What you pay for in a data-centre colocation bill

In a 2016 data-centre lease filed by Carbonite, the tenant could draw on 450 kilowatts of capacity from the start. Its first month’s base rent, however, was calculated using 100 kilowatts. The contract provided for rent to rise in stages, with faster increases if consumption over 30 consecutive days ran ahead of the schedule.[1]

The historical example exposes a distinction that matters when buying space for servers: available power, billable commitment and actual electricity consumption can be three different quantities. A price per kilowatt becomes meaningful when you know which one it multiplies.

Capacity is a promise; energy is a running total

In colocation, a customer places its hardware in a provider’s facility. ENERGY STAR describes charging arrangements that combine space and power capacity, as well as metered arrangements that charge for the customer’s electricity use and a share of the facility’s overhead. The charging model determines how much of a saving reaches the customer when its equipment uses less energy.[2]

A kilowatt, or kW, measures the rate at which equipment uses power. A kilowatt-hour, or kWh, measures energy accumulated over time. Equipment averaging 40 kW through a 30-day month uses 40 × 24 × 30 = 28,800 kWh. A 100 kW reservation provides headroom above that average; it does not mean the equipment consumed 100 kW continuously.

Read the unit on the capacity commitment carefully. Equinix’s power documentation defines its deployment’s contractual draw cap in kilovolt-amperes, or kVA, across the circuits together. The relationship between kVA and useful power in kW depends on the equipment’s power factor. Two redundant feeds also do not automatically entitle the customer to twice the contractual draw.[3]

For a prospective AI deployment, write down both the maximum simultaneous demand you need accommodated and the energy you expect to consume. Those are separate inputs to a quote. An average alone can conceal a peak that exceeds the agreed limit.

Where the cooling bill enters

Servers’ electricity is only part of a facility’s energy use. Cooling and power delivery add overhead. Power usage effectiveness, or PUE, expresses total facility energy divided by the energy delivered to IT equipment. A measured PUE of 1.30 means 1.30 units entered the facility for each unit delivered to the IT equipment, over the same measurement period.[2]

The billing rule still comes from the agreement. Carbonite’s filed lease, for example, made a cap on utility cost recovery conditional on operating practices including airflow containment and blanking empty rack positions. It specified a different shared-cost allocation when no tenant equipment was drawing power. Some commercial values were redacted, but the conditions remain visible. A facility’s efficiency figure alone would not tell a reader that tenant’s bill.[1]

The quote that wins can change with usage

Consider two invented monthly quotes for the same 100 kW reservation and service. These are a calculation exercise, not current provider prices. Quote A charges £100 per reserved kW per month, plus £0.15 per metered IT kWh multiplied by an agreed overhead factor of 1.30. Quote B charges £170 per reserved kW per month with energy included up to the agreed capacity. Assume a 30-day month, demand always within the limit, and equal omitted charges and taxes.

At 40 kW average load, A’s energy component is £5,616, leaving its total £1,384 below B. At 60 kW, the energy component becomes £8,424 and A costs £1,424 more. The break-even average is about 49.9 kW: £7,000 divided by 720 hours, £0.15 per kWh and the 1.30 multiplier.

That is why a useful comparison runs the same expected workload through both offers. A low-use month and a busy month reveal exposure that a single headline rate hides. Replace the invented assumptions with the actual minimum commitments, included allowances, energy rate and overhead calculation before using the method on a real quote.

Follow the rest of the invoice

Equinix’s invoice documentation separates space, power, interconnection and services such as Smart Hands. It distinguishes recurring charges from one-time work and prior-period adjustments. A connection or an engineer’s visit can therefore sit outside the line that first attracted your attention.[4]

Timing deserves its own check. The same documentation places planned price changes and billing ramps on the final page of the PDF invoice. New installations can also produce prorated or back-billed charges. Comparing a partial opening month with a full operating month can make a stable service appear to have become more expensive.[4]

Ask for a sample bill at the expected opening load and another after the deployment has grown, with the assumptions written beside each. Include the connections and hands-on support you intend to buy. A quote is most useful when you can explain how it turns into the next twelve invoices.

Report an error or suggest a correction