Verdigris vs Packet Power, Comparison

Illustrative example. Verdigris values (PNL-3-B, 412 A, +31% headroom) shown for shape; real deployments produce site-specific limits from 8 kHz waveform evidence over the measurement window.

Key takeaways

Both Verdigris and Packet Power deploy non-invasively on existing circuits, with no shutdowns.
Packet Power measures power and consumption (V, A, W, kWh, PF, Hz) at standard polling rates. Verdigris samples voltage and current waveforms at 8,000 Hz, applies ML models trained on failure modes, and returns validated operating limits.
The sampling-rate gap is multiple orders of magnitude. At standard polling, GPU power transients (8 ms ramps), early-stage rectifier capacitor degradation, and sub-cycle arcing precursors are invisible by design.
At T-Mobile, Verdigris detected equipment degradation in 4% of more than 800 monitored rectifiers, 21 days before failure. Zero standard BMS alarms had fired on those units.
Choose Packet Power for low-cost consumption visibility. Choose Verdigris for AI workloads, predictive maintenance on UPS infrastructure, and validated operating limits for capacity planning. The two can coexist.

Verdigris vs Packet Power: how the two platforms compare

Packet Power is a wireless monitoring platform with a portfolio of branch-circuit, PDU, DC busbar, and environmental sensors that report power and consumption metrics (V, A, W, kWh, PF, Hz) at standard polling rates. Verdigris is an electrical intelligence platform that samples voltage and current waveforms at 8,000 Hz and applies ML models to detect equipment degradation before it causes outages. Both products are non-invasive. Both clamp on existing circuits without downtime. The differentiation is what happens after measurement.

Both vendors share the same deployment story: wireless or clamp-on sensors, no shutdowns, no rewiring, fast time to first reading. That parity is real. What differs is the sampling rate, the kind of data captured, and the analytics applied on top.

Packet Power measures consumption at standard polling rates. Verdigris samples waveforms 8,000 times per second and applies ML models trained on failure modes like rectifier degradation, capacitor wear, and harmonic shifts. The price difference is the validated layer.

Capability comparison

Capability	Packet Power	Verdigris
Sampling/reporting rate	Standard polling (Packet Power does not publish a native sample rate)	8,000 Hz
Data type	Power and consumption metrics (V, A, W, kWh, PF, Hz)	Voltage and current waveforms
Analytics method	Threshold rules and basic anomaly detection on aggregated metrics	ML on waveforms
Waveform-physics predictive diagnostics	No	Yes
Failure mode detection	No	Rectifier degradation, capacitor wear, harmonic shifts
Deployment	Non-invasive wireless sensors	Non-invasive CT clamps
Hardware portfolio	Wireless sensor portfolio (PDU, branch circuit, environmental)	EV2 sensor (waveform-class)
Vendor neutrality	Any PDU	Any UPS, PDU, switchgear
Cloud platform	Yes	Yes
Price tier	Entry	Premium

Headers use Title Case. The Verdigris column is read last, because that is the differentiated value column.

The intelligence Verdigris adds

Electrical intelligence is a category of monitoring that converts high-frequency waveform samples into validated operating limits, predictive diagnostics, and early warnings about equipment health. Consumption monitoring is a category of monitoring that reports how much power was used. The two are not interchangeable. The architectural gap between waveform-class sampling at 8 kHz and standard polling is not incremental. At standard polling, you see the average current over a multi-second window. At 8 kHz, you see thousands of individual samples inside that same window: every distortion, every harmonic, every sub-cycle event.

That gap matters because equipment degradation lives in the waveform, not the average. A consumption reading tells you a rectifier drew 12.4 kW last second. A waveform tells you the 5th harmonic in that current is 18% above baseline, that the inrush profile is slewing 30 milliseconds slower than it did six months ago, and that the capacitor bank feeding it is on a degradation curve. Individual Harmonic Distortion and Total Harmonic Distortion are the entity-level signatures consumption meters do not resolve.

