Data Center Power Pricing: Demand Charges and Overage
The real price of a data center watt: how energy, demand, and capacity charges stack up, which colocation pricing model hides costs, and what overage fees do to your bill.

On this page
The price of a data center watt is not on the price list. A colocation quote for a half rack can read like a single simple monthly number, but the power line on the first invoice is built from several charges that were never shown together, and the difference can be a third of the bill. Power is also the line item most likely to grow: every new server, every GPU node, every denser chassis draws more, while the rent for the floor space barely moves. Teams that cannot say what a kilowatt actually costs them are negotiating the most open ended line item in their infrastructure budget with no number under it.
This article assembles that number from the components that actually appear on real bills. It covers the two meters on every power bill, the kilowatt hour and the kilowatt, and why both are charged. It explains the demand charge, the fee for the worst 15 minutes of your month, and the ratchets that make one bad peak expensive for a year. It walks through the three colocation pricing models, flat rate, metered, and committed with overage, and shows what each one does to your cost at different utilization levels, including the overage fee that can double the price of the watt you exceed. By the end you will know the all in price of a watt and the clauses to check before you sign.
At a Glance: Power Pricing Models Compared
| Flat rate | Metered | Committed + overage | |
|---|---|---|---|
| Billing basis | Fixed monthly fee for a circuit | Actual kilowatt hours consumed | Fixed fee for committed kW, overage above it |
| Ideal for | Small, stable, low density loads | Variable or spiky workloads | Medium to large footprints, dense compute |
| Key strengths | Predictable, simple to budget | You pay for what you use | Aligns cost with the resource that constrains the building |
| Watch out for | Paying for headroom you never use | Minimum commitments and PUE multipliers | Peak based overage and automatic ratchets |
kW and kWh: The Two Meters on the Bill
A kilowatt hour is a unit of energy, the amount of work a steady 1 kW load does in an hour. A kilowatt is a rate, the pace at which you draw that energy in a single moment, not a quantity. Most people read power bills with only the energy line in mind, but commercial billing prices both, because each one pays for something different. The energy charge compensates the operator for the fuel and generation, while the demand and capacity charges pay for the infrastructure that had to be sized and kept ready to serve your worst moment, not merely your average.
Run the arithmetic once and the two meters separate cleanly. A rack drawing a steady 5 kW all month consumes 3,600 kWh in a 720 hour month, and at 12 cents per kWh the energy portion of the bill is about $432. Draw an average of 2.5 kW instead and the energy line halves to roughly $216, which is exactly what the volume meter says it should. But set a night batch job that momentarily pulls the rack to 7 kW, and the demand meter reads a peak of 7 kW, not the 2.5 kW average, and the demand charge bills on that peak. Energy rewards restraint across the month; demand rewards restraint in a single interval.
The Demand Charge: Peak Power Is Expensive
A demand charge prices capacity rather than volume. Utilities serving large customers capture the highest average demand in any measurement interval, commonly 15 minutes, and bill a rate per kilowatt of that peak across the whole billing period. The logic is straightforward to defend: transformers, feeders, breakers, and the substation capacity behind them are all sized for the worst interval, not for the average, and that infrastructure sits ready whether or not you ever approach it again. One 15 minute window where every cooling coil and blower runs flat out can set the peak that the entire month is billed against, and the other 2,800 intervals of the month cannot change it.
Many tariffs then add a demand ratchet, a clause that fixes a floor on future billing demand. The common structure bills you for no less than a percentage, typically 70 to 85 percent, of the highest peak metered in the previous 11 or 12 months. The result is that a single hot afternoon, one chiller event, one simultaneous startup, can raise the demand line for a full year, and your energy use can drop every month after the spike without the demand charge following it down.
The ratchet exists because the utility built capacity for your worst day and has to recover its cost whether that day repeats or not. The same logic appears on the colocation side when a contract’s overage clause raises your committed kW after a breach, so the mental model transfers directly. When a facility passes utility costs through on a demand basis rather than per measured kilowatt hour, your bill inherits the building’s worst 15 minutes, which is one more reason to ask exactly how power costs are passed through before you sign.
How Colocation Prices Power
Colocation prices power in three broad models, and each one shifts where the risk sits. Understanding which one you are on matters more than comparing the headline rack rate.
Flat rate, or unmetered, power is a fixed monthly fee for a circuit or an included kW allocation, whether you draw 1 kW or the full allowance. It is the simplest model to budget and the one most punishing to low utilization, because the provider has priced in worst case usage and your headroom is your expense. For a small, stable, well understood load it is often the best deal; for a rack that idles at 30 percent while paying for 100 percent, it is a quiet subsidy.
Metered power bills the actual kilowatt hours your equipment consumes, usually at a rate close to the local utility tariff, which sounds fair and often is. The fine print lives in the minimums and the multipliers. Most providers require you to pay for a floor of the breakered capacity, commonly 40 to 50 percent, even at zero draw, because the circuit is reserved for you. Just as important is the PUE multiplier: a contract that applies the facility’s power usage effectiveness to your metered consumption makes you pay for 14 kilowatt hours of delivered power for every 10 your servers draw at a PUE of 1.4. Inserting PUE between your meter and your bill can raise your effective energy rate by the facility’s entire efficiency story, so confirm whether the rate you are quoted is raw or PUE adjusted.
Committed plus overage, the model that dominates retail colocation above a few kilowatts, bills a monthly fee for a committed number of kilowatts, the capacity the facility reserves for you, and then prices anything above the commit separately, with energy sometimes included and sometimes metered on top. Power is the scarcest resource in a modern data center, more than floor space, so pricing it as a capacity commitment aligns the bill with what the facility actually has to provision for you. In North American markets, committed kW pricing runs very roughly from $120 to $250 per kW per month depending on market, density, term, and what is bundled, so the number is real but only comparable once you strip out what is included.
