How to Select a Rack PDU for High-Density AI Data Centers

Rack Power Architecture

Before selecting a PDU, define the rack power architecture. Key questions:

• What is the total rack power draw (kW)? This determines the PDU input rating.
• What is the supply voltage and phase configuration? (Single-phase 230V, three-phase 400V, or other)
• Is A+B redundancy required? (Two independent PDU feeds per rack)
• What is the power source? (Busway tap-off, branch circuit from distribution board)
• What outlet types do the servers require? (IEC C13, IEC C19, NEMA, or other)

Input Configuration

The PDU input must match the power source. For high-density AI racks, three-phase input PDUs are common because they distribute the load across three phases, reducing the current per phase compared to a single-phase PDU at the same total power. Three-phase PDUs also allow the load to be balanced across the three phases of the supply.

The input connector type must match the power cable assembly or busway tap-off unit. IEC 60309 connectors are standard for European deployments. Confirm the input connector type and current rating with the busway or branch circuit specification before ordering.

Outlet Requirements

Outlet type and quantity must match the server power supply requirements. Common outlet types in data center PDUs:

For AI GPU servers, IEC C19 outlets are commonly required for high-power server power supplies. Confirm the outlet type and quantity required per server, and multiply by the number of servers in the rack to determine the total outlet requirement.

Metering and Monitoring

PDU metering capability ranges from no metering (basic PDU) to per-outlet metering (intelligent PDU). For AI data center deployments, the appropriate metering level depends on the monitoring requirements:

• Basic PDU (no metering): suitable where power monitoring is provided by upstream meters. No visibility at the rack or server level.
• Metered PDU (input metering): measures total PDU input current and power. Provides rack-level power visibility. Suitable for capacity planning.
• Monitored PDU (input + outlet metering): measures per-outlet current. Provides server-level power visibility. Suitable for GPU cluster power budget management.
• Intelligent PDU (per-outlet metering + remote management): adds remote outlet switching, environmental monitoring, and DCIM integration. Suitable for high-availability AI data center deployments.

Redundancy Considerations

AI GPU clusters are high-value infrastructure. A+B redundant power feeds — two independent PDU feeds per rack from separate power sources — protect against a single PDU or power feed failure. For A+B redundancy, each server must have two power supplies connected to separate PDUs. The PDU rating must be sufficient to carry the full rack load on a single feed in the event of a failure of the other feed (N+1 or 2N redundancy).

Integration with Busway

When the power source is an overhead busway, the PDU input connector must match the busway tap-off unit output connector. Confirm the tap-off unit output connector type and current rating before specifying the PDU input. The PDU input cable (whip) length must be sufficient to reach from the busway tap-off to the PDU input without excessive slack.