Industries / Enterprise Data Center

    Fewer alerts Earlier warning Defensible audit trail

    Operators get an average of 4,484 alerts per day. 67% are ignored. The signals that actually predict failure live below the measurement floor of incumbent BMS. Verdigris adds physics-validated early warning that cuts noise and gives you 21 days of advance notice before equipment fails.

    Enterprise data center with mixed-vendor infrastructure

    The average enterprise data center operator receives 4,484 alerts per day. Two-thirds get ignored. The signals that would actually predict an outage are mixed in with thousands of low-quality alarms from threshold-based monitoring that was designed for a different generation of equipment.

    Meanwhile, predictive-maintenance vendor adoption in enterprise data centers actually declined from 30% to 27% between 2024 and 2025. Schneider themselves acknowledge that vendors are "riding the marketing wave." What we mean by predictive maintenance is early warning grounded in waveform physics. Most vendors mean threshold-based pattern matching that adds to alert fatigue without resolving the underlying failure modes. The labels promise early warning. The methods rarely deliver.

    4,484

    Alerts received per day by the average DC operator

    67%

    Of those alerts ignored due to alert fatigue

    21 days

    Advance warning Verdigris delivered at T-Mobile, validated through waveform physics

    What enterprise operators are up against

    Most enterprise and telco data centers carry a mix of equipment vintages and vendors. Schneider, Eaton, Vertiv, ABB, Honeywell. Installed across multiple expansions over a decade or more. Each generation came with its own monitoring stack. Each vendor's "predictive maintenance" offering covers part of the failure surface but not all of it.

    These run together in three ways:

    • Alert noise compounds. Every threshold-based system on every vendor's gear contributes its own alarms. Real warnings get drowned out by the volume of nuisance alerts. Real failures get buried.
    • The failure modes that take you down aren't in the parameters being monitored. UPS rectifier capacitor degradation, transformer winding stress, harmonic shifts from new GPU tenants. These express themselves in waveform physics, not in temperature or vibration. By the time thermal sensors pick up the change, you're often weeks past comfortable lead time.
    • MTBF and service continuity get harder to defend. Without a validated layer that operates at the timescale failures actually develop on, fleet reliability metrics become reactive. Counting failures rather than preventing them.

    21-day warning, grounded in physics

    Verdigris adds a validated layer that sits below the noise floor of incumbent BMS. 8 kHz continuous waveform capture on the circuits that matter. UPS rectifiers, primary feeds, GPU-tenant zones. Resolves harmonic content through the 25th order. The Verdigris Signals layer translates physics shifts into a small number of high-quality early-warning signals.

    At T-Mobile, that meant 21 days of advance warning across an 800-rectifier UPS fleet (4% of units actively degrading inside the maintenance window). Caught by waveform physics, missed entirely by standard BMS thresholds and scheduled inspections. The maintenance team replaced the rectifiers on a planned schedule instead of reacting to a service-affecting event.

    Detection mode
    Detection horizon by mode — waveform physics, thermal, vibrationThree horizontal detection bands on a days-before-failure axis. Verdigris waveform-physics detection band starts at T-21 days. Incumbent BMS thermal sensor fires in the T-hours range as a consequence of the ESR rise that waveform analysis caught earlier. Vibration alarm fires in the T-minutes range as a consequence of mechanical swelling later still.Detection horizon by modeAll three modes overlaid. Earlier (left) is better. The modes catch the same failure at different points in the chain.Failure eventWaveform physicsT−21 daysThermal sensorT−hoursVibration alarmT−minutesEarlier ←Days before service-affecting failure→ LaterEach row is one detection mechanism. The earlier the band, the more days for your team to plan.Per Gasperi (1996) and IEEE 519: capacitor ESR drift shows up in harmonic ratios FIRST. Thermal and vibration come later in the failure chain.

    The causal chain

    Capacitor ESR drifts → harmonic ratio shifts (Verdigris waveform, T−21 days) → I²·ESR dissipation heats the component (BMS thermal sensor, T−hours) → electrolyte gas-out swells the case (BMS vibration alarm, T−minutes) → service-affecting failure. Verdigris catches the FIRST signal in that chain. Click a mode above to see its mechanism and physics citation.

    Verdigris is the validated layer, not a replacement BMS

    Verdigris doesn't replace your existing Schneider, Eaton, Vertiv, or ABB BMS. It layers physics-validated signals on top of them. Your operators keep the control surface they already trust. They get fewer, more accurate signals about what's actually degrading. Your audit trail captures the measurement-grade evidence that prediction was grounded in real waveform data, not pattern matching.

    What changes for enterprise operators

    Alert noise drops; alert quality rises

    Replace threshold-based pattern matching with physics-validated early warning. The signals that fire are signals worth investigating.

    21-day advance warning, fleet-wide

    Validated lead time before service-affecting failure. Plan replacements on your schedule instead of reacting to outages.

    Multi-vendor equipment, single validated layer

    Verdigris EV2 captures waveforms on the circuits feeding any vendor's equipment. The validated layer is consistent across Schneider, Eaton, Vertiv, ABB.

    Compliance-grade audit trail

    Every alert is grounded in measured waveform data with timestamped physics. Defensible in audit, regulatory review, and post-incident investigation.

    Read More

    Three Approaches to Predictive Maintenance in AI Data Centers
    Predictive Maintenance

    Three Approaches to Predictive Maintenance in AI Data Centers

    Why most vendor offerings don't deliver the lead time they promise. Schedule, condition, and physics-based comparison.

    Read more
    When training jobs die and the GPU logs don't explain it
    AI Data Center

    When training jobs die and the GPU logs don't explain it

    The sampling-gap failure mode at the edge of your fleet, where 1 Hz monitoring smooths over the events that actually trip protection.

    Read more
    What is 8 kHz monitoring?
    Fundamentals

    What is 8 kHz monitoring?

    Why sample rate is the bottleneck on early-warning fidelity, and what becomes possible above the BMS measurement floor.

    Read more
    The Hidden Costs of Inherited Data Center Infrastructure
    Data Centers

    The Hidden Costs of Inherited Data Center Infrastructure

    Taking over an aging facility or managing an environment built years ago? The biggest challenge is understanding the electrical environment you inherited.

    Read more
    The Three Eras of Data Center Reliability Engineering
    Data Centers

    The Three Eras of Data Center Reliability Engineering

    Reliability evolved through three eras: redundancy, monitoring, and intelligence. Most facilities are still in era 2. Era 3 changes the economics.

    Read more
    Pricing

    Pricing & SKUs

    Per-MW for Cloud and Sentinel. Cloud Light per-gateway for brownfield pilots. Perpetual license + maintenance for Forge.

    Read more

    Cut alert noise. Catch real failures earlier.

    Verdigris layers on top of your existing BMS. Same control surface. Validated alerts. 21-day lead time.

    Explore Verdigris SignalsTalk to an Engineer