Most contracts also carry escalation: power rates tied to a consumer price index plus a margin, electricity pass throughs when the utility raises rates, and caps that decide how much of an increase you absorb. The rate on page one is rarely the rate in year three, so the escalation clause belongs in the comparison you make between providers.
Overage: The Watt That Costs Double
The overage clause is where the true price of a watt is written. When your draw climbs above the committed kW, the excess is billed at a premium, commonly one and a half to two times the contracted rate, because the facility had to keep capacity standing by that you did not reserve. The premium alone can double the cost of the watt you exceed, and most contracts compute it from the worst peak in the billing period rather than your average.
The metering interval makes the trigger easier or harder to hit. A contract that monitors a rolling five minute peak punishes spiky workloads: batch processing, log aggregation, a GPU cluster ramping a training run, a backup window that momentarily doubles the rack. A contract that measures monthly average draw tolerates the same spikes completely. This single clause explains most of the variation in power overage bills between facilities, and it is worth reading before you place a workload with a pulse.
The automatic ratchet is the expensive surprise buried beneath the premium rate. Some contracts state that when your peak overages exceed a threshold, commonly 10 percent of the commitment in a month, your committed kW steps up to a new, higher floor. The bill rises to reflect the new commitment and never comes back down, even after the workload that caused the spike is gone. One bursty month today can raise your capacity charge for the rest of the term.
Remember too that a committed breaker rating is never fully usable. The same electrical rule that caps a continuous load at 80 percent of a breaker applies at the rack, so the commitment you sign should be sized to the sustained draw you actually expect, plus a defensible margin, not to the sum of nameplate ratings, which can overstate the real figure several times.
What a Watt Really Costs
The capacity side of a committed contract has a brutal utilization lesson. Commit to 4 kW at $220 per kW per month and the capacity line alone is $880 before a single kilowatt hour is consumed, and even that $880 assumes energy is priced separately. If your equipment draws only 40 percent of the commitment, you are effectively paying $550 for each real kilowatt you use; fill the commitment and the effective price falls toward $232. The cheapest watt in a colocation deal is the one that fills the capacity you already paid for, which is why underutilized racks are not an efficiency problem so much as a pricing problem.
Put the whole chain together and the true cost of a watt has four layers. Energy is the fuel you consume, and its rate depends on the local utility market, so the same rack can cost dramatically different amounts in a 20 cent market versus a 12 cent market before anyone touches a breaker. Demand is the peak penalty the facility itself pays and that your per kW rate absorbs. Capacity is the committed kilowatt reservation that appears on your bill whether you use it or not. And facility overhead, the PUE multiplier, the metering interval, the minimums, and the escalation, is the layer most contracts render in small type. The number worth negotiating is the all in cost per kilowatt you actually use, expressed as one figure: total monthly power cost divided by real average draw.
Controlling the Price You Pay
Measured draw beats nameplate in every negotiation. A monitored PDU reading the real per rack load turns the contract conversation from an estimate into a fact, and it is the same number you need to size a commitment you will actually use.
Shape your peaks, because the clauses that hurt, the demand ratchet, the overage trigger, the automatic step up, all respond to the worst interval of the month rather than your average. Sequence rack ups, stagger batch jobs, and push background work out of the peak window, and every clause that keys off peak gets easier to live under.
Match the model to the load shape. A steady, full load workload is served well by flat or committed power. A workload with seasonal or nightly variability is better on metered power with a minimum commitment you can actually hit. A spiky workload needs the metering interval examined before anything else, because a five minute measurement window and a monthly average are different contracts in everything but the name.
Negotiate the clauses, not just the rate. The overage multiplier, the ratchet percentage, the automatic step up threshold, the PUE application, the escalation cap, and the length of the lookback are the terms that decide which provider is actually cheaper. Two quotes at the same per kW rate can differ by 20 percent or more in year two once the escalation and overage language are priced.
The real price of a watt is not a single number, it is the interaction of a volume meter, a peak meter, and a handful of contract clauses. The same rack can cost half as much again under one pricing model versus another at the same quoted rate, and the difference is settled before installation by the terms you sign and the utilization you achieve. Know your measured draw, understand your peak, read the overage and ratchet clauses, and compare providers on all in cost per used kilowatt rather than the headline number.
The audit is the first move. Pull a year of power invoices, divide each month’s total power cost by the kilowatts you actually used, and circle the month where that number climbs. The month will point at a peak, an overage, or an escalation, and naming the cause is what turns the next contract negotiation into a conversation you can win.
Frequently Asked Questions
How much does a kilowatt of power cost in a data center?
What is a demand charge?
What are power overage charges in colocation?
Is flat rate or metered power better for colocation?
Why is my data center power bill higher than quoted?
Stop reaching for a spreadsheet
Obelinf keeps every subnet, device, circuit, and rack in one live source of truth, with audit logs and a topology view. Free for personal use.
Related Articles

Planned Maintenance Windows in Data Centers: Scheduling Around Risk
How to schedule data center maintenance windows around risk: redundancy limits, maintenance modes, business cycles, third party windows, and the change process and checklist that keep planned work safe.
Read more
Backup Power in Data Centers: UPS, Generators, Fuel
How to size a data center UPS, load test diesel generators against NFPA 110, and plan fuel storage and refueling so backup power holds through the outages that matter.
Read more
Direct to Chip vs Rear Door vs Immersion Cooling
How direct to chip, rear door, and immersion liquid cooling compare on density, PUE, cost, and retrofit fit, and how to choose between them for your data center.
Read more