Three specific examples of what consumption-rate measurement misses by design:

GPU power transients. Modern GPU racks ramp in 8 milliseconds during training step transitions. The NVIDIA H100 datasheet documents 700 W per-card draw with sub-millisecond response windows during workload changes. At standard polling rates, the entire transient happens between two readings. You see the steady-state average, not the spike. At 8 kHz, you see the ramp shape.

UPS rectifier capacitor degradation. Failing capacitors do not pull more power. They distort the input current waveform. The 5th and 7th harmonic content rises before the capacitance value moves, a signature documented in IEEE 519-2022 as the canonical indicator of degraded power-electronic switching. The companion standard IEEE 1459-2010 defines the power-measurement quantities under non-sinusoidal conditions that consumption-class meters do not resolve. A consumption meter sees no change. A waveform meter sees the harmonic shift weeks before failure.

Insulation breakdown precursors. Sub-cycle current spikes from incipient arcing faults last a few milliseconds and self-extinguish. At standard polling they are invisible. At 8 kHz they are a clear signature. The NFPA 70 (National Electrical Code) Article 210.12 recognizes arc-fault detection as a distinct safety category for exactly this reason: the events are real, fast, and undetected by consumption metering.

Both vendors clamp on the same circuits. Only one of them can see what is happening inside them.

When is Packet Power the right fit?

Packet Power is a good choice when basic per-rack visibility is the requirement and the deployment needs to be fast and inexpensive.

Common fits:

Small or mid-size data centers where AI workloads are not the primary concern
Operators who want environmental monitoring (temperature, humidity) alongside basic power readings
Sites where the existing PDU-level visibility is too coarse and you need branch circuit consumption data
Budget-constrained projects where lowest entry cost is the priority

Packet Power is established in the colocation and enterprise data center market. Their wireless sensors install in hours, their dashboards are familiar to facility teams, and the per-sensor pricing is the lowest in the category. For consumption visibility, it works.

When is Verdigris the right fit?

Verdigris is the right choice when the workload is AI, the equipment is critical, and the question is not how much power but whether the equipment is healthy. SREs and capacity engineers responsible for stranded-capacity recovery and AI-factory reliability operate at the intersection where consumption telemetry stops being enough.

Common fits:

AI factory and NeoCloud operators running GPU training jobs where 8 millisecond power transients define capacity
Predictive maintenance programs on UPS infrastructure where early warning grounded in 8 kHz waveform physics catches degradation weeks before failure
Vendor-neutral fleet visibility across Schneider, Eaton, Vertiv, Honeywell, and ABB equipment in the same power path
Capacity planning that needs validated operating limits, not nameplate ratings, to safely raise rack density
Sites where a single unplanned outage costs more than the difference in monitoring premium

The price difference is the validated layer: anomaly detection, early warning grounded in waveform physics, and validated operating limits. For AI infrastructure, the calculus is cost per prevented outage, not cost per sensor.

How it deploys in production: the T-Mobile rectifier fleet

T-Mobile deployed Verdigris across more than 800 rectifiers in their critical telecom power infrastructure. Within the monitored fleet, Verdigris detected equipment degradation in 4% of more than 800 monitored rectifiers that was undetected by the existing BMS. Zero standard alarms had fired on those units. The degradation lived in the harmonic content of the input current, not in the average power reading.

The advance warning window was 21 days. Consumption monitoring at any standard polling rate would have shown those rectifiers as nominal. The signature lived in the harmonic content of the input current, visible only at sampling rates fast enough to reconstruct the waveform.

In the words of the Senior Director of Infrastructure and Engineering at T-Mobile:

"Verdigris Signals gives us visibility into degradation patterns that [...] don't exist in [legacy] monitoring, enabling us to maintain our reliability leadership position while optimizing capital allocation."

That window matters for context. The Uptime Institute Global Data Center Survey reports that power-related failures continue to be the largest single cause of significant data center outages, and that mean-time-to-repair on power equipment has not improved meaningfully in five years. Twenty-one days of advance warning is the difference between a planned maintenance window and an emergency replacement.

