Summary
- Shell says NVIDIA has qualified its PG25 fluid as DSX Ready Coolant for direct liquid cooling.
- The propylene glycol formulation is designed for cold-plate systems and meets Open Compute Project PG25 coolant specifications.
- NVIDIA's public DSX Ready pages currently detail CDU and BESS categories rather than a separate public coolant directory.
Shell Lubricants says its direct liquid cooling fluid has received NVIDIA DSX Ready Coolant qualification, adding a new validation claim for the PG25 fluid used inside cold-plate cooling loops for high-density computing hardware.
The product is a propylene glycol based fluid that Shell launched in June 2025 as an extension of its data centre cooling portfolio. It circulates through liquid loops serving cold plates attached to processors or other high-heat components rather than immersing the complete server in a dielectric fluid.
Shell announced the NVIDIA qualification on 7 October and describes NVIDIA as the first organisation to announce qualification of the fluid. The specific coolant qualification should remain attributed to Shell because NVIDIA’s public DSX Ready pages currently describe qualified coolant distribution units and battery energy storage systems without providing the same public detail for a separate coolant category.
NVIDIA launched DSX Ready in September as a programme for infrastructure products that complete category-specific qualification requirements for AI factories. Its stated objective is to reduce integration risk by giving builders a defined set of power and cooling components aligned with the wider DSX architecture.
Fluid chemistry sits below the level of a complete coolant distribution unit, but it affects several interfaces inside the cooling loop. Corrosion, material compatibility, freeze protection, chemical stability and service life can all influence whether pumps, hoses, seals, cold plates and heat exchangers continue operating predictably over time.
PG25 chemistry reaches the cold plate
Shell DLC Fluid S3 uses a propylene glycol formulation and complies with the Open Compute Project’s PG25 coolant specification, according to Shell’s technical material.
The company says the fluid is formulated to protect aluminium, brass, cast iron, steel, solder and copper and is compatible with metals, elastomers, plastics and other wetted materials covered by OCP guidance. Those characteristics are relevant because a cooling loop can contain several materials with different corrosion behaviour.
Shell gives the fluid an expected service life of more than six years based on performance of similar chemistry in automotive applications operating under higher temperature and pressure. That figure is a manufacturer expectation rather than a guarantee of life in every data centre installation, where maintenance, contaminants and operating temperature can change fluid degradation.
A fluorescent green dye is included to make in-service leakage easier to identify, while the PG25 mixture provides freeze protection down to minus 10°C. External liquid-cooling infrastructure can experience lower ambient temperatures than the white space itself, particularly where parts of the loop or heat-rejection equipment sit outdoors.
Shell also says its direct liquid cooling approach can improve PUE by up to 27% compared with air cooling. That claim depends on the complete cooling design rather than the coolant alone, because pumping power, coolant temperatures, external heat rejection and the amount of server heat captured by cold plates all affect facility efficiency.
Cooling qualification becomes a system problem
Rising processor heat loads are pushing liquid cooling deeper into the server and facility design process. Once cold plates become part of the compute architecture, operators have to coordinate server hardware, coolant chemistry, CDUs, facility water systems, controls and maintenance rather than selecting room cooling independently from the IT load.
A fluid that transfers heat effectively can still create reliability problems if it attacks seals, encourages corrosion or changes materially as it ages. Conversely, a chemically stable coolant cannot compensate for undersized cold plates, poor flow control or insufficient heat rejection outside the data hall.
Qualification programmes are therefore becoming more relevant as operators try to assemble high-density systems from equipment supplied by several manufacturers. NVIDIA’s DSX Ready framework addresses part of that integration problem at the infrastructure level, even though its publicly documented categories currently focus on CDUs and battery systems.
Shell’s own portfolio now covers both direct liquid cooling and immersion cooling. The two approaches use liquid to remove heat more efficiently than conventional air systems but create different material and maintenance requirements.
Immersion cooling places servers directly inside electrically non-conductive fluid. Direct liquid cooling retains a more conventional server format and moves a water-based coolant through cold plates attached to selected components, making PG25 formulations relevant to a wider range of server designs already entering AI deployments.
The Shell announcement therefore adds a qualification claim around one component of that stack rather than establishing the efficiency of an entire data centre cooling system. Its operational value will depend on how the fluid performs over time alongside qualified cold plates, CDUs, pumps and facility heat-rejection equipment.
