AI Rack Design Is Becoming a System-Level Engineering Challenge

On October 12, 2026, Germany-based rack and systems supplier Rittal announced that it will show an architecture at the OCP Global Summit in San José (October 12–15) that combines the rack, direct liquid cooling, and DC power supply into a single system [1]. According to Rittal, its RiMatrix HPC arranges bayed OCP racks as a central coolant distribution unit (CDU), high-performance racks for compute modules, and a power sidecar placed next to the IT equipment to convert and distribute direct current [1]. 

One qualification matters. The power sidecar uses 800 V power electronics from Bosch, and Rittal itself calls it a development demonstrator. The release gives no availability date, customers, or pricing [1]. This is a showcased architecture, not evidence of broad deployment. 

 

Why rack design is becoming more integrated 

Rittal says cooling, rack, and power supply are integrated so closely, mechanically and functionally, that megawatt-range computing can fit within a few square meters and be easier to service [1]. As industry analysis shows, when density rises, the rack stops being a simple enclosure and becomes a structure that is tied to cooling and power delivery. 

 

Three engineering interfaces that matter 

The first is structure and mechanical integration. Bayed racks need stable alignment, and component positions and service aisles must be planned ahead. Rittal says its CDU uses modular units that can be replaced during operation [1], which means the mechanical design has to leave room for swapping them. The second is power. The sidecar sits directly beside the IT equipment and is integrated with Rittal’s RiLineX busbar platform [1], so space, connections, and layout between the sidecar and compute racks have to be coordinated. The third is cooling. Direct water-based chip cooling needs room for piping and connections, which affects layout. These are general engineering considerations, not design details published by Rittal. 

 

Why manufacturing has to be involved early 

Rittal stresses that the architecture is meant to be planned, reproduced, and used worldwide, and it is expanding factories in Europe and the U.S. with a €100 million investment [1]. Reproducibility depends on interface dimensions, tolerances, and service access being clearly defined first, then checked for fit and validated at the prototype stage. Not every AI rack follows the same design, and the OCP-oriented approach is one route among several. 

 

Our perspective 

We see rack design as more than a question of enclosure dimensions. From our perspective, once a rack becomes a system of bayed units, the mechanical interface through which equipment enters and leaves the rack deserves to be defined early, because it decides whether a unit can be pulled and refitted smoothly during service. We believe manufacturability and service access should be considered alongside power and thermal requirements, not after the design is frozen. 

 

Conclusion 

As AI infrastructure power density rises, structural, power, and thermal requirements can no longer be handled separately. Rittal’s architecture is still at the showcase stage, and real deployment remains to be seen, but it is a reminder that repeatable systems rest on clearly defined, verifiable mechanical and functional interfaces. 

Contact Hongguo Technology for more information about the company and its services.