Server Room Power Design: A Canadian Engineering Guide

Server room power design is the engineering of a continuous, redundant and code-compliant electrical path from the utility service to every rack. In Canada, that path is sized to measured peak IT load, protected by a UPS and, where the outage tolerance requires it, a generator. It is distributed through dedicated circuits, PDUs and rack PDUs, and it is installed under the Canadian Electrical Code, CSA C22.1, together with the provincial amendments that apply to the site. This guide sets out each stage in the order in which a designer addresses it.

Futuregenconsulting

futuregenconsulting.ca

Engineer Reliable Server Room Power

FutureGen Consulting designs electrical systems for Ontario projects, including power distribution, with over 30 years of engineering expertise.

Explore engineering services

TL;DR

Size the electrical system for the design-day peak of actual equipment draw, not nameplate ratings, and agree a kW-per-rack budget with the IT team before drawings begin.
Treat sustained IT load as continuous: under CEC Rule 8-104, keep it at or below 80% of the overcurrent device rating unless the equipment is marked for 100% operation.
Build redundancy as two independent paths (A and B), with dual-corded servers connected to different PDUs on different circuits.
Put cooling on the generator as well as the IT load, and bond racks, trays and the room to a telecommunications bonding busbar.
In Ontario, plans for larger services or standby generation must be submitted to the Electrical Safety Authority (ESA) for examination.

Start with load: how much power the room actually needs
Applicable codes and approvals in Canada
Distribution from the panel to the rack
UPS, generator and redundancy
Grounding, protection and safe working practice
Monitoring and efficiency
Perspective: why the electrical design should be engineered first
Engineering support for server room power
FAQ
Sources

Start with load: how much power the room actually needs

Every later decision, from breaker size to generator rating, derives from the load estimate, so the estimate comes first. The electrical system is sized for the design-day peak and not the annual average, because it must carry the worst case, as HSM University’s load estimation lesson explains. The same lesson lists the better inputs: measured power of real equipment running representative workloads, vendor power calculators, measurements from existing facilities with a similar workload, and a per-rack power budget agreed with the IT team or tenant.

Natural Resources Canada defines power density as the electricity demand of all data centre end uses divided by either floor area (kW/ft² or kW/m²) or the number of racks (kW/rack). Its Best Practice Guide for Canadian Data Centres recommends provisioning power and cooling to actual IT power usage rather than power supply nameplate ratings, which prevents oversizing of the electrical infrastructure.

A workable estimating sequence is:

List every device, with measured or vendor-calculated draw at peak workload.
Sum the draw per rack and compare it with the agreed rack budget.
Add cooling, lighting, UPS losses and any support loads that share the service.
Apply the continuous-load factor of 80% to determine the required circuit and equipment ratings.
Add a documented growth allowance, and decide where future capacity will be added: spare breaker positions, spare busway tap positions, or space for a second UPS module.

The cooling share matters at the service level. According to the NRCan 2021 market assessment cited in the guide’s appendices, IT equipment typically accounts for 65% to 85% of total data centre electricity use in Canada and cooling for 15% to 25%. A mechanical load of that size belongs in the electrical load calculation from the first draft. For the calculation method under Canadian code, see our guide to electrical load calculation, and for the cooling side, our HVAC system design services.

Applicable codes and approvals in Canada

Every electrical installation in a server room must comply with the electrical code in force where it is built. The national document is the Canadian Electrical Code, Part I, CSA C22.1:24, the twenty-sixth edition, approved in June 2023. It applies to electrical work and equipment at all voltages in buildings, structures and premises, and its revisions include arcing-fault clearing-time requirements in Section 2 and load calculations in Section 8.

Two requirements shape most server room designs:

Continuous load. Rule 8-104 limits the continuous load on a circuit to 80% of the rating of its overcurrent device, unless the device and assembly are marked for continuous operation at 100%. Conductor ampacities come from Rule 4-004 and Tables 1 to 4, with correction factors for ambient temperature and for more than three current-carrying conductors in a raceway.
Provincial approval. Provincial rules for licensed electrical work, permits and inspection apply on top of the national code. The same source notes that the 2024 Ontario amendments require plans to be submitted to and examined by the Electrical Safety Authority for three-phase services or standby generation of 400 A or more, single-phase services of 600 A or more, feeders greater than 1000 A, and emergency power supplies for life safety systems.

A server room of any real size therefore reaches the ESA plan examination threshold at the service or generator level, and the drawings must be permit-ready before the electrical contractor mobilizes. Where the room is part of a larger building, the power distribution design for the whole building, described in our P.Eng. guide to power distribution design for Ontario buildings, determines how much service capacity is available to it.

