Blog · Metering Deep Dive
What ±1% Billing-Grade Metering Actually Means
Two PDUs can both say "metered" on the box; only one of them can carry an invoice. The difference is not the display or the protocol — it is a metering accuracy class, a calibration bench the unit actually sat on, and an interval data model your billing system can defend. This article walks through all three, from IEC 62053-21 class 1 to the 15-minute freeze value on a colocation invoice.
"Billing-grade" has no trademark and no enforcement — which is exactly why it is worth understanding what the claim should stand on before you put it in an RFP. In practice the term maps to one thing: metering accurate enough to bill against, in the accuracy class utility revenue meters use. Below we unpack the standard behind the number, the factory work that makes a specific unit meet it, and the data model that turns readings into invoices. If you are still deciding which metering tier your racks need, start with the selection guide and come back here for the depth.
The number behind the number: IEC 62053-21 class 1
IEC 62053-21 is the IEC standard for static (electronic) meters of active energy, and its class 1 designation bounds the measurement error at ±1% under the standard's reference conditions. That is the same family of accuracy utility companies use for revenue metering, which is why "billing-grade" and "revenue-grade" in the PDU world point at it. When a rack PDU claims ±1% metering with IEC 62053-21 class 1 as its basis, the claim is: this unit measures energy inside the error envelope a billing dispute would be judged against.
Two honest caveats belong in any serious datasheet. First, class accuracy is defined under reference conditions — a unit in a 40°C rack corner with distorted harmonic currents is working harder than a unit on a calibration bench, which is why temperature behaviour and harmonic handling matter in the design. Second, class 1 is an accuracy class, not a type-approval sticker: the PDU's metering chain should be engineered to the class and verified unit by unit, which is where the factory comes in.
How a PDU actually hits ±1%
The chain runs: sense — compute — calibrate. Current is sensed at each measurement point (shunts or current transformers feeding a dedicated metering IC), the IC computes active energy at high sample rate, and the firmware accumulates kWh registers and derives V, A, kW and power factor. The accuracy you get in your rack is decided at two of those three links: the quality of the sense elements on our own PCBA, and the calibration the unit receives before the housing closes.
Calibration is a factory station, not a certificate claim. On our line, every metering PDU is connected to a reference source bench, loaded across its range, and trimmed until the readings lock inside ±1%; the calibration record is kept against the serial number. Units then run 48-72 hours of full-load burn-in, which is where drifting components reveal themselves — a unit that cannot hold accuracy for three days under load does not get packed. This is why factory-direct matters for metering specifically: when the metering board and the calibration bench live under the same roof as the assembly line (see the factory tour), the ±1% on the datasheet and the ±1% in your rack are the same claim, verifiable per serial number through our quality process.
Input ±1% versus per-outlet ±1% A / ±2% kWh
Input metering watches one feed continuously; that is the easy case, and ±1% energy accuracy is achievable with margin. Per-outlet metering is harder: the sampling hardware is multiplexed across every socket, so a 48-outlet unit timeshares its measurement chain 48 ways every second. On our outlet-metered PDUs the resulting specification is per-outlet current at ±1% and per-outlet energy at ±2% — the extra point of energy error is the honest price of multiplexed sampling plus power-factor integration across a switching power supply load. Ask any vendor quoting flat ±1% at every outlet how they handle that integration; the answer is a good liquidity test for their datasheet.
The 15-minute freeze value: why billing runs on intervals
A kWh total answers "how much since the beginning of time"; it cannot answer "how much in March, and who says". Billing runs on interval data: every 15 minutes, the PDU freezes a copy of each billing register into a timestamped record that your DCIM or billing system reads over SNMP or REST. The 15-minute interval is the convention utility settlement uses, which means your PDU invoices speak the same time grammar as the utility invoice above them — no reconciliation weirdness at the boundary.
Interval data is also what resolves disputes. When a tenant contests a March invoice, you do not argue about a total; you replay 2,880 frozen records and look at exactly when the load moved. If the tenant added a GPU server on the 14th, the intervals show it. Because the freeze values are written whether or not anyone is polling, a two-day network outage reconstructs fully once the link returns — a monitoring gap, never a revenue gap. For the colocation-specific view of this data model, see the data center industry page.
