Air Cooling Ran Out of Headroom
A conventional enterprise rack sits somewhere between 5 and 15 kW. AI training clusters built on current GPU architectures commonly draw 40 kW, 60 kW, or beyond 100 kW per rack. That is close to an order of magnitude in the same floor area, and no amount of additional air movement closes it — which is why liquid cooling has moved from a future consideration to a core requirement in UAE data centre design.
The UAE adds its own constraint. High ambient temperatures shorten the window in which dry coolers or free cooling can reject heat economically, which pushes facilities toward water-cooled chiller plant or a district cooling connection. The control strategy has to manage that handoff continuously against live IT load rather than running a fixed sequence.
Our position in this is specific: Intellicon is the controls and monitoring layer, not a cooling equipment supplier. We integrate the plant your mechanical contractor installs into a coherent BMS and own the points, graphics, trends, alarms and interlocks around it.
What We Monitor and Control
Coolant loop conditions
Supply and return temperature, flow rate and differential pressure trended per loop, with deviation alarms before thermal throttling begins.
CDU integration
Coolant Distribution Unit status, pump duty/standby rotation, filter differential and setpoint control brought onto the main head-end over BACnet/IP or Modbus TCP.
Zone-addressable leak detection
Rope and spot detection reporting location, not just a single alarm state — wired to support valve isolation as an interlock.
Per-rack thermal trending
Inlet and outlet conditions at rack level so high-density rows are visible individually rather than averaged into a hall reading.
Heat rejection sequencing
Facility water loop, chiller plant and district cooling interface sequenced against ambient conditions and live IT load.
Power quality at the rack
Voltage stability and harmonics monitoring alongside thermal data, delivered with our sister company King Link Electrical.
Direct-to-Chip, Immersion and Rear-Door
Direct-to-chip circulates coolant through cold plates on the processors. The controls task is per-loop temperature, flow and differential pressure, plus CDU interface and tight leak interlocking — fluid is inside the rack, so a leak is an IT availability event rather than a facilities one.
Immersion cooling submerges hardware in dielectric fluid. Monitoring shifts toward tank fluid temperature and level, pump and heat exchanger status, and fluid condition trending over time. Alarm philosophy differs because thermal mass buys more response time than a direct-to-chip loop does.
Rear-door heat exchangers are frequently the pragmatic retrofit path for an existing hall moving to higher density. They interface to the facility water loop and integrate cleanly into an existing BMS, which makes them a common first step before a full liquid build-out.
Most live UAE facilities end up hybrid — air-cooled legacy halls alongside liquid-cooled AI rows. The integration problem is usually presenting both on one set of graphics with one alarm hierarchy, rather than choosing between them. See our BMS retrofit and integration page for how we approach mixed estates.
Frequently Asked Questions
What is direct-to-chip liquid cooling, and what does the BMS need to do differently?
Direct-to-chip cooling circulates coolant through cold plates mounted on the processors themselves, removing heat at source rather than through room air. For the controls layer this changes three things: the BMS must monitor and trend coolant supply temperature, flow rate and differential pressure per loop instead of only room temperature; it must interface with the Coolant Distribution Units (CDUs) that sit between the facility water loop and the IT loop; and leak detection moves from a nice-to-have to a primary safety interlock, because coolant is now running inside the rack.
How much heat does an AI rack actually produce compared with a conventional one?
A conventional enterprise rack typically sits in the 5–15 kW range. AI training clusters built on current GPU architectures commonly draw 40 kW, 60 kW, or in dense configurations over 100 kW per rack depending on hardware generation and configuration. That is roughly an order of magnitude, and it is the reason air cooling alone stops being viable — a room-level CRAC strategy cannot remove that much heat from that small a footprint, regardless of how much air you move.
Can liquid cooling controls be integrated into an existing data centre BMS?
In most cases yes, subject to a protocol assessment during site survey. CDUs and liquid cooling plant from major vendors typically expose BACnet/IP or Modbus TCP points, which integrate into an existing head-end alongside the current CRAC, chiller and power monitoring. Intellicon works vendor-neutral across Siemens, Honeywell, Schneider and open-protocol platforms, so the liquid cooling layer becomes another subsystem on the same graphics and alarm structure rather than a separate island.
Why does leak detection matter more with liquid cooling?
Because the failure mode changes. In an air-cooled hall a chilled water leak is a facilities problem happening outside the racks. With direct-to-chip or immersion cooling, fluid is circulating within centimetres of live IT equipment, so a leak is an immediate IT availability event. Detection needs to be zone-addressable rather than a single loop, reported into the BMS with location, and wired so that it can trigger valve isolation rather than only raising an alarm.
Does the UAE climate change the design for liquid-cooled data centres?
Yes. High ambient temperatures reduce the hours per year in which free cooling or dry cooler rejection is viable, which pushes UAE facilities toward water-cooled chiller plant or district cooling connection for heat rejection. The control strategy has to manage that handoff — sequencing the facility water loop against ambient conditions, and maintaining coolant supply temperature to the CDUs within a tighter band than a conventional chilled water system would require.
Is Intellicon a liquid cooling equipment supplier?
No. Intellicon is the controls, monitoring and integration layer. We do not manufacture or supply CDUs, cold plates or immersion tanks. We integrate the cooling plant your mechanical contractor or vendor installs into a coherent BMS — points, graphics, trends, alarms, interlocks and reporting — and we handle the leak detection and power quality monitoring around it. Scope is always confirmed by site survey and design review.
Planning a High-Density Deployment?
We will review your cooling topology, points list and existing head-end, then scope the integration.
Related solutions and guides
- Data Centre BMS & DCIM IntegrationCRAC/CRAH sequencing, containment, PUE reporting and alarms.
- Guide: Liquid Cooling and the BMS in AI Data HallsWhat direct-to-chip and immersion cooling add to the point list.
- Custom Voltage Stabilisers for Data CentresServo, static and three-phase stabilisation.
- BMS Commissioning & HandoverPoint-to-point testing, sequences and handover documentation.
Want to know what is reachable in your building? Request a data-centre controls proposal