The conversation around data-centre cooling refrigerants often starts with a single number: global warming potential (GWP). For high-density AI facilities, however, refrigerant selection is an engineering constraint problem. The applicable GWP limit defines which options are legally available. Engineers must then design a system that delivers the required cooling capacity, efficiency, safety and reliability within that boundary.

This changes the order of the decision. A refrigerant that exceeds the applicable legal threshold cannot be justified by a smaller footprint or better efficiency unless a relevant exemption applies. Equally, meeting the GWP threshold does not establish that a refrigerant is suitable for the installation.

For facilities deploying racks at 40–100 kW or more, refrigerant decisions connect directly to plant layout, safety infrastructure and cooling-loop temperatures. Treating compliance as a final purchasing check risks discovering constraints after the building and mechanical systems have already been designed.

‍

Start with the cooling architecture

Refrigerant selection begins with understanding where refrigerant is used.

In a direct-to-chip cooling system, liquid carries heat away from components such as CPUs and GPUs. A coolant distribution unit (CDU) typically transfers that heat between the technology cooling loop and the facility water loop. The CDU and chiller perform different functions: a CDU manages circulation and heat transfer between loops, while a chiller uses a refrigerant circuit to cool a liquid.

The facility may reject heat through dry coolers, cooling towers, mechanical chillers or a combination of equipment. Some designs can operate without mechanical chilling for substantial periods, depending on the required water temperature and outdoor conditions.

The refrigerant circuit may therefore sit outside the data hall, separated from the IT equipment by water loops and heat exchangers. Its location matters because an outdoor packaged chiller presents different installation considerations from equipment inside a machinery room.

Before comparing gases, the design team should establish the heat-rejection arrangement, chiller capacity, operating temperatures and required redundancy. These define the conditions under which a refrigerant must perform.

‍

Treat GWP as a legal boundary condition

For stationary chillers placed on the EU market, Regulation (EU) 2024/573 establishes restrictions through Article 11 and Annex IV.

‍

From 1 January 2027:

• Stationary chillers with a rated capacity up to and including 12 kW cannot contain or rely on fluorinated greenhouse gases with a GWP of 150 or more, subject to the applicable safety exception.

• Stationary chillers above 12 kW cannot contain or rely on fluorinated greenhouse gases with a GWP of 750 or more, subject to the applicable safety exception.

‍

From 1 January 2032, the restriction for chillers up to and including 12 kW extends to fluorinated greenhouse gases generally, subject to the applicable safety exception.

Annex IV does not establish a general GWP 150 limit from 2032 for stationary chillers above 12 kW. That distinction matters for the larger chillers typically considered in data-centre projects. Requirements for other equipment categories should not be applied automatically to stationary chillers.

These provisions concern placing equipment on the market. They do not introduce a chiller-specific servicing ban in 2027 or require every existing installation to stop operating.

For a new project, the equipment category and applicable threshold should be established before the refrigerant shortlist is developed. Within the compliant options, engineers can then assess performance, safety, cost and environmental impact.

‍

Build safety requirements into the plant design

Refrigerant safety classification affects the installation from the outset. Under the ASHRAE classification system, A indicates lower toxicity and B indicates higher toxicity. Flammability classifications distinguish gases with no flame propagation from those with lower or higher flammability.

R-513A, for example, is classified A1 and has a published GWP of approximately 629. It falls below the 2027 GWP threshold for stationary chillers above 12 kW. Its selection still requires an assessment of equipment performance, refrigerant supply and the operator’s longer-term environmental objectives. It should not be described as becoming prohibited for that chiller category in 2032 under the current Annex IV schedule.

R-1234ze(E) offers a much lower GWP and is classified A2L, indicating lower toxicity and lower flammability. R-454B is also A2L, but its GWP is approximately 465. It is a sub-750 option, rather than a sub-150 or ultra-low-GWP refrigerant.

‍

Moving to an A2L refrigerant requires an installation-specific safety assessment using applicable standards, including EN 378 where relevant, alongside manufacturer instructions and local requirements. Depending on the refrigerant charge, equipment location, room characteristics and system configuration, the design may require:

• limits on refrigerant charge or changes to circuit arrangement;

• refrigerant detection and alarm systems;

• mechanical ventilation or emergency extraction;

• isolation measures and controls;

• management of potential ignition sources;

• suitable access and maintenance procedures.

‍

These measures are not identical for every A2L installation. They must be determined for the actual equipment and location.

Very low GWP also does not necessarily mean A2L. R-1233zd(E), for example, is classified A1 and is used in suitable chiller designs. Propane, ammonia and carbon dioxide introduce different combinations of flammability, toxicity and operating pressure. Each option requires a compatible system architecture.

