Every colocation negotiation covers the rate. Dirham or dollars per kilowatt-hour, per kilowatt of commitment, or some blend of the two. What gets far less scrutiny is the measurement underneath it: which device counts the energy, to what accuracy, how often it is read, and whether anyone other than the provider can check the number.
This is not a trivial detail. At one megawatt of continuous load, a one per cent measurement error is roughly 87,600 kilowatt-hours a year. The rate is negotiated down to fractions of a cent; the measurement that multiplies it is usually accepted without question.
Where the measurement actually happens
A facility has one utility meter at its boundary. Everything downstream of that is the provider's own instrumentation, and the architecture varies considerably between sites.
Branch circuit monitoring measures at the distribution panel or busway tap feeding a rack or cage. It is common, reasonably robust, and it captures everything downstream of that point.
Intelligent PDU metering measures at the rack power strip, sometimes per outlet. It offers the finest granularity and is useful operationally, but outlet-level instrumentation built into a PDU is designed primarily for monitoring rather than for revenue measurement.
Dedicated revenue-grade meters are installed specifically for billing, built to a recognised accuracy class, and calibrated. Where a tenant takes a large dedicated feed, this is the arrangement worth asking for.
Most facilities use a mixture. The question for a tenant is simply which one applies to their space, because the three are not equivalent for billing purposes.
Where error enters
Device accuracy class
Metering devices are built to stated accuracy classes, and the tolerance band differs meaningfully between a revenue-grade meter and general-purpose instrumentation. Neither is wrong; they are built for different jobs. The problem arises when monitoring instrumentation is used as the basis for an invoice without anyone acknowledging the distinction.
Current transformer sizing
Current transformers are most accurate across the upper part of their range. A CT sized generously for future growth, measuring a load well below its design point, sits in the least accurate part of its curve. This is a common and entirely unintentional source of drift, particularly in a space that was provisioned for expansion that has not yet happened.
Power factor and apparent versus real power
Billing should be based on real power in kilowatt-hours. Instrumentation that reports apparent power in kilovolt-amperes produces a different and generally higher figure. Confirm which quantity the invoice is derived from, and whether any power factor adjustment is applied.
Shared infrastructure allocation
Cooling, lighting, shared network equipment and losses in the distribution path are real costs that must be recovered. They are usually allocated through a facility overhead factor applied on top of measured IT load. That is a legitimate mechanism, but it is a calculation rather than a measurement, and the basis for it should be stated rather than embedded silently in a rate.
Reading interval and estimation
Where readings are taken monthly by hand rather than collected continuously, gaps get filled by estimation. Estimated periods followed by true-up adjustments are normal; what matters is that the invoice identifies which periods were estimated.
The questions worth asking before signing
- At what point in the distribution path is my consumption measured, and by what type of device?
- What accuracy class is that device, and when was it last calibrated?
- Is billing based on real power or apparent power?
- What facility overhead or loss factor is applied, and how is it derived?
- How often are readings taken, and are any periods estimated?
- Can I see my own interval data, and in what format?
- What is the process if I dispute a reading, and who bears the cost of verification?
None of these questions is adversarial. A provider operating well-maintained revenue-grade metering will answer all seven without difficulty, and the answers tell a tenant a great deal about how the facility is run.
A worked example of why the percentage matters
The arithmetic is worth doing once, because it reframes how much attention the measurement deserves relative to the rate.
A continuous one megawatt load consumes 8,760,000 kilowatt-hours in a year. A measurement error of one per cent is 87,600 kilowatt-hours. At any realistic commercial rate that is a material sum, and it arrives silently, spread across twelve invoices that each look unremarkable.
Set that against the negotiation. Teams will spend weeks moving the rate by a fraction of a cent, which is entirely rational. The point is simply that the multiplier deserves a fraction of the same attention, and usually receives none.
The same logic applies in reverse. A tenant who assumes the worst about metering accuracy without evidence is also mispricing the deal. The objective is not suspicion; it is knowing which of the two numbers in the calculation has been verified.
What "revenue grade" actually means
The phrase appears in sales material more often than it appears in specifications, and it is worth being precise about.
A revenue-grade meter is built and tested to a recognised accuracy class under a defined standard, and that classification is traceable to a test certificate for the specific unit or model. The relevant characteristics are the accuracy class itself, whether the device has been calibrated and when, and whether the current transformers feeding it are matched to the meter and correctly sized for the load.
A meter can carry an excellent accuracy class and still produce poor readings if its CTs are oversized for the actual load, because the device is only as good as the signal reaching it. This is why asking about the meter alone is insufficient; the installation is part of the measurement.
Monitoring instrumentation built into intelligent PDUs is genuinely useful for capacity planning, trend analysis and alerting. It is designed for those purposes. Using it as the basis for an invoice is a different application, and the honest answer from a provider is to say which type is in use rather than to describe everything as revenue grade.
How this interacts with the commercial structure
The significance of metering depends on how the contract charges for power. Under a pure committed-capacity model, measurement matters less because the charge is fixed. Under metered or hybrid models it matters a great deal. Our comparison of colocation pricing models sets out how these structures differ.
The gap between what is contracted and what is actually drawn is itself a recurring source of billing surprise, which we cover in contracted capacity versus actual draw.
The same principle applies to connectivity, where the measurement method is as consequential as the rate. Our explanation of 95th percentile bandwidth billing covers the equivalent question for network charges.
In the UAE specifically, the underlying tariff structure a facility buys on shapes what it can offer downstream. Our overview of DEWA and EWEC power tariffs explains the supply side of that relationship.
Verification in practice
Tenants with meaningful load should reconcile rather than simply pay. Compare invoiced consumption against your own equipment-level telemetry, which most modern servers and PDUs expose. The two figures will never match exactly, because the provider's measurement includes distribution losses and overhead that your servers do not see. What matters is whether the relationship between them stays stable month to month.
A ratio that drifts is the signal worth investigating. It may indicate a CT operating outside its accurate range, a configuration change, or simply a change in your own load profile, but a stable baseline makes an anomaly visible.
Where verification rights matter, they belong in the agreement alongside the other commercial protections discussed in our guide to deposits and guarantees in UAE colocation contracts, and in the wider colocation buyer's checklist.
Power is usually the largest line on a colocation invoice. It is worth understanding how the number is produced before agreeing to pay it for three years.