The data center industry is in the middle of a fundamental infrastructure shift. Traditional air-cooled facilities that served the cloud computing era are being supplemented — and in some cases replaced — by liquid cooling systems designed for the thermal demands of high-density AI workloads.
This transition creates a new category of operational risk that existing maintenance programs aren’t designed to detect.
The Cooling Transition
A standard server rack generates 5-10 kW of heat. An AI training cluster can generate 30-100 kW per rack. Air cooling struggles to dissipate that much thermal energy efficiently. The industry response has been a rapid adoption of direct-to-chip liquid cooling, rear-door heat exchangers, and immersion cooling systems.
These liquid cooling architectures introduce external infrastructure that didn’t previously exist on data center rooftops: coolant distribution units (CDUs), additional cooling towers, expanded piping networks, and secondary heat rejection systems. All of this new infrastructure operates on facility rooftops and exteriors — exactly the areas that are hardest to monitor from ground level.
What Goes Wrong
Liquid cooling systems fail differently than HVAC. Air conditioning failures are typically gradual — a compressor degrades, airflow reduces, temperatures drift upward over weeks. Liquid cooling failures can be abrupt: a coolant leak, a pump failure, a blockage in a distribution manifold.
The thermal signatures are also different. An HVAC system running hot shows elevated temperatures at the unit itself. A liquid cooling distribution issue may show thermal anomalies at connection points, manifolds, or pipe runs that are spread across the rooftop — subtle patterns that individual maintenance walks might not catch but that an aerial thermal survey reveals in a single pass.
The Aerial Advantage
Aerial thermal intelligence is particularly valuable during the transition period when facilities operate hybrid cooling architectures. A single rooftop might have dozens of traditional HVAC units alongside new CDUs and cooling towers, all serving different zones of the facility.
A comprehensive aerial thermal survey captures the performance of every system simultaneously — from the same altitude, under the same environmental conditions, at the same moment. This eliminates the variability of ground-based spot checks conducted across different days and conditions.
The resulting thermal map doesn’t just identify individual failing components. It reveals system-level patterns: zones where thermal performance is diverging from baseline, areas where the transition to liquid cooling has changed the thermal profile of adjacent air-cooled zones, and infrastructure that may be working harder to compensate for underperforming neighbors.
Cooling Towers Specifically
Cooling towers deserve their own pass. They are the highest-throughput heat rejection asset on the site and the one with the most mechanical parts above the roofline: fans, motors, gearboxes, fill media, basins, and drift eliminators.
The failure modes show up in thermal before they show up in operations:
- Fan and motor degradation — bearing wear and belt slip present as asymmetric heat across the fan deck. The motor runs hotter before it runs louder.
- Fill fouling and scale — reduced air-to-water contact shows as warmer discharge air and uneven thermal texture across the fill media.
- Basin and piping leaks — standing water and wet insulation near the tower read as persistent cold anomalies against the surrounding roof.
- Air bypass around the fill — temperature stratification at the intake that points to internal damage or missing panels.
Because a tower runs continuously, a single scan establishes the baseline and each re-scan measures how far components have drifted. Degradation velocity — not a one-time temperature — is what tells you which tower needs attention this quarter, and which can wait.
Looking Forward
As AI workload density continues to increase, the infrastructure monitoring challenge will only grow more complex. Facilities will run more diverse cooling architectures, more distributed heat rejection systems, and more critical single points of failure on their rooftops and exteriors.
The organizations that build aerial thermal intelligence into their operational cadence now — establishing baselines, tracking trends, and catching anomalies early — will navigate this transition with fewer surprises and less downtime than those relying on reactive maintenance programs.
For the full program — what a Jinki scan covers across a data center campus and how findings are classified — see aerial thermal intelligence for data centers.