware baselines, expansion planning and end-of-life disposition.
Technical Services
Support that goes past the delivery dock
Hardware arriving on a pallet is not capacity. Our engineering and logistics teams stay engaged through installation, validation and the operating life of the fleet — either directly or alongside your integrator and facility partners.
- Pre-sales architecture review and configuration validation
- Factory integration, imaging and burn-in prior to shipment
- On-site rack-and-stack, structured cabling and power commissioning
- Cluster bring-up support and acceptance test documentation
- Regional spares depots and RMA coordination with OEMs
- Refresh, expansion and decommissioning programmes
Commercial terms, lead times and eligible destinations are confirmed per project and subject to applicable export control and sanctions regulations.
Company
Noventraxis Global Limited
Noventraxis is a Hong Kong–based technology trading and services company focused on the equipment layer of AI infrastructure. We operate where three problems meet: getting the right hardware, getting it across borders lawfully, and getting it working on the floor.
Our model is deliberately narrow. We do not try to be a general electronics distributor — we concentrate on AI compute, the network and storage around it, and the power and cooling underneath it, for customers building or expanding real capacity.
Coverage
Where we operate
Sourcing and delivery capability across major data centre markets, with routing and entity structure selected per destination.
Availability is subject to compliance screening and applicable trade regulations in each jurisdiction.
Contact
Tell us what you're building
Send a bill of materials, a rough specification, or just the target capacity and timeline. We will come back with availability, lead time and an indicative landed cost.