Project snapshot
| Sector | Hyperscale data centre — cloud / colocation compute hall |
| Region | North America |
| Application | Server hall distribution under an N+1 redundant topology |
| Product | SCB13 cast resin dry-type (K-13 rated) / SCBH15 amorphous alloy |
| Rating | 2,500 kVA per unit × [ ] units |
| Voltage ratio | 13.8 kV / 480 V |
| Vector group | Dyn11 |
| Impedance voltage | 6 % |
| K-factor rating | K-13（UL 1561 / IEEE C57.110） |
| Insulation / rise | Class F (155 °C), 100 K average winding rise |
| Cooling | AN / AF |
| Enclosure | IP23 |
| Monitoring | PT100 winding sensors + RS485 to BMS |
| Typical load factor | 40–60 % of rated |
| Standards | IEC 60076-11, IEEE C57.110-2018, UL 1561 |
| Delivered | [Q · 20 ] |
The challenge
A hyperscale hall is an unusual electrical environment. Three characteristics drove this specification, and none of them is about raw capacity.
Every amp is non-sinusoidal. Server power supplies, UPS rectifiers and cooling drives all draw current in pulses rather than a smooth sinusoid. Harmonic currents heat a transformer in two distinct ways: I²R loss rises with total RMS current, and eddy-current loss rises with the square of the harmonic order — a fifth-harmonic component generates roughly 25 times the eddy loss of the same magnitude at fundamental frequency. A transformer designed only for 60 Hz sinusoidal load will run hot in this environment even when it is nowhere near its nameplate rating.
Fire safety governs placement. The units had to sit indoors, close to the load they serve. That rules out flammable insulating liquid: an oil-filled design would have required bundled containment, fire-rated separation and oil monitoring that an indoor electrical room was never going to accommodate.
Uptime is the product. The facility runs an N+1 topology, so the transformers were specified as redundant units rather than as a single bank sized to the load. And because the hall is energised 8,760 hours a year, no-load loss matters more here than on almost any other application — it is incurred whether the servers are busy or idle.
There is also a sizing subtlety: data centre transformers rarely run anywhere near full load. Typical load factors sit at 40–60 % of rated capacity, which happens to be where transformer efficiency peaks. Specifying for that operating band — rather than for a headline kVA figure — is what keeps lifetime energy cost down.
Our solution
K-13 rated cast resin construction
We supplied K-13 rated cast resin dry-type transformers. The K-factor is defined in IEEE C57.110 as the sum of (Ih/I1)² × h² across the harmonic spectrum, with UL 1561 recognising discrete ratings of 1, 4, 9, 13, 20, 30, 40 and 50. K-13 covers installations where 50–100 % of the load generates harmonics — dense computing, UPS without input filtering, electronically controlled lighting.
Two things are worth stating plainly. First, a K-rating is a heat-survival rating, not a filter: it does not reduce the harmonic content the load draws, it ensures the transformer stays within its temperature rise limit while carrying it. Second, the rating is only correct if it matches the measured spectrum — which is why we ask for harmonic data at specification stage rather than assuming a default.
Cast resin construction handles the fire-safety side: windings vacuum-cast in epoxy, self-extinguishing and containing no oil. See our comparison of oil-immersed vs dry-type transformers for the full trade-off, and our cast resin dry-type transformers for data centers page for the application-level reasoning.
Amorphous alloy cores for the 24/7 no-load penalty
For units on continuous duty we supplied SCBH15 amorphous alloy units alongside the SCB13 design. Amorphous alloy has no crystal structure, roughly 3–6 times the resistivity of silicon steel, and unit eddy-current loss of only about 20–30 % of silicon steel. The practical result, per the manufacturer’s published data, is no-load loss roughly 70 % below a conventional 10-series dry-type design.
That figure is worth converting. At the 1,000 kVA reference point, published comparisons put amorphous no-load loss near 800 W against roughly 1,800 W for a comparable silicon-steel design — a difference of about 1,000 W that accrues for 8,760 hours a year, or roughly 8,760 kWh per year per unit, regardless of how busy the hall is. Amorphous cores carry a higher first cost (commonly quoted at 30–50 % above a comparable silicon-steel unit), with payback typically inside three to five years at commercial electricity rates.