Consumption monitoring would have shown those degrading units as nominal. The current draw was within spec. The voltage was within spec. The power factor was within spec. The degradation signature was in the 5th and 7th harmonic content of the input current, visible only when you sample fast enough to reconstruct the waveform.

This is the architectural difference. A consumption-class meter reports averages. An 8 kHz waveform sensor reports the shape of every cycle. Equipment degradation shows up in the shape long before it shows up in the average. That is what waveform-class resolution actually buys. And it directly translates to stranded-capacity recovery for SREs who cannot derate the fleet on guesswork.

Running both together

These systems are not mutually exclusive. Data center operators often start with Packet Power for basic visibility and add Verdigris for the equipment that matters most, like UPS systems feeding AI training racks.

Verdigris EV2 sensors clamp on existing circuits independently of any wireless monitoring already in place. The two systems do not interfere. Verdigris ingests data from third-party sources through a REST API, so Packet Power consumption data can feed into the same intelligence layer that analyzes Verdigris waveforms.

The pattern we see most often: Packet Power for the broad fleet, Verdigris for the critical power path. Consumption monitoring everywhere, waveform intelligence where it matters.

A second case illustrates what waveform-level visibility surfaces beyond rectifier health. At Orange Silicon Valley, Verdigris uncovered more than $40,000 in annual energy savings inside an innovation lab where consumption telemetry had previously reported the facility as nominal. Scott Dworkis, Infrastructure Engineer at Orange Silicon Valley, framed the finding directly:

"You're likely to learn surprising [...] once you start digging into the data, [findings] that go straight to the bottom line, and this was certainly the case for us."

The point is not the dollar figure. The point is that consumption-class polling treated the facility as already-optimized. Waveform-class analysis disagreed.

What if Packet Power adds waveform capture or ML analytics?

Packet Power's current offering is consumption monitoring at modest cadence. Continuous waveform retention plus ML failure-mode analytics is a different architecture entirely, not a feature toggle on the existing platform.

Packet Power captures consumption metrics (V, A, W, kWh, PF, Hz) through a wireless sensor portfolio spanning PDU, branch circuit, DC busbar, and environmental monitoring. The cloud ingests at consumption-monitoring cadence, which is appropriate for that data type but not for waveform-at-scale. Even if Packet Power shipped a higher-frequency sensor tomorrow, the pipeline isn't built for continuous waveform retention across a fleet, and the analytics aren't trained on failure-mode signatures.

Verdigris's pretrained ML for rectifier degradation, capacitor ESR drift, and harmonic failure signatures was trained on years of waveform data across the customer base. That training corpus is the moat, not the sensor.

Buy what Packet Power offers today: consumption monitoring with wireless sensors. Verdigris is a different category. It's waveform-grade electrical intelligence with pretrained ML for equipment failure modes. If you already have Packet Power, layer Verdigris on top for the validated layer they don't provide.

Related vendors

Verdigris + Schneider Electric

Augment EcoStruxure with 8 kHz validation across the full power path.

Verdigris + Eaton

Keep Brightlayer and PredictPulse. Add 8 kHz validation across every vendor.

Verdigris + Vertiv

The validated layer above Environet and Liebert; 21-day early warning on rectifier degradation.

Verdigris + ABB

Proven Ekip Smart Fit and ABB Ability integration; 8 kHz validation across the path.

Verdigris + Siemens

Add 8 kHz validation to Building X, Desigo CC, and Sivacon panels.

Verdigris + Honeywell

Stream 8 kHz electrical intelligence to Forge on top of EBI.

Verdigris vs SkySpark

Rules on 1 Hz BMS data vs ML on 8 kHz waveforms. How the architectures differ.

Pricing & SKUs

Cloud, Forge, and Sentinel. Per-MW basis with Cloud Light brownfield on-ramp for retrofit pilots.

Packet Power meters consumption Verdigris validates load capacity