Large computational loads in Ontario

Rooms rated above 10 MW fall under a separate regime. The IESO’s technical requirements for large computational loads apply to loads connected to the Ontario transmission system or to a distribution system whose final System Impact Assessment and Notification of Conditional Approval are issued after June 30, 2026. They include ride-through of two consecutive voltage disturbances within 30 seconds and tripping of all loads when system frequency drops below 58.8 Hz for 10 seconds. In Alberta, the AESO’s connection guide for transmission-connected data centres requires submission of the UPS type, the duration and portion of load supported, and a medium voltage single-line diagram. Most enterprise server rooms are far below these thresholds, but the documents show the level of detail that utilities expect from backup design.

Distribution from the panel to the rack

The distribution chain determines how easily a room can be maintained, expanded and balanced. A typical North American arrangement is described in an Electrical Engineering Portal white paper: floor-mounted PDUs, rated 50 kW to 500 kW, contain a main input breaker, branch circuit panelboards, a transformer, surge arrestor and monitoring, and step 480 Vac down to 120/208 Vac. Remote power panels resemble PDUs without the transformer and feed the racks. Rack PDUs then distribute power directly to the IT equipment, in single-phase or three-phase form depending on expected rack density.

For smaller rooms, the same logic is applied at reduced scale: a dedicated panel for the room, dedicated circuits, and rack PDUs, with no shared office loads. Overhead busway with plug-in tap boxes is an alternative to cable whips where racks are added or moved often; it adds up-front cost, and the A and B busway runs must be planned so that a single fault cannot affect both.

Dedicated circuits and A/B feeds

Three rules recur in well-designed rooms:

Server loads do not share circuits with office equipment, and each PDU has its own dedicated circuit.
Dual-corded servers provide redundancy only when their two cords connect to different PDUs on different circuits, ideally from different sources.
Power bars are never chained to power bars, nor a UPS to a power bar, since the overloaded cords and receptacles are a recognized cause of trips and fires.

The University of Toronto’s data centre design considerations apply the same principle: each server power supply on a separate circuit breaker, and each cabinet with redundant PDUs fed from different sources, at least one protected by a UPS.

Sizing PDUs and branch circuits

Branch circuit capacity is calculated as rated voltage times rated current for single-phase, and the square root of three times rated voltage times rated current for three-phase, according to the Server Technology iPDU Handbook. The handbook also notes that a three-phase PDU divides its branch circuit wires into pairs of single-phase circuits serving the outlet receptacles.

A worked example shows the effect of the continuous-load factor. A breaker rated for 7.36 kW can carry 5.9 kW continuously, which is 80% of its rating, as rack-density.com demonstrates with a 32 A, 230 V circuit. Under the Canadian 120/208 V arrangement, the same arithmetic applies with the local voltage. The guide adds that the 80% principle extends to PDU input ratings and to the UPS, and that peak draw is the sum of what devices actually pull at their busiest, not the sum of nameplates.

Design step	Basis	Practical result
Rack budget	Agreed kW per rack, measured draw	Defines rack PDU type and phase
Circuit rating	Continuous load at or below 80% of breaker rating	Breaker and conductor selection
Phase balance	Equal loading on three phases	Lower heat, fewer upstream trips
Redundancy	Separate A and B paths	Each cord on a different PDU

UPS, generator and redundancy

The backup chain follows a defined sequence. The utility service feeds main switchgear, where utility and generator sources meet through automatic transfer switches. A UPS bridges the seconds between a utility failure and generator pickup. Power then passes through PDUs, remote power panels or busway to the rows, and finally to rack PDUs where dual power supplies connect. Toronto Metropolitan University’s specification for rack-mounted power protection illustrates the UPS function for a communications room: a continuous-duty, on-line, solid-state unit that protects equipment from voltage fluctuations, brown-outs, blackouts, surges and sags.

Matching redundancy to the outage tolerance

The University of Toronto document describes the four tier classifications: Tier I provides a dedicated UPS, cooling and generator; Tier II adds redundant critical components; Tier III permits maintenance without shutdown through redundant delivery paths; and Tier IV adds fault tolerance. It also states that a permanent generator is needed for full capacity during outages longer than 5 minutes.

At the large end, the Markham data centre operated by Core Data Centres uses A and B power distribution to Tier 3 standards, with 3 x 1100 kW UPS units providing 10 minutes of backup at full load, 3 x 1.5 MW generators in an N+1 configuration, and five days of runtime at full load. Few enterprise rooms need that scale, but the architecture of independent paths, UPS ride-through and generator runtime is the same one applied at a smaller rating.

