Specifications
Architecture
NVIDIA MGX modular server architecture (Chenbro is an NVIDIA MGX partner)
Chassis Form Factors
1U and 2U compute trays, 2U AI server chassis, 4U air-cooled AI chassis, 21-inch OCP ORV3 series
4U AI Chassis GPU Support
Up to 8x double-width GPGPUs (NVIDIA H100, H200, B200, L40 class)
4U Cooling Layout
5x front-mounted + 5x mid-mounted 80 x 80 mm fan brackets for 8-GPU airflow
2U AI Server Chassis
Air-cooled, modular bays, configurations for up to 4 PCIe GPUs in a standard EIA 19-inch rack
1U Compute Tray
For NVIDIA GB200 NVL72 — 2x GB200 Superchips per tray, 18 compute trays per rack
1U Rack Density
36x NVIDIA Grace CPUs + 72x NVIDIA Blackwell GPUs per NVL72 rack
2U Compute Tray
For NVIDIA GB200 NVL36 — 2x GB200 Superchips per tray, 9 compute trays per rack
2U Rack Density
18x NVIDIA Grace CPUs + 36x NVIDIA Blackwell GPUs per NVL36 rack
Tray Cooling
Blind-mate liquid-cooled design for rack-scale AI factory deployment
CPU Compatibility
NVIDIA Grace CPUs and x86 architectures
Networking
NVIDIA BlueField DPU integration; OCP-standard-compatible I/O modules
Rack Standard
EIA 19-inch standard racks; 1U, 2U and 4U server support
Configuration Base
MGX supports over 160 system configurations from shared modular building blocks
Launch Status
MGX chassis line launched Q3 2025
Services
Open chassis supply plus JDM, ODM and OEM Plus customisation for system integrators and server brands
Applications
Generative AI training and inference, HPC, digital twins, cloud services, 5G edge
Overview
Chenbro builds its AI server chassis line directly on NVIDIA's MGX modular architecture, the reference design NVIDIA created so that OEMs, ODMs and system integrators can assemble rack-scale AI systems from a shared pool of mechanical, thermal and I/O building blocks. Because Chenbro is an MGX partner, its chassis accept the same compute trays, PCIe add-in modules and BlueField DPU carriers as the rest of the MGX ecosystem.
The portfolio spans the full rack-scale stack. At the top of the density curve are the 1U and 2U liquid-cooled compute trays built for NVIDIA GB200 NVL72 and NVL36 racks — each tray carries two GB200 Superchips and mates to the rack manifold through a blind-mate liquid-cooling interface. For air-cooled deployments Chenbro offers a 4U chassis that accommodates eight double-width GPGPUs behind ten 80 mm fan brackets, and a modular 2U server chassis that fits up to four PCIe GPUs in a standard 19-inch EIA rack. A 21-inch OCP ORV3 series addresses OCP-standard rack deployments.
All variants are supplied through Chenbro's open-chassis, JDM/ODM and OEM Plus service models, which allows buyers to take a validated mechanical platform and add their own branding, motherboard and firmware. QS Compute supplies Chenbro MGX chassis and compute trays with volume pricing and 15-day sample lead times.
Key Benefits
Validated MGX mechanicals — NVIDIA reference architecture means trays, risers and DPU carriers drop in without bespoke tooling. Both cooling paths — blind-mate liquid cooling for GB200 racks and high-airflow 4U cooling for PCIe GPU builds. 160+ configurations from one modular pool shortens integration time and reduces NRE cost. Forward compatible — Grace and x86 CPU support plus future GPU generations protect the platform investment.
Applications
Rack-scale AI factories and LLM training clusters (GB200 NVL72/NVL36), 8-GPU air-cooled AI training and inference servers, HPC and scientific computing nodes, digital twin and simulation farms, cloud and CSP deployments, 5G and edge AI infrastructure, and ODM/OEM barebone programmes.
Request a Quote — CHENBRO NVIDIA MGX AI SERVER CHASSIS — 1U/2U/4U FOR GB200 NVL72 & 8-GPU AIR-COOLED RACKS
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →