3 Things to Consider When Planning a Data Center — An Electrician’s Perspective

TL;DR: After a decade living in switchgear rooms and server aisles, the three decisions that make or break a data center are: (1) your power architecture and capacity plan, (2) how you design for reliability, safety, and maintenance, and (3) the quiet foundations—grounding, bonding, power quality, and monitoring. If you get those right, everything else gets easier.

A quick story to set the stakes

It’s 2:07 a.m. on an integrated systems test. We simulate a utility failure. The UPS holds, generators pick up, IT stays humming—and then the room starts getting hot. Why? A beautifully redundant electrical design… forgot the chillers. They weren’t on the emergency bus. No flames, no headlines, just a near miss and a hard lesson.

That night sums up data center work: you don’t just keep servers on—you keep the whole ecosystem alive. With that in mind, here’s a practical, readable playbook from an electrician’s chair.

1) Power Architecture & Capacity Planning

The idea (plain and simple)

Your one-line diagram is your story. It says where power comes from, how it’s cleaned, how it’s distributed, and how it grows. Get it wrong and you buy trouble you can’t see. Get it right and you get boring nights and predictable mornings—which is success in this world.

What to think about first

• Where power comes from: One utility or two? What fault current can it deliver? Can it meet day-1 load and year-5 growth without a major rebuild?

• How you ride through: UPS (for seconds to minutes) + generators (for hours). UPS cleans the power; generators carry the load. You generally need both.

• How you split risk: A/B paths—two independent routes from source to rack. Dual-corded IT gear plugs into both.

• How you maintain: Bypass paths around UPS and critical panels, so technicians can work without dropping the room.
  

A simple (readable) single-line

         Utility

            |

      [Main Switchgear]

        |           |

     [UPS A]     [UPS B]

       |             |

    [PDU/RPP A]   [PDU/RPP B]

       |             |

   Racks (A PDU)  Racks (B PDU)

   Generators -> Paralleling Switchgear -> Feed Main Switchgear (on outage)

   Maintenance Bypass around UPS and critical panels for safe servicing

How I usually phase capacity (less stress later)

• Day-1 at ~60–70% utilization of major gear. That buffer covers real-world diversity, growth, and weird edge cases.

• Scalable blocks: Add UPS modules, PDUs, and RPPs in predictable steps.

• Cooling power is “critical,” too: Put CRAC/CRAH controls and the chiller plant on the emergency source, or you’ll keep servers alive while cooking them.
  

Mini case: “We’re only adding 10 racks…”

A client asked for 10 more racks “no problem.” Nameplate said 4 kW each, reality turned into 8–10 kW average. The PDUs could handle it, the RPP breakers were fine, but the whips and cable tray fill weren’t. We re-pulled feeders, moved some load to the B side, and added branch-circuit monitoring so the next change was based on real numbers, not guesses.

Quick checklist (use this in your design review)

• Utility fault current verified; short-circuit and coordination studies run

• UPS chosen for the right job (double-conversion for core, line-interactive only for edge/light duty)

• Generators sized for N+1 or 2N—aligned with your business tier target

• A/B distribution is real (not “two outputs from the same UPS”); dual-corded IT planned

• Maintenance bypass around UPS and key panels

• CRAC/CRAH/chillers and their controls ride through on emergency power

• Clear growth plan: UPS modules, PDUs, RPPs, whips, and panel spaces reserved
  

Common gotchas (I see these a lot)

• Fake A/B: Two paths that meet too soon or share a hidden component (same control power, same PLC, same breaker).

• Brittle UPS choices: Efficiency features are great—until transfer behavior surprises you. Test them.

• Forgetting the “little” conductors: Rack PDUs get attention; whips and neutrals quietly overheat if you don’t size for reality.
  

Budget/impact (plain talk)

• Moving from N to N+1 adds noticeable cost now and saves heartburn later. 2N costs more, but buys you true path independence. If uptime is revenue, that math usually works. Think in blocks so upgrades are surgical, not invasive.

