Rack Circuit Capacity Calculator
Turn a rack's feeds into the number that decides what you can install: usable kW after redundancy and the 80% continuous-load derate, plus how many servers of a given class that budget supports.
Per circuit
14.49
kVA · 13.77 kW
Usable capacity
14.49
kVA · 13.77 kW · 1 of 2 circuits
After derate
11.01
kW continuous · 11.59 kVA
Servers that fit
1
at 10.20 kW each · 0.81 kW spare
Single-phase kW = V × A × PF ÷ 1000. Three-phase kW = V × A × √3 × PF ÷ 1000, using the line-to-line voltage. Power factor defaults to 0.95; drop it from either formula to get kVA.
How a Rack Feed Becomes a kW Budget
A rack circuit is described by three numbers: its voltage, its breaker rating in amps, and whether it is single-phase or three-phase. Multiply the first two and you get apparent power in volt-amps, which is what the breaker responds to. For a single-phase circuit that is simply V × A ÷ 1000 kVA. A three-phase circuit carries current on three conductors, so the same arithmetic is multiplied by √3 (1.732) when the voltage quoted is line-to-line, which is how three-phase circuits are normally specified.
Apparent power is not the same as the real power your equipment consumes. Real power in kilowatts is apparent power multiplied by the power factor, the fraction of the current that does useful work rather than circulating as reactive current. Server power supplies with active power factor correction sit between 0.95 and 0.99, which is why 0.95 is the default here. A rack quoted at 14.49 kVA therefore delivers about 13.77 kW of real power, and the difference between those two numbers is why "kVA" and "kW" cannot be used interchangeably in a capacity plan.
The last adjustment is the continuous-load derate. Electrical codes treat any load running for three hours or more as continuous and require the circuit to be sized at 125% of it, which in practice means loading a breaker to no more than 80% of its rating. A 30A circuit is a 24A circuit for planning purposes. The derating field above defaults to 80% for that reason, and you can change it if your local code or your facility's own standard differs.
Order matters. Calculate apparent power per circuit first, then decide how many circuits are genuinely available under your redundancy scheme, then apply the power factor and the derate. Applying the derate before redundancy produces the same number here, but applying redundancy after you have already committed equipment is how racks end up over-provisioned on paper and fragile in practice.
What N, N+1 and 2N Do to Usable Capacity
Redundancy decides how much of the installed capacity you are allowed to spend. The circuits that are not spending capacity are still installed, still metered, and still counted by anyone reading the panel schedule instead of the design intent.
| Scheme | Circuits that carry load | What survives a failure |
|---|---|---|
| N | All of them | Nothing. Losing a circuit drops whatever it fed. |
| N+1 | All but one | Any single circuit failure, with the spare picking up its share. |
| 2N | Half of them | An entire distribution path, because the mirror side carries the rack alone. |
The 2N row is where most capacity plans go wrong. Two 32A feeds into a rack give you one 32A feed's worth of load with a second one standing by. Treating them as 64A works right up until an upstream breaker, a UPS module or a maintenance window takes one side away, at which point the surviving side sees twice its design load and opens. A dual-corded server plugged into both sides behaves the same way: each supply has to be able to run the machine by itself.
Two Real Questions, Worked Through
Three pairs of 210V 30A single-phase primary/redundant circuits
- Per circuit: 210 V × 30 A ÷ 1000 = 6.30 kVA.
- Six circuits installed: 6 × 6.30 = 37.80 kVA on the panel.
- Three pairs, primary and redundant, is 2N. Half the circuits carry load: 3 × 6.30 = 18.90 kVA usable.
- Real power at 0.95 power factor: 18.90 × 0.95 ≈ 17.95 kW.
- At the 80% continuous derate: 17.95 × 0.8 ≈ 14.36 kW to commit.
The rack is often quoted as a 37.8 kVA rack. It is a 14.36 kW rack once redundancy and the derate are honoured, and the 18.9 kVA difference is the price of surviving a feed failure.
An H100 rack on 230V single-phase 63A
- Apparent power: 230 V × 63 A ÷ 1000 = 14.49 kVA.
- Real power at 0.95 power factor: 14.49 × 0.95 = 13.77 kW, the ~13.7 kW figure this feed is usually quoted at.
- At the 80% continuous derate: 13.77 × 0.8 = 11.01 kW.
- An 8-GPU H100 or H200 SXM5 node draws up to 10.2 kW per 4U chassis, so a single derated 63A feed supports one node, with 0.81 kW left for the switch and the PDU's own overhead.
