The Thermal Threshold

Air cooling works – until compute density crosses the thermal threshold.

For years, incremental upgrades were enough. Higher airflow. Stronger fans. Better containment and more equipment. When density rises beyond a certain point, air becomes a limiting medium.

Heat accumulates faster than it can be removed.

Beyond that point, you’re no longer removing heat – you’re chasing it.

The Response Gap

AI doesn’t break cooling systems. Density does.

The challenge is not artificial intelligence.

It is the concentration of compute within the same physical footprint.

When kilowatts per rack rise sharply, response time becomes critical. When kilowatts per rack rise faster than response time, the system destabilises.

From Product to Behaviour

We didn’t just improve a rear-door cooler. We changed how the system behaves.

Cooling can be treated as equipment – or as system behaviour. At high density, stability depends on how components interact, not how they perform in isolation.

NGC’s system is designed to anticipate thermal change – not just react to it.

At some point you stop designing components and start designing behaviour.

Autonomous, Not Isolated

Each rear door is autonomous – but it never acts alone.

Every rack regulates its own flow. But stability emerges from coordination. Units monitor conditions locally while compensating collectively. The room remains thermally neutral – even under sudden load changes.

Neighbour help isn’t a feature. It’s a behaviour.

Stability Over Peak Numbers

What matters isn’t peak performance. It’s how quickly the system stabilises.

Specifications describe capacity. Operation reveals dynamics. Under volatile workloads, stability is measured in seconds – not kilowatts. Reaction time defines whether a system absorbs change or amplifies it. We don’t design for perfect conditions.

We design for what actually happens.

At Data Center World, we were approached by an engineer from Aramco with a specific challenge.

A quantum computing installation required a stable 15 °C water supply. The existing infrastructure delivered 10 °C. The system had to bridge that gap without structural rebuild or interruption to ongoing operations.

The question was simple:

Can you convert 10 °C supply water into 15 °C with a flow of 3 m³/h at 1.3 bar differential pressure – reliably, and with precision?

Within weeks, we designed a modified Cooling Distribution Unit capable of delivering up to 5 bar and 5 m³/h, incorporating targeted design adjustments to improve precision and control.

We manufactured the unit and shipped it directly to Saudi Arabia. Installation was carried out by the customer’s engineers, supported remotely by our team.

No local presence was required.

During commissioning, a monitoring-board issue was identified and corrected. The thermal control architecture itself performed as designed from first activation.

The CDU has now operated continuously for four months, maintaining supply temperature within ±0.1 °C.

In quantum computing environments, temperature stability is not an optimisation target – it is a prerequisite.

The project reflects a central principle of our approach:

When system architecture is understood at thermodynamic and hydraulic level, precision is achieved through control – not through excess infrastructure or unnecessary complexity.

It requires engineers who enter the customer’s context, question assumptions, and allow the solution to take shape through dialogue.

NGC took time to understand the challenge before proposing the solution

A customer approached us when their server room began to overheat. Newly installed servers delivered higher performance – and significantly higher heat density – than the space had been designed to manage.

We proposed rear-door cooling as an alternative

Ceiling height prevented additional fan coils. A downflow CRAH solution was considered, but could not be accommodated. In-row cooling was evaluated, yet available whitespace was limited.

The installation centred on a 42U, 800 mm rack with a 50 kW load. Room temperature was maintained at 23 °C. Server exhaust air was estimated at 45 °C.

We implemented an RDHx with integrated circulation pump, combined with a dry cooler and chiller in a single fluid circuit. The system was engineered to balance airflow, pressure drop, and return temperature – avoiding excessive fan or pump energy while maintaining stable thermal behaviour.

At maximum load, fan power was 735 W and pump consumption 95 W. Under realistic operating conditions, indoor pPUE (partial Power Usage Effectiveness) was 1.0136.

Including dry cooler and chiller operation, total cooling sub-system pPUE reached 1.0294. The dry cooler handled 94% of annual cooling demand based on local ambient temperature data.

• The solution required no structural rebuild.
• It stabilised rack-level thermal behaviour.
• It reduced energy consumption without increasing system complexity.

NGC understood our constraints – and engineered an energyefficient cooling system to match.