‍

Assess volumetric capacity and plant footprint

Refrigerants differ in the cooling capacity they can deliver for a given compressor displacement. This affects compressor selection, refrigerant flow, heat-exchanger design and pipe sizing.

Some lower-GWP refrigerants have lower volumetric refrigeration capacity than the refrigerants they replace. Depending on the application, delivering the same duty may require greater compressor displacement or a different equipment configuration. These changes can influence plant dimensions and service clearances.

However, lower GWP does not automatically mean a larger chiller. R-454B, for example, was developed to provide capacity characteristics close to R-410A in suitable equipment. Comparisons must use specific refrigerants and complete chiller designs at the same operating conditions.

For data centres, the practical question is how much installed cooling capacity the proposed plant can deliver within the available space. That assessment should include the chiller footprint, maintenance access, safety provisions, acoustic treatment and redundancy requirements.

A compact unit may still require substantial surrounding space. Conversely, a different compressor or heat-rejection arrangement may accommodate a low-GWP refrigerant without the footprint penalty assumed from refrigerant properties alone.

‍

Let direct-to-chip temperatures shape the design

Liquid-cooled AI hardware can support higher coolant temperatures than conventional chilled-water systems, depending on the server manufacturer’s specifications. Some designs operate with supply temperatures in the region of 30°C to 45°C.

The design team must identify which loop those temperatures describe. Technology-loop and facility-loop temperatures differ because heat exchangers require a temperature difference to transfer heat. A server coolant specification cannot simply be copied into a chiller selection.

Higher allowable water temperatures can reduce compressor lift, meaning the temperature difference between refrigerant evaporation and condensation. They can also increase the number of hours when outdoor conditions allow heat rejection without full mechanical chilling.

‍

This shifts the performance assessment away from a single nominal efficiency figure at a traditional 7°C chilled-water setpoint. More useful questions include:

• How efficiently does the chiller operate at the project’s actual water temperatures?

• What is its minimum stable load?

• How does it perform during partial mechanical cooling?

• How many annual hours can use free cooling?

• What are the combined energy requirements of compressors, pumps and fans?

• Can the system maintain capacity during the design summer conditions and equipment outages?

‍

Free cooling still consumes energy through pumps and fans, and its availability depends on climate, heat-exchanger approach temperatures and control strategy. Annual system performance should therefore be modelled using the site’s weather and operating profile.

‍

Keep HFC supply separate from equipment compliance

A chiller can meet the applicable equipment GWP threshold while using an HFC or an HFC-containing blend. That makes refrigerant supply a separate consideration.

The EU HFC quota system limits covered gases placed on the market in CO₂-equivalent terms. It operates alongside equipment restrictions, rather than replacing them.

For operators, this means reviewing the refrigerant charge, expected servicing requirements, supplier arrangements and relevant compliance documentation. Quota reductions do not establish a specific future price or guarantee a shortage, but they belong in the lifecycle assessment of an HFC-based plant.

Equipment compliance, refrigerant availability and servicing permissions should each be checked against the installation and supply route.

‍

Use a coordinated selection process

A practical refrigerant assessment should bring facilities engineers, MEP designers, procurement teams and equipment manufacturers together early.

1. Map the cooling architecture. Identify the technology loop, CDU, facility loop and heat-rejection equipment.

2. Establish the legal boundary. Confirm the equipment category, rated capacity, placing-on-the-market date and applicable restrictions.

3. Define operating conditions. Record water temperatures, ambient design conditions, cooling duty and redundancy requirements.

4. Assess safety architecture. Determine the measures required for the refrigerant, charge and installation.

5. Compare complete equipment designs. Review capacity, footprint, access requirements and performance at project conditions.

6. Model annual operation. Include free cooling, partial loads, pumping, ventilation and peak-weather performance.

7. Plan lifecycle supply and servicing. Review refrigerant procurement, maintenance capability and compliance documentation.

This process makes refrigerant selection part of the engineering brief before equipment is ordered.

‍

Design compliance into the system from day one

For AI data centres, GWP is a legal boundary condition before it becomes a comparison metric. Once that boundary is established, refrigerant properties influence safety infrastructure, compressor selection, plant layout and operating performance.

The strongest design connects these decisions to the actual cooling architecture. Elevated water temperatures, free-cooling opportunities and equipment location can be as consequential as the refrigerant itself.

AFS Cooling supports businesses with refrigerant procurement, quota management and F-gas compliance. Contact AFS Cooling to discuss refrigerant requirements and supply planning for your data-centre cooling systems.