A data centre is the one commercial building type where the transformer is not really a cost item — it is a reliability decision. If you are specifying a dry type transformer for data center duty, three things drive the specification ahead of price: the redundancy architecture the facility is designed to (N, N+1 or 2N), the harmonic content of the load behind it, and the fact that a large part of that load arrives through a UPS rather than straight from the utility.
Get those three right and the rest — impedance, temperature rise, enclosure, sound level — follows logically. Get them wrong and you end up with a transformer that runs hot at 60% load, or a 2N design that collapses to a single point of failure at the bus tie. This guide covers the architecture, the harmonic arithmetic, and the sizing rules that actually apply.
Two reasons, and neither is about performance. The first is fire and life safety: a cast resin or VPI unit contains no flammable liquid, so it can sit inside an occupied building, on a technical floor or immediately adjacent to a UPS room, without NEC Article 450 vault construction, oil containment or fire separation. The second is maintenance: no oil sampling, no leak inspection, no liquid disposal. In a facility where an unplanned shutdown is measured in business interruption, removing a maintenance regime is a reliability gain, not just a cost saving.
The trade-off is honest. Dry-type units are physically larger for the same kVA, they need clean cool air, and they are noisier than an equivalent liquid-filled design — which matters when the electrical room shares a wall with a colocation cage.
Utility power arrives at medium voltage, typically 10 kV to 35 kV, and is transformed twice before it reaches a server. The first stage is the service entrance: a medium-voltage step-down outside the building, usually an outdoor liquid-filled or pad-mounted unit. The second stage is inside, where dry-type transformers step down to the voltage the UPS and PDU chain expects — commonly 400, 415, 480 or 600 V.
Data centre units in that interior stage typically run from 500 kVA to 3,750 kVA per unit, with larger blocks for hyperscale halls supplied as engineered unit substations. Integrating the MV incoming section, the transformer and the LV distribution into one coordinated assembly is what makes the redundancy scheme workable, and it is worth aligning the transformer package with the switchgear scope early — our switchgear and substation solutions overview sets out how that interface is usually drawn.
Redundancy is a property of the whole power path, not of the transformer alone. That said, the transformer is where the loading assumptions change most visibly between architectures.
| Architecture | Transformer arrangement | Typical normal loading | Harmonic rating that fits |
|---|---|---|---|
| N | One transformer carries the whole load | 70–85% | K-13 to K-20 |
| N+1 | One spare unit in the group; two of three active in a typical block scheme | 60–70% | K-7 to K-13 |
| 2N | Two fully independent transformer and UPS paths, each sized for 100% of critical load | 40–50% | K-4 to K-7 |
Notice how the loading falls as redundancy rises. This is the part that catches people out on efficiency: a 2N design means each transformer spends its life at 40–50% load, and transformer efficiency at part load is not the same as at nameplate. That is why loss capitalisation matters more in a 2N data centre than in almost any other application — you are buying two transformers to carry half a load each, twenty-four hours a day.
Two transformers do not add up to 2N if both later feed the same downstream switchgear section. Redundancy has to survive the bus tie, the maintenance bypass and the cross-connection. The same discipline applies to capacity: an N+1 group sized at the handover date can become N once load grows, so capacity and redundancy have to be reviewed together rather than signed off once.
Server switch-mode power supplies, UPS rectifiers and LED drivers do not draw a clean sine wave. They draw current in short pulses near the voltage peak, and that waveform is the fundamental plus a series of harmonics at 3rd, 5th, 7th and higher orders. The consequence for a transformer is not that RMS current rises — it is that eddy-current losses rise roughly with the square of the harmonic order. A 5th harmonic current causes about 25 times the eddy loss of the same magnitude at fundamental frequency.
A transformer sized on nameplate kVA alone can therefore run hot and lose insulation life while its RMS current never exceeds rating. IEEE C57.110 is the recommended practice that defines how much a transformer can carry under non-sinusoidal load, and it is the reference your protection and thermal studies should use.
