Circuit-Level vs Panel-Level Monitoring in Data Centers

I have deployed both panel-level and circuit-level monitoring in data centers. I have watched both work. And I have watched panel-level monitoring miss degradation that circuit-level monitoring flagged weeks earlier. The distinction is not about having more data points. It's about whether the electrical signal you need survives aggregation.

Here is the practical difference, and when each approach is the right one.

What panel-level monitoring gives you

A panel-level meter monitors the main feed to a panelboard or PDU. It sees the aggregate of everything downstream: total current, total power, overall power factor, aggregate harmonic content.

That is useful for:

• Total load tracking per panel
• Capacity utilization percentage
• Load balancing across phases
• Billing and energy accounting
• Basic alarming (overcurrent, undervoltage)

Panel-level monitoring is the default for a reason. It gives you visibility into load distribution without the complexity of instrumenting every individual circuit. For capacity and billing purposes, it works well.

But here is the limitation I keep running into with customers: equipment degradation is a per-unit phenomenon. A UPS panelboard typically feeds 20 rectifier circuits. If one rectifier starts degrading, its harmonic profile shifts. At the panel level, that shift gets mixed with the other 19 circuits. The current signal gets diluted. In most cases, it disappears.

The math on electrical signal dilution

Let me put numbers on this because it matters for how you evaluate monitoring proposals.

Take a 20-circuit panelboard feeding 20 identical UPS rectifiers. One develops capacitor degradation, causing its 5th harmonic current to increase by 15%. That's a clear degradation signature at the individual circuit.

At the panel level, that 15% increase on one circuit out of 20 translates to less than a 1% change in the panel's aggregate 5th harmonic reading. That is within normal measurement noise. You will not see it. Your alarm will not fire. The rectifier keeps degrading.

At the circuit level, the 15% change is clearly visible. You see exactly which circuit changed, when it changed, and how that change compares to the other 19 identical circuits.

This is not theoretical. This is the math that determines whether you catch a degrading rectifier three weeks before failure or three hours before failure.

What circuit-level monitoring adds

Circuit-level monitoring places sensors on individual breakers within the panelboard. Each circuit is monitored independently: its own current, its own harmonic content, its own behavior over time.

You get everything panel-level provides, plus:

• Identification of which specific load is drawing abnormal current
• Degradation detection in individual equipment (specific UPS rectifiers, specific motors)
• Harmonic analysis per circuit, not aggregated across the panel
• Equipment-specific baseline comparison
• Fleet-wide statistical analysis across identical circuits

That last point is underappreciated. When you monitor hundreds of identical rectifiers at the circuit level, you can compare each one against the fleet. Statistical outliers become obvious. The rectifier that is drifting from the population norm stands out, even if its absolute readings are still within spec.

The electrical signal dilution problem in practice

Return to the math from the earlier section. A 15% harmonic shift on one circuit out of 20 produces less than a 1% change at the panel. That is within measurement noise. Now consider a fleet of 200 rectifiers across 10 panels. At the panel level, you have 10 aggregate readings. At the circuit level, you have 200 individual readings.

The per-circuit data lets you run fleet-wide statistical comparison. Each rectifier builds its own harmonic baseline over time. A unit whose 7th harmonic amplitude drifts above the 95th percentile of the fleet distribution is flagged, even if its absolute value remains within the manufacturer's specification. The anomaly is statistical, not threshold-based. Panel-level data does not support this analysis because the individual electrical signal is already lost in the aggregate.

Without circuit-level identification, the options for addressing degradation are inspecting the entire fleet (expensive, disruptive) or waiting for the degradation to progress to failure (worse). With it, maintenance targets the specific units that need attention.

When each approach is the right one

I'm not going to tell you that everyone needs circuit-level monitoring. That is not honest. Here is how I think about it.

Panel-level is sufficient when:

• Your primary use case is capacity tracking and billing
• Redundancy covers any single equipment failure, and you can tolerate the time to identify which unit failed
• Maintenance follows calendar-based schedules regardless of condition data
• The facility runs low-density, steady-state workloads with low downtime cost

Circuit-level becomes necessary when:

• Equipment reliability directly affects revenue or SLA (AI training, GPU compute)
• The fleet is large enough that individual inspections are impractical without targeting data
• You want condition-based maintenance rather than calendar-based
• Capacity validation requires per-circuit confidence, not panel-level averages
• The load profile is complex enough that aggregate measurements mask individual behavior

The cost question

Circuit-level requires more sensors. That is the trade-off. Instead of one meter per panel, you need sensors on individual circuits.

The Verdigris approach addresses this with EV2 hardware that uses split-core current transformers. Installation does not require de-energizing the panel or modifying existing wiring. Multiple circuits per panelboard are monitored from a single installation point.

The cost comparison is not sensors versus no sensors. It is the cost of per-circuit monitoring versus the cost of unplanned downtime from degradation that panel-level monitoring structurally does not resolve. For AI data centers, where a single UPS failure can interrupt training jobs costing thousands of GPU-hours, that comparison tends to resolve quickly.

Here's what I would ask your vendor

If you are evaluating monitoring solutions, ask: at what level of granularity does your system identify degrading equipment? If the answer is panel-level or PDU-level, ask the follow-up: how does a 15% harmonic shift on one circuit out of 20 surface in your alerting? If they cannot answer that specifically, you know which detection gap you are carrying.

Key takeaways

• Panel-level monitoring measures the aggregate of everything downstream. Circuit-level monitoring measures each load independently.
• A 15% harmonic shift on one of 20 circuits becomes ~0.75% at the panel meter, within typical measurement noise. The per-load electrical signal is lost in aggregation.
• Circuit-level data supports fleet-wide statistical comparison: each rectifier builds its own baseline, and outliers are flagged against the fleet distribution rather than against absolute thresholds.
• The cost comparison is not sensors versus no sensors. It is the cost of per-circuit monitoring versus the cost of unplanned downtime from degradation that panel-level monitoring structurally does not resolve.