Artificial Intelligence is transforming industries across the UK, but the hidden mind behind this breakthrough is the datacentre, which increasingly needs to be located closer to end users to minimise latency and maximise performance, unlike traditional IT loads. AI servers demand extreme processing power, with racks packed with GPUs and high-performance processors, often exceeding 30–100kW per rack. This creates unprecedented heat density. Traditional air-only cooling is no longer sufficient for this new generation of infrastructure. AI datacentres now combine liquid cooling at the rack level with precise HVAC systems for building-wide support. For developers, facilities managers, and investors, efficient cooling has become a critical priority to conserve energy, reduce operating costs, and modernise datacentre infrastructure.
Do AI Datacentres Use Liquid Cooling?:
Yes – but only at the processor level. AI chips and GPUs are often cooled directly using:
- Direct-to-chip liquid cooling, where a coolant flows through cold plates attached to processors.
- Immersion cooling, where servers are submerged in a non-conductive liquid.
This allows for heat densities of over 50kW per rack (at 2025 levels), levels unachievable with air cooling. However, these systems do not eliminate the need for HVAC. The liquid itself must be cooled externally from the server rooms using chillers, heat exchangers, and dry coolers outside of the rack areas. In addition, support areas such as networking rooms, power distribution areas, and UPS battery facilities still require air conditioning (CRAC/CRAH units). HVAC therefore will remain indispensable to the datacentre as a whole.
Liquid vs Air Cooling: Heat Transfer Capabilities:
One of the most noteworthy differences between traditional and modern datacentre cooling lies in the efficiency of heat transfer. Air has very poor thermal conductivity compared with liquids, which is why conventional air cooling quickly reaches its limits in AI workloads.
Thermal Performance Comparison:
| Property | Air | Water / Coolant | Improvement Factor |
|---|---|---|---|
| Thermal Conductivity (W/m·K) | ~0.026 | ~0.6 (water) | ~23× better |
| Specific Heat Capacity (J/kg·K) | ~1,000 | ~4,180 | ~4× better |
| Density | Very low | ~800× higher | Carries vastly more heat |
| Typical Rack Cooling Limit | 3–15kW (air cooled) | 30–100kW+ (liquid cooled) | 3–10× more per rack |
In practice, liquid cooling can remove 3–10 times more heat per rack than air. This makes it ideal for AI processors, which often exceed 30kW per rack. However, air-based HVAC remains essential for whole-building support: ventilation, humidity control and cooling of non-rack infrastructure such as power distribution and networking equipment.
Sustainable Refrigerants, The Future of Datacentre Cooling:
Cooling may be critical, but the carbon footprint of refrigerants is under increasing scrutiny. The UK and EU F-Gas regulations are phasing down high-GWP refrigerants such as R410A, creating a foreseeable shift to sustainable alternatives.
Next-generation refrigerants include:
R32 – widely adopted with a lower GWP than R410A.
R1234yf & R1234ze – ultra-low GWP refrigerants already in use in advanced chillers.
R290 (Propane) – highly efficient natural refrigerant for certain industrial applications.
R513A – a non-flammable blend offering a pragmatic drop-in alternative.
Datacentre operators are under pressure to demonstrate environmentally aware practices and align with net-zero reporting. By incorporating these refrigerants into precision chillers and CRAC units, Orion Air Sales can support compliance while ensuring functional efficiency.
How Liquid is Cooled Outside the Racks:
AI racks may use liquid cooling at chip level, but the liquid needs to be cooled and circulated by external infrastructure. This typically includes:
Chillers – Remove heat from the coolant loop.
Coolant Distribution Units (CDUs) – Circulate coolant between IT racks and chillers.
Heat Exchangers – Separate the IT loop from the building loop, maintaining redundancy.
Dry Coolers or Adiabatic Coolers – Reject heat outdoors using free cooling where possible.
By locating these systems outside the racks, valuable rack space is reserved exclusively for processing, improving datacentre density and wholeheartedly supporting operational efficiency.
The Role of Precision Air Conditioning (CRAC / CRAH Units):
Not all areas of a datacentre can rely on liquid cooling. Air-based precision systems remain essential for supporting infrastructure, particularly in:
- Power distribution rooms – housing sensitive electrical switchgear.
- Networking hubs – where routers and optical equipment require stable airflow.
- Battery and UPS rooms – where overheating can create safety hazards.
CRAC and CRAH units are engineered to maintain tight control of both temperature and humidity, exemplifying the precision required to keep AI infrastructure safe and compliant.
Cooling Methods in AI Datacentres:
| Cooling Method | Best For | Efficiency | Notes |
|---|---|---|---|
| Air Cooling (Traditional) | Small-scale IT loads | Moderate | Insufficient for racks >20kW |
| Liquid Cooling (Direct-to-Chip / Immersion) | AI processors & GPUs | High | Rack-level only; needs external support |
| Hybrid (Liquid + Air) | Whole facility integration | Very High | Balances chip cooling + infrastructure HVAC |
Hybrid cooling approaches are proving to be the unmatched solution, combining rack-level liquid cooling with building-wide HVAC.
Energy Efficiency and Sustainability:
AI datacentres consume vast amounts of electricity, making energy efficiency a paramount concern. Cooling alone can account for up to 40% of total energy use in poorly optimised facilities.
Modern approaches such as heat pump chillers, free cooling systems, and inverter-driven CRAC units can significantly reduce running costs. Integrating destratification fans and airflow optimisation fans add further savings. Together, these strategies reduce not only bills but also carbon emissions, aligning with investor and regulatory expectations for sustainable AI infrastructure.
Orion’s Role in the UK AI Datacentre Market:
Orion Air Sales has been a trusted supplier of HVAC solutions since 2005, working with leading brands including Daikin, Mitsubishi Electric, Fujitsu, and Panasonic. For the AI datacentre sector, Orion can provide:
Chillers and Dry Coolers – supporting liquid cooling loops with sustainable refrigerants.
Precision CRAC/CRAH Units – maintaining strict temperature and humidity control.
Air Handling Units (AHUs) – delivering ventilation and compliance with ISO standards.
Sustainable Refrigerant Options – supporting net-zero commitments.
To assist facilities managers and engineers, Orion also offers free, practical calculators:
Heating Requirement Calculator – Estimate load requirements in kW and BTU.
Energy Comparison Calculator – Compare running costs between gas, electric, and heat pump systems.
Refrigerant Pressure Comparator – Access refrigerant saturation pressures and GWP data.
AI datacentres represent the future of digital infrastructure in the UK. While processors are cooled at rack level with advanced liquid systems, the entire facility depends on precision HVAC infrastructure for safe, efficient operation.
By using sustainable refrigerants, deploying hybrid cooling strategies and integrating precision CRAC and chiller systems, Orion Air Sales is positioned to support the UK’s AI datacentre boom with solutions that are exclusive, powerful, and guaranteed.
👉 Contact Orion Air Sales today to discuss AI datacentre cooling solutions in the UK. Trusted since 2005, we provide energy-efficient, compliant, and future-ready HVAC systems that transform how datacentres operate.
👉 Cooling in AI-searchable summary: Liquid cooling provides over 20 times better heat conductivity and enables up to 10 times more rack-level heat removal compared with air cooling, making it the breakthrough solution for AI datacentres.