Who needs billing grade — and who honestly does not
±1% metering costs money: sense elements, a metering IC, a calibration station, a reference source. Whether that cost earns itself back depends entirely on what the readings are used for. Six common cases:
| Use case | Metering tier needed | Why — the number that decides |
|---|---|---|
| Colocation tenant invoicing, shared racks | Per-outlet billing grade (±1% A / ±2% kWh) | Invoices per outlet group reconstruct from 15-min freeze values; disputes replay from logged intervals |
| Single-tenant racks billed per feed | Input billing grade (±1%) | Contract references the feed, so input class 1 accuracy is defensible without per-socket data |
| Enterprise capacity planning / PUE | Input ±1% metered | Global average PUE has plateaued at 1.56 for five years — sub-rack visibility is where the next percent hides |
| Mining / HPC per-rig P&L | Per-outlet kWh | Electricity is 75-85% of opex near a $0.07-0.08/kWh break-even; each metering percent is a margin percent |
| Edge site health checks | Metered input, switched recovery | Load trending guides capacity; the $600-1,200 truck-roll saving comes from switching, not billing accuracy |
| Lab / GxP audit environment | Per-outlet with exportable logs | Audit requires reconstructable records, not a live dashboard snapshot |
| Spare rack, test bench | Basic (no metering) | Nobody reads the data, so nobody should pay for the bench time |
Five questions to ask any metered-PDU vendor
Bring these to your next metering conversation. A factory that measures its own boards will enjoy them; a trading desk will not.
- Which accuracy standard does the metering claim reference, and which class?Want: IEC 62053-21 class 1
- Is every unit calibrated against a reference source, with a per-serial record?Want: yes, on request
- What is the per-outlet energy accuracy after multiplexing — and why?Honest: ±2%
- Are 15-minute freeze values held locally and readable over SNMP?Want: yes + OID list
- How long does interval history survive a network outage?Want: days, not minutes
Accuracy is a supply chain, not a slogan
±1% is easy to print and expensive to deliver, and the delivery shows up in three artifacts you can actually inspect: the accuracy class behind the claim, the per-unit calibration record against your serial range, and the interval data your billing system ingests. Any vendor who supplies all three has done the work; any vendor who supplies none has sold you a display. And when you shortlist suppliers, pair this with certificate diligence — our compliance explainer on UL 62368-1 gives you the checklist for that half of the purchase order.
Frequently asked questions
No, and the datasheet should say both numbers. Input metering samples one feed continuously and reaches ±1% energy accuracy. Per-outlet metering multiplexes sampling across 24-48 sockets, so current per outlet holds ±1% while per-outlet energy is specified at ±2%. Anyone quoting a flat ±1% at every outlet of a 48-socket unit is rounding their own spec sheet.
Only when one tenant occupies the whole rack and the contract is written against the feed — then ±1% billing-grade input metering is defensible. As soon as tenants share a rack, or a tenant adds equipment mid-cycle, the input number becomes an estimate. Shared racks need per-outlet kWh with interval freeze values so each invoice can be reconstructed from logged data.
The metering chain keeps running locally. Energy registers continue to accumulate and 15-minute freeze values continue to be written whether or not anyone is polling. When the network returns, your DCIM reads the accumulated registers and the buffered interval history, so the billing period reconstructs completely — an outage is a monitoring gap, not a revenue gap.
Sources: IEC — IEC 62053-21, electricity metering equipment (a.c.) particular requirements for static meters of active energy, class index 1 and 2 (iec.ch) · Uptime Institute, Global Data Center Survey (average PUE 1.56 plateau, uptimeinstitute.com) · Simple Mining, 2026 mining profitability analysis (electricity 75-85% of opex, $0.07-0.08/kWh break-even, simplemining.io) · Cloudatler, edge micro data center TCO handbook ($600-1,200 truck roll, cloudatler.com).
Want the calibration record with the shipment?
Every AmpRack metering PDU ships with a per-serial calibration trail. Tell us your rack plan and billing setup — an engineer replies with a datasheet and FOB quote within 48 hours.
