Data Center Power: A/B Feeds, Redundancy Tiers, and Capacity Planning Without Guesswork

Power is the one dependency in a data center that nothing else survives without. Yet in most colocation cages, the weakest link is not the utility feed or the generator — it is what happens in the last three metres, inside the rack, where cabling and capacity decisions are made in a hurry.

This guide covers how redundant power is actually delivered in a colocation environment, what the common redundancy notations mean, and how to plan rack capacity so you are not the reason a "fully redundant" facility goes dark for your equipment.

What A/B feeds really mean

A colocation cabinet is typically delivered with two independent power feeds, labelled A and B. Each feed should trace back through separate breakers, separate PDUs, and ideally separate UPS systems and distribution paths. The intent is simple: either feed can be lost — for maintenance or failure — and your equipment keeps running on the other.

That intent only holds if three conditions are true:

• Every device with dual power supplies has one cord on A and one cord on B.
• Each feed alone can carry the full load of the cabinet.
• Single-corded devices are fed through a transfer switch, not simply plugged into whichever outlet was free.

The third point is where most cages fail an audit. Firewalls, older switches, KVMs, and appliances often ship with a single power supply. Plugging them into feed A means a routine A-side maintenance window takes them offline, regardless of how redundant the building is. An automatic transfer switch (ATS) in the rack solves this by presenting one output fed from both A and B.

Redundancy notation, in plain terms

• N — exactly enough capacity to carry the load, with no spare. A single failure is an outage.
• N+1 — one spare component beyond what the load requires. One UPS module or one generator can fail or be serviced without loss.
• 2N — two complete, independent systems. Each can carry the entire load alone. This is what genuine A/B distribution is built on.
• 2N+1 — two complete systems, each with a spare component. Found in facilities designed for concurrent maintainability with a failure margin.

Facility marketing sometimes describes redundancy at the UPS level while the distribution downstream is shared. Ask specifically where the redundancy ends: the UPS, the PDU, the busway, or the outlet in your cabinet.

The 50 percent rule for dual-feed loads

If both feeds are live and sharing load, each feed should carry no more than roughly half of the cabinet's rated capacity. The reason is arithmetic: when one feed drops, the other must absorb everything instantly. A cabinet drawing 80 percent on each side is comfortable until an A-side event pushes the B side past its breaker rating, and a maintenance window becomes an outage.

In practice this means a 5 kW cabinet with A/B feeds is a 5 kW cabinet — not a 10 kW cabinet. Plan the equipment budget against the single-feed limit, not the sum.

Capacity planning that survives contact with reality

Nameplate ratings on a server power supply describe the maximum the supply can deliver, not what the machine draws. Planning against nameplate wastes contracted capacity; planning against idle draw invites breaker trips during boot or heavy load. A workable approach:

1. Measure actual draw at the PDU under representative load, not at idle.
2. Add headroom for boot-time inrush, which is briefly higher than steady state.
3. Reserve capacity for the growth you already know about — the next node, the next shelf of drives.
4. Re-measure after every significant deployment and update the rack elevation record.

Metered or switched PDUs with per-outlet reporting make this straightforward. Without them, capacity planning is estimation, and estimation is how cabinets end up over-committed.

Where dense hardware changes the maths

GPU and high-density compute nodes have shifted typical rack draw well beyond the 3–5 kW that older cage designs assumed. A single dense node can consume what an entire legacy cabinet once did. Before deploying dense hardware into existing space, confirm three things with the facility: available power per cabinet, cooling capacity for that heat load, and whether the busway or distribution serving your row can support the increase. Power is often available before cooling is.

Documentation is part of the power system

A redundant design that nobody can verify is not operationally redundant. Every cabinet should have a current record of which outlet feeds which device, on which feed, through which breaker. When something trips at 3 AM, that record is the difference between a five-minute fix and an hour of tracing cords by flashlight.

At minimum, keep: a rack elevation, an outlet-to-device map, per-feed measured load, and photos of the front and rear of the cabinet after any change.

Common failures we find during audits

• Dual-corded servers with both cords on the same feed
• Single-corded devices with no in-rack transfer switch
• Feeds loaded past 50 percent, so failover trips a breaker
• PDU outlet maps that no longer match reality after months of changes
• Power cords routed through the cable pathway, sharing space with copper runs

None of these are exotic. All of them are found regularly, and all of them are fixable during a scheduled visit rather than during an incident.

How StackTrue helps

We audit rack power distribution, correct A/B cord assignment, install in-rack transfer switches for single-corded equipment, measure and document per-feed load, and produce an accurate outlet-to-device record you can hand to any technician. Where capacity is the constraint, we scope what your current cabinet can support before you buy hardware that will not fit within it.

StackTrue provides remote hands, smart hands, rack & stack, structured cabling, and 24/7 break-fix across Vancouver, Burnaby, Richmond, Surrey, and the Fraser Valley, with two-hour emergency response in Metro Vancouver.