2) Reliability, Safety & Maintainability

The big picture

The best data center is the one you can safely maintain without downtime. That’s not just gear—it’s procedures, labeling, access, and training. Reliability isn’t luck. It’s design + ops, proven by testing.

Redundancy that means something

• N: no spare.

• N+1: one spare of each critical component (e.g., 3+1 UPS).

• 2N: two full, independent systems. Pick one on purpose. If your SLAs scream “no downtime,” budget for at least N+1 on power and cooling, and design for concurrent maintainability.

  
  

Maintenance without drama

• Wrap-around bypass at UPS and around critical boards: service the UPS while IT runs on utility or generator.

• Static Transfer Switches (STS) where needed: move a downstream panel between sources without a blip.

• Clearances and egress: Don’t bury panels; plan door swings and aisles so people and breakers can both move freely.
  

Safety you can feel (and see)

• Arc-flash study & labels: Know the incident energy before anyone opens a cover.

• PPE and training: The right gear is only helpful if people wear it and know why.

• Lockout/Tagout: Written procedures, real locks, and a culture that enforces them.

• Thermal scans: Infrared imaging of live gear catches loose lugs and hot spots before they fail.
  

  

Commissioning that proves it (not a paperwork exercise)

• FAT/SAT for major equipment so you don’t learn hard truths on site.

• Integrated systems testing (IST): Kill normal power and watch everything do what the one-line promised—UPS carry, generators start, ATS operate, cooling stays online, alarms land, and trends are logged.

• Load-bank tests: Don’t just “assume” the generator can carry the room. Make it sweat.
  

Mini case: The “silent trip”

A facility had perfect paper coordination—until a real downstream fault hit. An upstream breaker tripped first, dropping a whole PDU. The cause? Settings never made it into the field. We pulled the as-left settings from each breaker, updated the one-line, and repeated testing. Lesson: paper ≠ practice; verify in metal.

Quick checklist

• Redundancy model chosen (N / N+1 / 2N) and actually implemented

• UPS/STSs have documented, tested bypass paths

• Arc-flash study done; labels current; PPE stocked and training logged

• Lockout/Tagout procedures written and used

• Time-current coordination settings documented and loaded into breakers

• IR scans and periodic generator/ATS tests on the calendar

• IST completed with real failure scenarios; issues tracked to closure
  

Common gotchas

• Controls SPOFs: Two power paths controlled by one non-redundant brain.

• Old labels & drawings: If the one-line isn’t updated, you’re flying by memory.

• No “maintenance mode”: High arc energy at main gear with no way to temporarily reduce it during energized work.
  

Budget/impact

• Safety/commissioning is the cheapest insurance you’ll ever buy. A single avoided outage can pay for years of IR scans, labels, training, and ISTs.

3) Grounding, Bonding & Power Quality (and Monitoring)

Why it matters (the quiet heroes)

Most outages are obvious. Most nuisance issues aren’t: random server reboots, sensitive gear acting “possessed,” or a PDU that runs hot. Those often trace back to bonding, harmonics, and power quality—and you only see them if you’re measuring.

Grounding & bonding that keeps the peace

• One clear reference: A robust equipment grounding system and a telecom bonding backbone keep everything at the same electrical potential. That reduces noise and ensures faults clear fast.

• Bond what you install: Ladder tray, cabinets, racks—bond them all. Label the bonds. Keep it neat so future you can follow it.

  
  

Surge & transient defense

• Layered surge protection (SPD): Service entrance, distribution, and sensitive panels. Think of it like three goalies. If the first misses, the next two keep the game alive.

  
  

Harmonics & neutrals (don’t cook your wiring)

• Non-linear loads (UPS rectifiers, server power supplies, VFDs) distort waveforms and heat neutrals.

• How to manage: Oversize neutrals where needed, consider K-rated transformers, and use harmonic filters if THD is high.

