Low-voltage switchboards, busway and PDUs

Below the transformer, power enters the low-voltage (LV) system: everything under 1000 V. This is where the data centre's electrical design becomes most visible and most dense. Large switchboards line the electrical rooms, UPS systems sit beside them, and power fans out through PDUs, panelboards and overhead busways to thousands of rack circuits. Most of the equipment an operator works on day to day is here, and so are most of the design decisions that decide how much IT load each rack can actually use.

This chapter follows the LV chain from the transformer secondary to the rack: switchboards and the breakers in them, transfer switches and UPS boards, PDUs and remote power panels, overhead busway, rack PDUs, and the choice between 208 V and 415 V distribution. It ends with how A and B power paths are loaded and how capacity becomes stranded.

The low-voltage chain

In a typical Canadian design, the transformer secondary at 600Y/347 V feeds a main switchboard, which also connects to the standby generator through an automatic transfer switch or paralleling switchgear. The main switchboard feeds the UPS systems and the mechanical loads (chillers, pumps, air handlers), which are backed by generators but usually not by the UPS. The UPS output feeds a critical distribution switchboard, and from there power is stepped down to 208Y/120 V or 415Y/240 V for the IT equipment.

Switchboards, switchgear and breakers

Low-voltage assemblies come in two main classes, built to different standards:

• Switchboards (certified to CSA C22.2 No. 244 in Canada) are compact assemblies, usually with fixed-mounted breakers. They are economical and common for most distribution.
• Low-voltage switchgear (certified to CSA C22.2 No. 31) uses draw-out power circuit breakers in separate compartments and is tested to withstand fault current for longer, which makes selective coordination easier (Lesson 7). It is preferred for main and critical distribution where maintainability and selectivity matter most.

The breakers in these assemblies fall into three families. Molded-case circuit breakers (MCCBs) are sealed units used for feeders and branch circuits. Insulated-case breakers are larger, often draw-out, used for mains and large feeders. Low-voltage power circuit breakers are the heavy-duty draw-out type used in switchgear. Larger breakers have electronic trip units with adjustable settings, usually described by the letters LSIG: long-time (overload), short-time (moderate faults, with a delay for coordination), instantaneous (heavy faults) and ground fault.

Breakers are either standard (80%) rated or 100% rated. A standard-rated breaker may carry continuous load (three hours or more, which describes all data centre IT load) only up to 80% of its rating. A 100%-rated breaker, tested in its assembly for continuous full-load operation, can carry 100%. Because IT load is continuous by nature, this distinction appears in every sizing calculation.

Transfer switches and bypasses

An automatic transfer switch (ATS) moves a load between the utility and a generator source. When utility voltage fails, the ATS signals the generator to start, waits for it to reach stable voltage and frequency (typically under 10 to 15 seconds), and transfers the load. When the utility returns and is stable, it transfers back, usually after a delay. Large sites use paralleling switchgear instead, which synchronizes many generators onto a common bus. ATSs for critical loads often come with a bypass-isolation switch so the ATS itself can be maintained without an outage. Similarly, every UPS has a maintenance bypass, an arrangement of breakers that routes power around the UPS so it can be serviced while the load stays on (raw utility or generator) power. Lesson 11 and DCE-340 discuss these further.

PDUs and remote power panels

The traditional way to deliver power to racks is through power distribution units (PDUs) on the data hall floor. A floor-standing PDU contains a step-down transformer, typically 150 to 300 kVA, and one or more panelboards with dozens of branch circuit breakers. Each breaker feeds a cable (overhead or under a raised floor) to a receptacle at a rack. Remote power panels (RPPs) are cabinets containing extra panelboards, without transformers, fed from a PDU and placed closer to the racks to shorten branch circuits and add circuit capacity.

In Canada, the PDU transformer steps 600 V down to 208Y/120 V. Each PDU transformer is a separately derived system with its own neutral-to-ground bond, so it also provides a local, clean grounding reference for the IT equipment.