The K-factor compresses the whole harmonic spectrum into one number, weighting each harmonic by both its magnitude and its order: K = Σ (Ih / I1)² × h². A perfectly linear load gives K = 1, which is what a standard distribution transformer effectively is.
| Rating | Non-linear load it can carry | Typical fit |
|---|---|---|
| K-1 | Linear loads only | Motors, resistance heating — not a data hall |
| K-4 | Up to about 35% | Lightly electronic panels; 2N paths where loading stays low |
| K-9 | 35–50% | Mixed commercial and light IT |
| K-13 | Up to about 75% | Dense office computing, UPS without input filtering, institutional IT |
| K-20 | Up to 100% | Data centres, server halls, PDU circuits, mission-critical UPS |
| K-30 and above | Extreme, known spectra | Special industrial duty; large, costly and rarely justified |
UL recognises the K-1, K-4, K-9, K-13, K-20, K-30, K-40 and K-50 steps, and the test requirements sit in UL 1561. Two practical notes. First, always round up — a calculated K of 5.9 means you specify K-9, and critical facilities often go one step further. Second, a higher K-rating is not free: K-rated units are larger, run lower efficiency at light load, and typically have lower impedance, which raises available fault current downstream.
The 3rd harmonic and its odd multiples — 9th, 15th, 21st — behave differently from the rest. On a four-wire wye system they are in phase across all three phases, so instead of cancelling at the neutral they add. In a panel dominated by single-phase electronic load, the neutral can carry more current than any phase conductor, theoretically up to about 173% of phase current in the worst case. That is why K-rated dry-type transformers are built with a neutral bus rated for at least 200% of full-load current, and why the neutral is often doubled from the transformer to the first distribution point.
Where harmonic reduction — not just harmonic tolerance — is the objective, a harmonic mitigating transformer is the tool. Designs that present a low impedance path to triplen and zero-sequence currents, or that deliberately phase-shift the secondary relative to the primary, cancel specific harmonics rather than simply survive them. A common arrangement is to feed two similar non-linear loads from transformers with different phase displacements — one delta-wye at 30°, one delta-delta at 0° — so that 5th and 7th harmonics cancel on the shared primary. This is the cleanest way to hold IEEE 519 limits at the point of common coupling without active filters.
The transformer feeding a UPS chain is not sized on IT load alone. Three additional components belong in the calculation.
Where transformers sit behind emergency generation, energisation matters. Transformer magnetising inrush can reach many times rated current for a few cycles, and a generator that has to absorb that inrush while also picking up the critical block may trip protection. The fix is usually a design one: specify a lower inrush design, stage the transfer, or size the generator with the transformer energisation case included. This is a coordination exercise between two scopes, and it is exactly the kind of interface that industrial power distribution and protection design exists to close.
| Parameter | Typical data center value | Why it is specified |
|---|---|---|
| Rated power | 500–3,750 kVA per unit | Matches UPS block size and failure domain strategy |
| Primary / secondary voltage | 10–35 kV / 400, 415, 480 or 600 V | Matches utility supply and UPS input |
| Insulation system | Cast resin or VPI, Class F or H | Indoor fire safety and thermal margin |
| Cooling | AN with AF available; AF typically adds around 40% capacity | Part-load efficiency normally, headroom in contingency |
| Harmonic rating | K-13 to K-20, higher on request | Matches the UPS and server load spectrum |
| Impedance | Around 6%, coordinated with switchgear | Balances fault current against voltage regulation |
| Neutral bus | 200% of full-load current | Triplen harmonic and single-phase load return |
| Temperature rise | Declared rise, often specified lower than the class limit | Insulation life under continuous high duty |
| Enclosure and IP rating | IP20 or IP23, matching the room | Clearance and dust protection |
| Sound level | Low-noise design per NEMA ST-20 or local limit | Occupied building, adjoining tenancies |
| Thermal monitoring | RTD / PT100 windings sensors with fan control | Feeds DCIM or BMS trending |
| Standards | IEC 60076-11 and IEEE C57.12.01; TIA-942 or tier scheme where applicable | Compliance and acceptance |
Two of these deserve a note before you send the RFQ. The physical envelope matters in a data hall — a 2,000 kVA cast resin unit is a substantial object to route into a technical floor, and the clearance and access path should be checked against the dry type transformer sizes and dimensions chart at design stage, not on delivery day. And the electrical detail behind the thermal and environmental classes is worth reading in full in our dry type transformer specification guide.