Cooling belongs on the generator

Servers survive a transfer on UPS power, whereas chillers, CRAH units and pumps usually restart only after the generator picks up. If they restart slowly, room temperature becomes the limiting factor of the outage. The critical mechanical loads are therefore part of the electrical design, and the generator, the transfer scheme and any load-shedding logic must account for them. Our articles on emergency versus standby power, generator load shedding under CSA C282:24 and generator load bank testing cover these topics, and our emergency power design service applies them to individual projects.

Upgrading an existing room

Many server rooms are added to buildings whose switchboards and supplies were never intended for a sustained IT load. Assessing the existing supply, metering, surge protection and generator connection before adding racks is normally the first task; examples of this scope are outlined in material on data centre power upgrades. In Ontario or Alberta, the resulting design must still be prepared and stamped under local requirements and the Canadian Electrical Code.

Grounding, protection and safe working practice

Protective measures for a server room are straightforward to specify and expensive to retrofit:

A telecommunications bonding busbar, with racks and cable trays bonded to it, and a direct connection to the main building grounding system.
Surge protection at the service and at the distribution panel, in addition to the UPS.
A UPS with bypass, so that maintenance does not interrupt the load.
Labelled receptacles and PDUs, updated panel schedules and a one-line diagram kept inside the room door.

The University of Calgary’s electrical systems design standard (Version 5.0, March 2026) offers a published institutional reference: wiring beneath a raised access floor in EMT conduit to underfloor distribution boxes, liquid-tight flexible conduit to equipment and outlets, and, in telecommunications rooms, at least two dedicated 20 A, 120 V duplex outlets on emergency power with direct connection to the main building grounding.

Where energized work is anticipated, arc flash practice under CSA Z462 and the applicable provincial rules for licensed work apply. Arcing-fault clearing time also appears in the 2024 CEC revisions, which is a reason to coordinate protective devices through a power system study before equipment is purchased.

Monitoring and efficiency

Metered rack PDUs provide remote monitoring of connected loads, with user-defined alarms that warn of circuit overloads before the IT load drops. The Server Technology handbook also explains that unbalanced three-phase loads generate heat, raise cooling costs and increase the chance of tripping a PDU or upstream breaker, so equalizing the three phases is both an efficiency and a reliability measure.

At facility level, NRCan recommends periodic reporting of standardized indicators, including PUE, ERE, CUE, WUE and cooling efficiency ratio, and a dashboard that centralizes the efficiency data. PUE ranges from a minimum of 1.0, the most efficient, upward without limit. Metering designed in at the panel, PDU and rack level supplies the data for those indicators at no later retrofit cost.

Perspective: why the electrical design should be engineered first

We regard the server room as a mechanical and electrical problem that happens to host IT equipment. The load estimate, the A and B paths, the generator scope and the cooling restart sequence are interdependent, and a decision taken in one of them changes the others.

In our experience of Ontario and Alberta projects, the drawings that move quickly through permitting are those in which the load calculation, single-line diagram, protective device coordination and mechanical loads were prepared together by the same engineering team. That is the way we approach power distribution, emergency power and HVAC design, and it is why we recommend engaging a P.Eng. before equipment is specified and not after.

A room designed to a documented kW-per-rack budget, with metering and spare capacity planned from the start, can be expanded without reopening the electrical design each time.

Sanjay

Engineering support for server room power

FutureGen Consulting Inc. prepares MEP design for residential and commercial projects in Ontario, including power distribution, emergency power, HVAC, power system studies and electrical construction cost estimation, with permit-ready engineer-stamped drawings and building code compliance. For a server room, these services can be combined under one scope, from load calculation to field review.

Project teams that need a code-compliant design can review our engineering services or begin with our power distribution engineering services.

FAQ

How much power does a server rack need?

The figure depends on the equipment, so it is set as a per-rack budget agreed with the IT team and based on measured or vendor-calculated draw at peak workload. NRCan expresses power density in kW per rack for this purpose. The circuit is then rated so that this load stays within 80% of the breaker rating.

Does a server room need a generator?

Where outages longer than a few minutes cannot be tolerated, yes. The University of Toronto design considerations state that a permanent generator is needed for full capacity during outages longer than 5 minutes. The generator must also carry the cooling, not only the IT load.

Which code governs server room wiring in Canada?

The Canadian Electrical Code, CSA C22.1:24, applies, together with provincial amendments, permits and inspection. In Ontario, plans for services or standby generation at the stated thresholds are examined by the Electrical Safety Authority.

Why use two PDUs per rack?

Dual-corded servers remain powered during a failure or maintenance only if each cord connects to a different PDU on a different circuit. Ideally the two circuits come from separate sources, with at least one protected by a UPS.