Three numbers, three different answers
Take one rack of mixed equipment and ask what it draws. The labels on the back add up to 9.2 kW. The PDU average over a busy hour is 4.6. The tallest minute in the log is 5.8. Every one of those is a true statement about the same rack, and choosing between them decides how much switchgear gets bought.
Run the defaults on this page and the three answers separate cleanly. Against a 64.2 kW feed figure, the metered average fits 13 racks, the peak sample fits 11, and the nameplate sum fits 6. That is a factor of two between the first and the last, on identical hardware, with nothing but the definition changed.
Where the feed figure comes from
A 225 A breaker at 208 V three-phase carries 81.1 kVA on the arithmetic — volts times amps times the square root of three, divided by a thousand. At the 0.99 power factor metered on the panel that is 80.2 kW, and at the 80 percent continuous-load allowance the engineer is working to, 64.2 kW. Every one of those three steps is a number somebody supplied. This page does not know what continuous-load allowance applies to your installation, and it will not guess: the code your authority has adopted, the conductor, the terminations and the ambient all sit inside that percentage, and the person who signs the load calculation owns it.
Watts per square foot is a denominator argument
The eight default racks draw 36.8 kW. Over the 210 sq ft the form counts — cabinets plus their share of both aisles — that is 175 W per square foot. Count only the cabinet footprints, six square feet each, and the same row reads 767. Count the whole room including the walkway to the door and it drops under a hundred.
None of those is wrong and none of them are comparable. When a specification sheet, a lease or a colocation quote names a watts per square foot figure, it is worth a phone call to find out what went in the denominator, because the number moves by a factor of five before anybody has changed a server.
What is left, and what "left" is not
Eight racks at 4.6 kW leave 27.4 kW against the 64.2 kW figure, which is five more racks at the same draw. That is a subtraction, not a finding. It says nothing about whether the conductors, the terminations, the ambient and the harmonic content let the feed run at that load for years, and it says nothing about what happens when the second feed takes over — a redundant path is a claim that has to be proved by transferring the load with the room live, not by adding two numbers.
The one thing the subtraction is genuinely good for is spotting stranded capacity. A row provisioned on nameplates and running at meter typically sits at half the power it was built for, with the difference locked up behind breakers that will never trip. That gap is the cheapest floor space in the building, and finding it takes a PDU log rather than a purchase order.
Questions people ask
How many servers can I put on a 30 amp circuit?
This page will not answer that, and neither should any page. What it does is the arithmetic in the open: a 30 A single-phase feed at 208 V is 6.24 kVA, and whatever continuous-load allowance and power factor you enter cuts it from there. Divide by the draw you metered and you get a count. Whether that circuit actually carries it is a load calculation against the code your authority has adopted, done by somebody who has seen the panel.
Should I size from nameplate or from metered draw?
Meter, and then decide what margin you want on top. Nameplate is the figure the internal supply is designed never to exceed, so it sits far above service draw — in the defaults here, 9.2 kW of labels against 4.6 kW measured. Sizing everything from nameplate is how rooms end up half empty with no capacity left on paper. Sizing from a single quiet hour of meter data is how they trip during a backup window. The honest input is a week of logging that includes a month end.
What is a normal kW per rack?
There is no normal, and the published averages are not much use because they mix a row of network switches drawing under a kilowatt with a row of GPU chassis drawing thirty. The number that matters is yours, from your own PDUs, per rack rather than averaged across the room — the room average hides the two cabinets that are actually the problem.
Why does the peak column matter if the average fits?
Because breakers respond to current, not to averages. A row that sits comfortably at 4.6 kW a rack and spikes to 5.8 during a nightly backup is drawing 26 percent more for the length of that window, and the count of racks the feed figure covers drops from 13 to 11. If your PDU keeps a minute-by-minute history, the tallest sample is the input worth arguing about.
Does a second feed double the capacity?
Not in any sense this page will endorse. Two feeds are two feeds; whether either one carries the whole row when the other is gone depends on how the load is split, on what the transfer equipment does, and on somebody having tested it under load. This calculator adds and divides. Redundancy is a design and a commissioning result, not an arithmetic one.