Sized for N+1, not for nameplate
Under an N+1 topology the capacity question is not “how much load” but “how much load with one unit down”. A hall requiring 2,000 kVA would be served by three 1,000 kVA units — two carrying the load, one standing hot as a spare that picks up automatically through transfer switching. That is why this project was supplied as multiple 2,500 kVA units rather than a single larger bank: the rating of each unit is set by the redundancy case, not by the steady-state case.
AN / AF cooling was specified so that forced air gives roughly 40 % additional capacity for peak and failover conditions, rather than being needed continuously.
Thermal monitoring wired into the BMS
PT100 sensors in the windings feed a temperature controller with alarm and trip contacts, exposed over RS485 to the building management system. For a facility that cannot dispatch someone to read a dial, winding temperature has to arrive as data. Fan control, alarm escalation and protective tripping all run off the same signal.
Technical specification
| Parameter | Value |
|---|---|
| Rated power | 2,500 kVA |
| Primary / secondary | 13.8 kV / 480 V |
| Frequency | 60 Hz |
| Vector group | Dyn11 |
| Impedance voltage | 6 % |
| K-factor rating | K-13（UL 1561 / IEEE C57.110-2018） |
| Insulation class | F (155 °C) |
| Cooling | AN / AF |
| Enclosure | IP23 |
| Monitoring | PT100 + RS485 |
| Measured THD (if available) | 20 % |
| Standards | IEC 60076-11, IEEE C57.110-2018, UL 1561 |
Testing and verification
Every unit passed routine tests per IEC 60076 before release: winding resistance, voltage ratio and vector group, no-load loss and current, load loss and impedance voltage, applied and induced voltage withstand, and insulation resistance. Dry-type specific requirements were met per IEC 60076-11.
K-13 capability was established against IEEE C57.110, the recommended practice for transformer capability under non-sinusoidal load current. Temperature rise was verified within the Class F limit of 100 K average winding rise / 155 °C hot-spot, measured at the IEC reference ambient of 40 °C maximum.
Our guide to transformer routine testing sets out what to expect before shipment, and transformer cooling classes explains how AN and AF ratings differ.
Result
Fire-safe indoor distribution with harmonic tolerance matched to the measured load, redundant capacity that holds the hall up with a unit down, and no-load loss cut to the level a 24/7 facility actually needs. Winding temperature arrives at the BMS as data rather than as a surprise.
Planning a hyperscale, colocation or AI compute hall? Review our cast resin dry-type transformers for data centers, or send us your load profile and harmonic data for a proposal.
Frequently asked questions
What is a K-factor rating, and is K-13 or K-20 right for my data centre?
The K-factor quantifies how much additional heating a transformer can dissipate from harmonic currents. It is defined in IEEE C57.110 as the sum of (Ih/I1)² × h² over the harmonic spectrum, and UL 1561 recognises ratings of 1, 4, 9, 13, 20, 30, 40 and 50. K-13 suits installations where 50–100 % of the load generates harmonics — dense computing, UPS without input filtering. K-20 is specified where 75–100 % of the load is non-linear and the transformer is more heavily loaded, which is how several manufacturers classify dedicated server rooms. The honest answer is that the rating should follow a measured harmonic spectrum, not a rule of thumb — if you have power quality data, that is what determines the specification.
Does a K-rated transformer reduce harmonics?
No. A K-rating is a heat-survival rating, not a filter. It does not change the harmonic content your load draws; it ensures the transformer stays within its temperature rise limit while carrying it. Achieving that means more core and coil material, subdivided conductors to limit skin effect, a neutral rated at up to 200 % of phase conductor capacity for triplen harmonics, and an electrostatic shield. If you need the harmonic content itself reduced, that is a harmonic mitigating transformer — a different product.
Why dry-type for a data centre rather than oil-filled?
Indoor placement. Data centre electrical rooms sit close to the load, and an oil-filled unit brings bundled containment, fire-rated separation, oil monitoring and disposal with it. Cast resin windings are self-extinguishing and contain no flammable liquid, so the units can be placed where they are actually needed. Cast resin also tolerates indoor humidity better and runs 100 % humidity without re-drying.
Is an amorphous alloy core worth the extra cost?
It depends on energised hours and load factor, not on capacity. Amorphous alloy cuts no-load loss by roughly 70 % against a conventional 10-series dry-type design — and no-load loss is incurred every hour the transformer is energised, whether the hall is busy or idle. That is why it pays off on 24/7 duty and much less on intermittent loads. First cost runs 30–50 % higher, with payback typically inside three to five years. If your load factor is very low or the unit is rarely energised, a silicon-steel design is usually the better value.