In a data centre, transformer losses hit twice: once on the electricity bill, and again on the cooling bill, because every watt lost as heat has to be removed by the HVAC system. That second effect is why loss capitalisation — valuing no-load and load losses over the equipment life using local energy cost, realistic load factor and a cooling multiplier — is standard practice in critical facility procurement.
It also explains why oversizing is not automatically good. A much larger transformer than required does give capacity headroom, but it carries higher no-load loss for the entire service life, whether or not the load ever materialises. On long evaluation horizons, a modular growth strategy using several right-sized blocks is often better than one very large unit, because it keeps insulation level, part-load efficiency and failure domain all in reasonable proportion. Where the loss figures drive the decision, the cost-structure discussion in our power transformer price guide is a useful cross-check on what you are actually paying the premium for.
For a data centre, start from the architecture and work backwards. Decide N, N+1 or 2N, which sets the normal loading and therefore the harmonic rating and the loss economics. Then size for the worst realistic hour — UPS recharge and generator transfer included — rather than for steady state. Specify the K-factor from a measured or expected spectrum rather than a default, oversize the neutral to 200% of full-load current if the load is single-phase electronic, and price the losses instead of only the equipment.
For units built to those classes with declared rises, the cast resin range in our SCB series dry type transformer data is a useful starting reference, and we are happy to work through a redundancy and harmonic study against your actual load profile.
Because they are installed indoors, close to the load and often next to occupied space. A dry-type unit has no flammable liquid, so it avoids vault construction, oil containment and fire separation, and it removes oil sampling and leak inspection from the maintenance regime.
K-20 is the normal specification for a dedicated data hall or PDU circuit at 100% non-linear load; K-13 suits dense office computing and UPS systems without input filtering. Where a 2N architecture keeps each unit at 40–50% loading, K-4 to K-7 may be adequate — but the rating should follow the measured or expected harmonic spectrum, not a habit.
Two fully independent transformer and UPS paths, each sized to carry 100% of the critical load, with each unit normally loaded at 40–50%. True 2N requires the independence to survive the bus tie, the maintenance bypass and any cross-connection.
UPS rectifiers and switch-mode supplies draw pulsed current containing odd harmonics. Winding eddy-current losses rise roughly with the square of harmonic order, so the transformer can overheat while RMS current stays within nameplate. IEEE C57.110 defines the loading limits, and the K-factor rating defines what the unit is built to withstand.
They do different jobs. A K-rated transformer is built to tolerate harmonic heating without derating. A harmonic mitigating transformer reduces the harmonics themselves, often by presenting a low-impedance path to triplen currents or by phase-shifting two secondaries so that 5th and 7th harmonics cancel on the shared primary. Choose based on whether the problem is transformer life or IEEE 519 limits at the point of common coupling.
Start with the critical IT load, add UPS conversion losses, then add the battery recharge demand that appears after a utility interruption. Apply the harmonic rating to the resulting current waveform, and check the generator energisation case if the transformer sits behind standby generation.
Usually yes. Continuous 24/7 duty leaves less room for thermal recovery than an intermittent commercial load, so a design with a declared rise below the class limit buys insulation life and overload headroom — at the cost of a larger, heavier and more expensive unit.