This is why dense GPU racks move to three-phase. The same 63A at 400V three-phase is 43.65 kVA, 41.47 kW at 0.95, and 33.17 kW after the derate: three nodes instead of one, on the same breaker rating.
The 10.2 kW per 4U chassis figure is the one used in our liquid cooling monitoring guide. For anything else, measure the machine or enter the vendor's figure rather than trusting a nameplate PSU rating, which commonly overstates real draw.
What This Calculator Assumes
Every result above rests on inputs you can change and on three assumptions you should check against your own site before you commit equipment to a rack.
Unbalanced three-phase load is the assumption worth checking first. The √3 formula describes a circuit whose three phases carry roughly equal current. Single-phase outlets distributed unevenly across a three-phase rack PDU can load one phase far harder than the others, so the rack trips on one pole while the calculated total still looks comfortable. Per-phase current readings from the PDU are the only way to see that.
Rack Circuit Capacity Questions
How many kVA does a rack with 3 pairs of 210V 30A single-phase redundant circuits support?
Each circuit is 210 × 30 ÷ 1000 = 6.3 kVA, so six circuits represent 37.8 kVA installed. Three pairs in a primary/redundant (2N) arrangement means only one circuit of each pair carries load, giving 18.9 kVA usable, about 17.95 kW at a 0.95 power factor. Derated to 80% for continuous load, plan to roughly 14.36 kW.
How do you calculate kW from volts and amps for a rack circuit?
For single-phase, kW = volts × amps × power factor ÷ 1000. For three-phase, multiply by √3 (1.732): kW = volts × amps × 1.732 × power factor ÷ 1000, using the line-to-line voltage. Drop the power factor from either formula and you get kVA, the apparent power the breaker actually sees.
How much power can a 230V 63A single-phase rack circuit deliver?
230 × 63 = 14,490 VA, or 14.49 kVA. At a 0.95 power factor that is 13.77 kW of real power, which is the figure people are looking for when they ask what a 63A GPU rack feed is worth. Applying the 80% continuous-load derate leaves about 11.01 kW to commit.
What is the difference between N, N+1 and 2N rack power redundancy?
N means every installed circuit carries load and nothing is held in reserve. N+1 keeps one spare circuit so the rack survives losing any single one. 2N mirrors the whole feed, so half the installed circuits are standby. Usable capacity is N circuits, N−1 circuits, and half the circuits respectively.
Why doesn't 2N redundancy double a rack's usable capacity?
Under 2N each side has to carry the entire rack on its own, because the point of the second side is to take over when the first fails. Installing twice the circuits buys availability rather than headroom. Sizing a rack to the installed total is how a rack that looks healthy trips the moment one feed drops.
What power factor should I assume for rack power calculations?
Server power supplies with active power factor correction typically run between 0.95 and 0.99, so 0.95 is a conservative planning default and the one used here. Use 1.0 only if the equipment is genuinely unity power factor, and use a measured value from the PDU when you have one.
What is the 80% continuous-load rule?
Electrical codes derate a breaker for loads that run for three hours or more, which covers essentially everything in a data center rack. A 30A circuit should therefore carry no more than 24A continuously. Planning to the full breaker rating is the most common reason a rack trips under sustained load rather than at install time.
How many GPU servers fit on a rack circuit?
Divide the derated usable kW by the per-server draw. An 8-GPU H100 or H200 node at 10.2 kW needs more than a single 230V 63A feed once the 80% derate is applied, so dense GPU racks are normally fed by three-phase circuits or several circuits per rack.
What a Calculator Can and Cannot Tell You
This page estimates. It takes the numbers printed on a panel schedule and a vendor datasheet and turns them into a planning budget. It cannot tell you what the rack is drawing right now, whether the three phases are balanced, or whether a server is running 40% below the figure you entered.
Sensaka DCOS measures those things in production. It reads per-outlet and per-phase current from intelligent rack PDUs over SNMP, and per-server power draw from each machine's BMC over IPMI and Redfish, so the same rack you planned here reports what it actually consumes. The gap between the two is usually large enough to matter: teams that commit capacity against nameplate ratings tend to leave a third of a rack's usable kW stranded.
For the inventory side of the same problem: which machine sits in which U of which rack, on which circuit, and what changed since the last audit, see data center asset management.
Related calculators and guides
Also useful: Rack Unit Calculator, What Is a PDU?, and PUE Calculator.