• Verify, don’t assume: Measure THD at commissioning and after big IT refreshes.

  
  

Monitoring that pays for itself

• EPMS (Electrical Power Monitoring System): Meter feeders, PDUs/RPPs, and—if you can—branch circuits.

• What you get: Real-time load, alarms, and quality (sags/swells, THD).

• Why it matters: You spot problems early, trend capacity honestly, and make change management safer.

  
  

Mini case: The mystery reboot

A row of servers rebooted “randomly” once a week. Main meters showed nothing. Branch monitoring found brief voltage sags on one RPP during a chiller ramp-up. We re-balanced loads, tuned the VFD settings, and added a filter. The “random” became “never again.”

Quick checklist

• All racks, trays, and cabinets bonded to a clear backbone

• SPDs at service, distribution, and sensitive panels with status monitoring

• Neutrals sized for non-linear loads; K-rated transformers where appropriate

• Harmonic filters considered where THD is high

• EPMS specified and integrated; alarming and trending enabled

• Commissioning includes power-quality baselines (save those reports)
  

Common gotchas

• Dead SPDs: Surge modules fail quietly; without monitoring, you don’t know.

• “It’s just a ground wire”: Poor bonds create noisy references and weird behavior.

• No data: If you aren’t logging power quality, every post-incident meeting starts with opinions, not facts.
  

Budget/impact

• EPMS can feel like a “nice to have” until it saves a single incident. After that, no one argues. Start with feeder and PDU/RPP meters; add branch monitoring as you scale.

Quick comparison: Redundancy models

Pulling it together: a readable plan

1. Start with the one-line. Put it on the wall. Red-line it in meetings. Everyone should understand the flow: utility → switchgear → UPS → PDUs/RPPs → racks (with generators and bypass paths).

2. Choose your redundancy on purpose. Tie it to business impact, not ego. If the CFO can’t explain why you’re buying 2N, you’re buying arguments later.

3. Design for maintenance. Bypass paths, working space, clear labeling. If a tech can’t safely swap a breaker on a Tuesday afternoon, you designed a Saturday night problem.

4. Keep cooling in the picture. Electrical and mechanical are married. Treat them that way.

5. Measure. EPMS and branch-circuit monitoring turn opinions into facts.

6. Test like you mean it. IST doesn’t end when “the lights stayed on.” It ends when alarms, trends, transfers, and cooling all behaved as designed—and the team can repeat it.

FAQ (practical and short)

Isn’t a single big UPS simpler than two paths? It’s simpler until it isn’t. When you have to maintain it—or when it fails—you’ll wish you had an independent second path.

Can we skip generators if our utility is “reliable”? You can. But your downtime risk becomes tied to every storm, car-pole hit, and upstream breaker trip you don’t control. UPS covers short rides; generators cover long ones.

Do we need an arc-flash study for a small room? Size doesn’t matter. Fault current and clearing time do. If people open panels, they deserve to know the risk and wear the right PPE.

What’s the first thing to meter if budget is tight? Main feeders and PDUs/RPPs. That gives you load, imbalance, and basic power quality. Add branch monitoring later.

How much headroom is “enough”? I like day-1 at ~60–70% of major gear. Reality varies, but it’s a good starting rule while you gather real load data.

A last word (from the aisle, not a brochure)

The best compliment I can give a data center is that it’s boring. No surprises during a storm. No drama in maintenance windows. No mystery alarms at 2 a.m. Boring comes from clear one-lines, honest capacity planning, clean A/B paths, safe procedures, and monitoring that tells the truth.

If you’re planning a new build or a serious upgrade, bring your electrician, mechanical lead, and IT ops into the same room early. Draw the picture together. Fix the arguments on paper. And test until the picture matches the building.

Disclaimer: This article is general guidance. Always consult your AHJ and a licensed electrician or professional engineer for local requirements and sealed designs.