Tools / Power

What does 10MW mean?

Turn a megawatt figure into energy, racks and accelerator counts. Every assumption is shown and you can change it.

Illustrative arithmetic, not a measurement. The numbers below follow from the assumptions on this page. The assumptions are our own round figures for illustration. They are not sourced facts about any real site, chip or operator, and real sites differ widely.

Your figure and assumptions

Figure in MW

From 0.01 to 100,000.

What does that figure describe?

IT load is the power reaching the computing equipment. Facility load adds cooling and losses. A grid connection is usually quoted in MVA.

Assumptions you can edit

PUE (facility load divided by IT load)

Assumed 1.3. Allowed 1 to 3.

Power factor (used to move between MW and MVA)

Assumed 0.95. Allowed 0.5 to 1.

Average load as a share of facility load (%)

Assumed 80. Used for annual energy. Allowed 1 to 100.

kW of IT load per rack

Assumed 40. Dense AI racks can be far higher, older racks far lower.

Share of IT load that is accelerator servers (%)

Assumed 70. The rest is storage, networking and general servers.

Accelerators per server

Assumed 8. Whole number, 1 to 72.

Lowest kW per accelerator server

Assumed 6. Gives the high end of the GPU range.

Highest kW per accelerator server

Assumed 14. Gives the low end of the GPU range.

Reset

What 10MW of it load comes to

Illustrative arithmetic, not a measurement
What	Result	How it was worked out
IT load	10 MW	Your figure.
Facility load	13 MW	IT load multiplied by PUE (1.3).
Overhead (cooling, power losses, lighting)	3 MW	Facility load minus IT load.
Equivalent in MVA at the assumed power factor	13.7 MVA	Facility load divided by power factor (0.95).
Energy in a year	91.1 GWh	Facility load x 80% average load x 8,760 hours.
Racks	250	IT load divided by 40 kW per rack.
Accelerator servers	500 to 1,167	70% of IT load, divided by 14 kW and 6 kW per server.
Accelerators (GPUs or similar)	4,000 to 9,333	Servers x 8 accelerators per server.

Ranges come from the two kW per server assumptions. Nothing here adds capacity across different metrics or sites: it is one figure worked through one set of assumptions.

Link to this result

https://computefinder.net/tools/what-does-100mw-mean?mw=10&metric=it&pue=1.3&pf=0.95&load=80&kwrack=40&srvmin=6&srvmax=14&gps=8&share=70

The link holds your figure and every assumption, so anyone opening it sees the same arithmetic.

MW and MVA in plain English

A megawatt (MW) measures real power: the part that does useful work, such as running servers and cooling. A megavolt-ampere (MVA) measures apparent power: the total the wires and transformers have to carry. The two differ by the power factor, so MW is MVA multiplied by the power factor. At a power factor of 0.95, a 100MVA connection carries about 95MW of real power.

Grid connections and transformers are usually sized in MVA, while data centre loads are usually described in MW. When someone says "a 100MW data centre", ask which they mean: IT load, facility load or a grid connection, and whether the figure is MW or MVA. They can differ by a third or more. This is a general explanation, not a statement about any named project.