Powernexu launches 3200W CRPS supply for AI servers

Powernexu has added a 3200W 54V CRPS power supply to its server power portfolio, aimed at high-density AI and GPU server platforms.

The new model, the CRPS3200CL5, is designed for server platforms, accelerator infrastructure and modular computing systems, where rising compute density is increasing demands on power delivery. It uses a long-body CRPS form factor measuring about 265 x 73.5 x 40 mm, with a 54V main output and a standby output for system management and auxiliary control functions.

The launch reflects a wider shift in AI server design. As total system power rises, engineers are focusing on more than a power supply unit's wattage rating. In these environments, power distribution, connector temperatures, redundancy behaviour, voltage drop, airflow and transient performance can all affect whether a server power subsystem operates reliably.

Distribution challenge

As server power rises, the current carried through connectors, cables, busbars and power-distribution boards becomes more significant. At a given power level, a higher intermediate distribution voltage reduces the required current, cutting resistive losses, easing stress on connectors and reducing localised heating.

This is particularly relevant in high-density AI systems because resistive losses rise with the square of current. In practical terms, even small amounts of resistance at a connector, joint or distribution path can become significant when a system is handling hundreds of amps.

Engineers working on multi-kilowatt AI servers therefore need to assess the full power path rather than treat the power supply as a standalone component. A higher-voltage bus can help lower current, but it does not remove the need to validate downstream conversion, protection, connector ratings, thermal behaviour and mechanical integration at platform level.

Redundancy limits

Powernexu also highlighted the importance of testing redundant CRPS architectures under failure conditions. In many servers, multiple PSU modules share the load during normal operation. If one module fails, is removed or loses input power, the remaining units must immediately take on more demand.

As a result, installed PSU capacity is not the same as usable redundant capacity. Engineers need to determine how much qualified capacity remains after the loss of a module or input source, and whether the surviving path stays within electrical and thermal limits while maintaining the workload.

Such testing should go beyond checking whether a system stays switched on. Relevant measurements include individual PSU current, common-bus voltage, connector temperature, current-sharing behaviour and the system response when a module is removed or reinserted.

Dynamic loads

AI and GPU workloads add another variable because power demand can change quickly. A server that appears stable under a fixed load may behave differently when accelerator demand shifts sharply, especially if that change happens during a redundancy event.

For that reason, qualifying server power systems can require controlled load-step testing as well as steady-state measurement. Parameters such as bus-voltage deviation, current-sharing balance, transient recovery, connector and busbar temperature, protection status and management telemetry can help engineers determine whether the subsystem remains within operating limits during both standard conditions and fault transitions.

Thermal link

The release of a higher-output CRPS model also underlines how closely electrical and thermal design are linked in dense server platforms. A connector or busbar may continue to function electrically while operating at a temperature that creates reliability concerns.

Increased contact resistance from mechanical wear, imperfect mating or other factors can create local hot spots, particularly at high-current cable terminations and power-distribution interfaces. Thermal imaging is one method used to identify these issues, especially where overall system temperature readings do not reveal localised heating.

Powernexu describes this wider change in AI infrastructure as a shift toward treating power delivery as a system-level engineering problem. That approach covers the route from facility or rack input through CRPS modules, isolation and redundancy stages, power-distribution hardware, the intermediate bus, downstream DC-DC conversion and the accelerator load itself.

The company supplies server and industrial power products, including CRPS server power supplies, redundant power systems and power-distribution products for data centres, AI and GPU computing, cloud platforms, enterprise IT and industrial applications. The CRPS3200CL5 is part of that range for environments where system designers must account for continuous and transient load, thermal limits and failure behaviour across the full power path.