The age of artificial intelligence (AI) is upon us, and with it comes an insatiable demand for computing power. At the heart of this revolution lies the data center, the engine of modern AI.
As AI workloads grow more complex and scale toward “AI factory” models, data center infrastructure must evolve in response, requiring faster time to market, greater scalability, and consistent performance.
NVIDIA’s MGX™ platform addresses these challenges with a modular reference design that brings new levels of flexibility and efficiency to data center architectures, enabling manufacturers to accelerate deployment and reshape how future AI data centers are built.
The power of NVIDIA MGX
NVIDIA MGX is a modular reference architecture for accelerated computing, helping system builders design future-compatible systems faster, reducing costs, accelerating time to market, and improving ROI. MGX supports hundreds of server combinations, from single nodes to NVIDIA Vera Rubin third-generation rack-scale architectures for AI, high-performance computing, and NVIDIA Omniverse™ workloads.
With NVIDIA MGX, manufacturers can efficiently create customized systems for diverse AI workloads using a modular architecture that minimizes development effort. By reusing proven design building blocks, partners can save millions in engineering resources and reduce time to market by months, enabling faster response to rapidly evolving AI demands.
Danfoss: The foundation of MGX performance
Danfoss Power Solutions has been a significant contributor to the MGX-powered racks, the NVIDIA GB200 NVL72 and GB300 NVL72, with products that support the thermal and fluid management requirements needed to operate at extreme compute densities. This includes evolving quick disconnect technologies, where Danfoss has played a key role in design and specification development.
Inside the MGX GB200 NVL72 Rack
For the NVIDIA GB200 NVL72 Rack with MGX, Danfoss is providing a suite of our industry-leading solutions, including:
- UQD04, UQDB04, and FD83 1-inch Quick Disconnect Couplings: These couplings are essential for robust and reliable liquid cooling systems, ensuring leak-free operation and optimal thermal management.
- FF4514 Swivel Adapter: Our FF4514 swivel adapter offers unparalleled flexibility in routing and connectivity. This 90-degree adapter is designed for easy and fast connections in liquid cooling hose assemblies, enabling a compact and efficient system design.
- EHW194-24 Hose: This high-quality hose is designed to withstand the demanding conditions of data center environments, ensuring the integrity of the liquid cooling loop.
Inside the MGX GB300 NVL72 Rack
The powerhouse MGX GB300 NVL72 Rack also features a range of Danfoss components designed for maximum performance and reliability. MGX GB300 NVL72 builds on the robust ecosystem established for MGX GB200 NVL72.
- MQD03, MQD04, UQD04, UQDB04, and FD83 1-inch Quick Disconnect Couplings: This combination of our advanced quick disconnect couplings ensures that the GB300 rack can handle the immense thermal loads generated by the latest generation of GPUs.
Inside the NVIDIA MGX rack architecture for Vera Rubin NVL72
The highly anticipated NVIDIA Vera Rubin NVL72 rack features larger couplings to accommodate higher thermal loads.
- MQD03, MQDB03, MQD04, MQDB04, UQD08, UQDB08, and FD83 2-inch Quick Disconnect Couplings: MGX rack features MQD as well as the newly redesigned UQD according to OCP’s V2 specification.
The Danfoss difference: Engineering excellence for the AI era
The inclusion of Danfoss components in NVIDIA's MGX racks is a recognition of our decades of expertise in thermal management and fluid conveyance. Our solutions help enable higher compute density, improved system reliability, and efficient cooling performance — critical factors in scaling modern AI infrastructure.
As AI and HPC continue to push the boundaries of what's possible, Danfoss will be there to provide the foundational technologies that enable the next wave of innovation. We are proud to partner with NVIDIA in building the future of computing.
