Rack power density is the amount of electrical power that the equipment in a single rack or cabinet draws, or is allowed to draw, usually expressed in kilowatts (kW) per rack. In a data center it describes how concentrated the power load is. In a colocation contract it is often the figure that limits how much equipment you can actually put in your space, because a cabinet can run out of power and cooling long before it runs out of room.
At a glance
- Measured in kilowatts per rack or cabinet, either as a contracted allowance or as measured draw.
- Nearly all the power IT equipment draws becomes heat, so supportable density depends on both power delivery and cooling.
- GPU and AI servers have pushed typical densities up sharply, and not every facility can support them.
- In colocation, density often drives cost and space planning more than square footage does.
What problem it solves
Buying data center space by floor area or cabinet count no longer tells you whether your equipment will fit. Modern servers pack more processing into each rack unit, and accelerated hardware such as GPU servers draws far more power than the general-purpose servers many facilities were designed around. A cabinet that is physically half empty may already be at its power or cooling limit.
Rack power density gives buyers and providers a shared number to plan against. It lets you check whether a facility can support your hardware, how many cabinets you will need, and what power commitment to sign. It also exposes a common source of waste: paying for cabinets you cannot fill because the power allocation runs out first.
How it works
Power delivery. A provider delivers one or more circuits to each cabinet, often a primary and a redundant circuit from separate power paths, and your equipment plugs into power distribution units (PDUs) mounted in the rack. The circuits’ rating sets a ceiling; providers commonly cap continuous load below that rating, in line with local electrical rules, so the usable figure is lower than the circuit’s nameplate. Where redundant circuits are provided, the load is usually planned so one circuit can carry it alone if the other fails.
Cooling. Almost all the power your equipment uses ends up as heat. Air cooling with hot-aisle and cold-aisle layouts works well up to a point; beyond it, facilities use measures such as aisle containment, rear-door heat exchangers or liquid cooling. The density a facility can support in practice is the lower of what its power and its cooling can handle at your location on the floor.
Contracting and billing. Colocation power is commonly sold per circuit, per committed kilowatt or as metered usage, sometimes in combination. Contracts may set a maximum density per cabinet and charge more for high-density space. How overhead such as cooling is priced varies; facility efficiency, measured by power usage effectiveness (PUE), can affect what you pay where overhead is passed through.
Measuring it. Your draw can be read from metered PDUs, from the provider’s monitoring or from the power data your servers report. Planning from equipment nameplate ratings tends to overstate real draw, while planning from today’s idle load can understate peak draw, so measured peak figures are the better basis.
For help sizing power and comparing facilities, see our Colocation solution page.
When it matters for buyers
- Deploying GPU or AI infrastructure. Accelerated servers may need densities that many existing facilities, or parts of them, cannot support. Confirm before you sign; buying capacity through GPU as a service is an alternative if you would rather not host the hardware.
- Hardware refreshes. New servers often draw more power per unit than the ones they replace, so a refresh can break an existing power allocation.
- First move into colocation. Sizing by space alone can lead to too few cabinets or too little power; size by measured kW.
- Contract renewals. Unused committed power and half-filled cabinets are common sources of waste worth renegotiating.
Questions to ask vendors
- What density can you support per cabinet in the specific space you are offering, and is that limited by power, cooling or both?
- How is power priced: per circuit, per committed kW or metered, and how is cooling overhead reflected?
- Are redundant circuits from separate power paths included, and how much load can we draw before losing redundancy?
- What cooling is available for high-density cabinets, and is liquid cooling supported or planned?
- Can we increase density later in the same space, and at what cost and lead time?
- How will we see our actual power draw, and how often is it reported?
How it differs from power usage effectiveness
Rack power density and power usage effectiveness are both power metrics but measure different things. Density is how much power the IT equipment in one rack uses; it tells you whether your hardware fits. PUE is the ratio of a whole facility’s energy to the energy its IT equipment uses; it tells you how much overhead the building adds. A facility can support high densities and still have a poor PUE, or the reverse.