Overhead busway

Many modern data halls replace PDU branch cabling with overhead busway: a prefabricated enclosed track of copper or aluminium busbars running above each row of racks. Plug-in tap-off boxes, each containing a breaker and a cable or receptacle for one rack, can be added, removed or moved without shutting the busway down. Busway is fed at one end (or the middle) from a switchboard or transformer. Its advantages are flexibility (racks can be added or changed in minutes), speed of installation, and the elimination of hundreds of individual branch cables under the floor or in trays. Its disadvantages are a higher cost at low densities and the need to plan tap-off ratings and phase balance carefully.

Voltage drop along a busway is usually small because the bars are large and the run short, but it should still be checked for long rows. Phase balance matters too: tap-off boxes are distributed across the three phases so that each phase carries a similar share.

208 V or 415 V?

For decades, North American data centres delivered IT power at 208Y/120 V through PDU transformers, because that is the standard North American commercial voltage. Today many new designs distribute at 415Y/240 V instead. In Canada the 600 V system is stepped down to 415Y/240 V by a transformer at the switchboard or busway feed, and the busway or panelboards then carry 415 V all the way to the racks.

The case for 415Y/240 V is strong:

• Server power supplies accept 200 to 240 V. On a 415Y/240 V system, each power supply is connected line to neutral at 240 V, which is near the top of its range and its most efficient operating point.
• The same current carries twice as much power: a 415 V circuit delivers 2.0 times the power of a 208 V circuit of the same ampere rating, so rack circuits, connectors and cables shrink.
• It removes a stage of transformation close to the racks (the PDU transformer), saving roughly 1 to 2% of losses and the floor space and heat of PDU transformers.

The trade-offs are higher available fault current at the racks, which raises interrupting ratings and arc flash energy, and the fact that 415Y/240 V equipment (panelboards, breakers, receptacles) must be suitable for that voltage. 415 V is also the standard voltage in much of the world, so equipment is readily available.

Rack PDUs

Inside the rack, power strips called rack PDUs distribute power to each server. A typical rack has two (A and B), each fed from its own circuit, mounted vertically at the back of the rack. Common types are basic (power distribution only), metered (showing total current, so operators can see loading), monitored per outlet, and switched (allowing remote power cycling of individual outlets). Outlets are usually IEC 60320 C13 (up to about 10 to 15 A) and C19 (up to about 16 to 20 A) types; inputs use locking or industrial connectors such as NEMA L-series or IEC 60309 "pin and sleeve" plugs. A three-phase rack PDU arranges its outlets in groups across the phases, and keeping those groups balanced keeps the upstream circuit balanced.

A and B paths and stranded capacity

Almost every server has two power supplies, each plugged into a different rack PDU fed from a different path. In normal operation the two supplies share the load roughly equally; if one path fails, the other must carry the whole load. This has a profound effect on how much of the installed capacity can actually be used.

The same logic applies all the way up the chain. In a 2N system, every UPS, switchboard and transformer normally runs below 50% of its rating, because its partner may need to hand over its whole load. If loads are not balanced between paths, or if some racks draw far less than the capacity reserved for them, capacity that has been built and paid for cannot be used. This is called stranded capacity, and it is one of the main reasons data centres monitor power at every level, from the switchboard to the individual outlet, and why designs such as distributed redundant and block redundant (DCE-340) aim to use more of the installed equipment.

Power for AI racks

AI training racks draw far more power than conventional racks: tens of kilowatts was once considered high density, but racks of liquid-cooled accelerators now draw over 100 kW, and designs for several hundred kilowatts per rack are in development. At these densities, AC distribution at 415 V needs very large conductors and connectors at each rack, and the conversion stages inside the rack become a significant source of loss and heat. Rack-level power shelves already convert AC to 48 to 54 V DC busbars feeding the servers, and in 2025 major suppliers announced higher-voltage DC architectures (for example around 800 V DC) for future AI racks. Electrical designers of AI facilities should expect distribution voltages and topologies to keep changing over the life of a building, and leave space and pathways accordingly.