PRE-FIRE DETECTION SYSTEM / DALLAS LEGACY CO-LO, BUILDING 2

The wire burns
before anyone knows
it's on fire.

INTRCPT GLB-1 watches the conductor — not the rack, not the room. Four physically-independent sensing modalities. Eleven threat classes. Weeks of lead time before catastrophe becomes loss.

11 THREAT CLASSES
4 SENSING MODALITIES
WKS LEAD TIME TIER 2
24 CDUs ACTIVE
LIVE ZONE STATUS 2s refresh
ZONE-DAL-CDU-06 0.14 NORMAL
ΔI 0.10Z(f) 0.13FFT 0.18EC 0.14
ZONE-DAL-CDU-01 0.14 NORMAL
ΔI 0.14Z(f) 0.12FFT 0.18EC 0.11
ZONE-DAL-CDU-05 0.13 NORMAL
ΔI 0.14Z(f) 0.15FFT 0.15EC 0.09
ZONE-DAL-CDU-02 0.12 NORMAL
ΔI 0.09Z(f) 0.14FFT 0.11EC 0.16
192 racks · 8 cooling zones · last telemetry 9:55 PM
42%
of data center loss costs come from fires.
FM Global, 15-year data center loss study — 2026 update
24%
come from liquid-related events.
FM Global, 15-year data center loss study — 2026 update
0
Current NFPA 75-mandated sensors that detect conductor-level threats from coolant contact.
NFPA 75 standard, current revision
HOW IT WORKS

One sensor. Four modalities. One composite signal.

INTRCPT GLB-1 installs at the cooling distribution unit and observes the shared conductor that feeds downstream racks. Four physically-independent sensing modalities watch for the precursors of conductor failure. No single modality dominates; all four contribute to the composite anomaly score.

ΔI
Hall Effect Differential Current SENSING MODALITY 01

Detects current imbalance at the cooling distribution unit, in real time. Partial wetness from coolant ingress creates phase asymmetry — ΔI catches it before the fault propagates.

T01 Coolant Ingress on Conductor T02 Ground Fault Progression T03 Differential Current Imbalance
Z(f)
Impedance Spectroscopy SENSING MODALITY 02

Sweeps the conductor across a frequency range and maps the impedance signature. Insulation degradation doesn't show up as resistance change — it shows up as impedance drift weeks before the fault window opens.

T04 Insulation Thermal Fatigue — 4–12 wk lead T07 Connector & Busbar Oxidation — 8–16 wk lead
FFT
Frequency Spectrum Analysis SENSING MODALITY 03

Monitors the spectral footprint of the electrical signal on the conductor. Arc-precursor signatures and vibration wear patterns generate characteristic frequency signatures — FFT catches them before the arc forms.

T07 Busbar Oxidation (FFT component) T08–T11 Severe failure modes — NDA required
EC
Coolant Conductivity SENSING MODALITY 04

Measures the conductivity of the coolant at the CDU level. Chemistry drift, ion ingress, and contamination change the conductivity baseline — EC detects it before it reaches the conductor interface.

T05 Coolant Conductivity Drift — days of lead T06 Micro-Condensation on Conductors T09 Coolant-chemistry transition — NDA
THREAT MATRIX

What every CDU is watching for, every second.

TIER 1 — CRITICAL Real-time detection
01
Coolant Ingress on Conductor Liquid contact on energized conductor. The event NFPA 75 misses entirely.
02
Ground Fault Progression Slow leakage path development to ground. Identifiable before it becomes a fault.
03
Differential Current Imbalance Partial wetness creating phase asymmetry. ΔI catches it in real time.
TIER 2 — COMMON ML trend · weeks of lead time
04
Insulation Thermal Fatigue Heat-cycle degradation. 4–12 week lead time via Z(f) impedance drift.
05
Coolant Conductivity Drift Fluid chemistry change. Days of lead time via EC monitoring.
06
Micro-Condensation on Conductors Diurnal moisture pattern detection. Hours of lead time.
07
Connector & Busbar Oxidation Vapor-induced contact resistance rise. 8–16 week lead time.
TIER 3 — SEVERE NDA required for technical depth
08
Propagation-Mode Fault Detection Distributed-architecture fault topology requires cross-modality analysis. NDA Required
09
Coolant-Chemistry Transition Risk Edge-case detection for facilities migrating between cooling fluid classes. NDA Required
10
High-Voltage Propagation-Mode Anomaly Sub-minute detection window. No commercial equivalent for liquid-cooled AI infrastructure. NDA Required
11
Mechanical-Electrical Interface Degradation Long-horizon wear pattern recognition. NDA Required
LIVE VERIFICATION WINDOW

Verifiable, not theatrical.

Every alert backed by a cryptographic commitment. INTRCPT seals its anomaly schedule at boot with a SHA256 commit hash. The hash is public from day 1. The full schedule reveals after the verification window closes — anyone can recompute the hash and prove no events were inserted, removed, or moved during the test.

COMMIT HASH
SHA256 sealed at window open — revealed at window close
30 DAY WINDOW
8 COOLING ZONES
24 CDUs MONITORED
192 RACKS COVERED
ARCHITECTURE
N+1 Redundancy
3 GLB-1 per zone
Lockstep CDU monitoring
Vertiv 360AI / GB200 reference pattern
WHO IT'S FOR

Three audiences. One verification window.

Insurance Carriers

Reduce 66% of data center loss costs at the source. Pre-fire detection for the two largest categories of paid loss: fire (42%) and liquid (24%).

Loss Reduction Contractual Mandates Risk Pricing

Standards Bodies

Fill the NFPA 75 gap for liquid-cooled AI infrastructure. Current sensor mandates predate liquid-cooled high-voltage DC clusters. INTRCPT covers what the standard doesn't yet specify.

NFPA 75 ASHRAE W-Class Standards Advocacy

Datacenter Operators

See the failures you don't see today. Weeks of lead time on slow failures. Sub-minute on the catastrophic ones. Drop-in at the CDU.

Operator Dashboard N+1 Redundancy Real-Time Alerts

The standard was written before the wire got wet.
We wrote the sensor that watches what the standard doesn't cover.

CONTACT intercptio@gmail.com