Liquid Cooling Technologies Driving Data Centre Efficiency

Motivair introduces scalable CDU for AI data centres

Motivair, a provider of liquid cooling systems for data centres, owned by Schneider Electric, has announced a new coolant distribution unit designed to support high-density data centre cooling requirements, including large-scale AI and high-performance computing deployments. The new CDU, MCDU-70, has a nominal capacity of 2.5 MW and is intended for use in liquid-cooled environments where compute density continues to increase. Motivair says the system can be deployed as part of a centralised cooling architecture and scaled beyond 10 MW through multiple units operating together. According to the company, the CDU is designed to support current and future GPU-based workloads, where heat output is significantly higher than traditional CPU-based infrastructure. It notes that rack power densities in AI environments are expected to approach one megawatt and above, increasing the need for liquid cooling approaches. Designed for scalable, high-density cooling Motivair states that the new CDU integrates with Schneider Electric’s EcoStruxure platform, allowing multiple units to operate as part of a coordinated system. The design is intended to support phased expansion as cooling demand grows, without requiring major redesign of the wider plant. Rich Whitmore, CEO of Motivair by Schneider Electric, comments, “Our solutions are designed to keep pace with chip and silicon evolution. Data centre success now depends on delivering scalable, reliable infrastructure that aligns with next-generation AI factory deployments.” The CDU forms part of Schneider Electric’s wider liquid cooling portfolio, which includes systems ranging from lower-capacity deployments through to multi-megawatt installations. Motivair says the units are designed as modular building blocks, enabling operators to select and combine systems based on specific performance and redundancy requirements. The system is manufactured through Schneider Electric's facilities in North America, Europe, and Asia, and is intended to provide high flow rates and pressure within a compact footprint. The company adds that the design supports parallel filtration, real-time monitoring, and integration with other cooling components to support efficient operation across the data centre. The MCDU-70 is now available to order globally. For more from Schneider Electric, click here.

Sabey Data Centers partners with OptiCool Technologies

Sabey Data Centers, a data centre developer, owner, and operator, has announced a partnership with OptiCool Technologies, a US manufacturer of refrigerant-based cooling systems for data centres, to support higher-density computing requirements across its US facilities. The collaboration sits within Sabey’s integrated cooling programme, which aims to ease adoption of liquid cooling approaches as processing demand increases, particularly for AI applications. Sabey says the partnership will broaden the range of cooling technologies available to customers across its portfolio, providing a practical route to denser deployments. OptiCool supplies two-phase refrigerant pumped systems designed for data centre use. The non-conductive refrigerant absorbs heat at the rack through phase change, removing heat without chilled water, large mechanical infrastructure, or significant data hall changes. Sabey states that the method can support increased density while reducing energy use and simplifying plant design. Supporting higher-density and liquid cooling uptake John Sasser, Chief Technical Officer at Sabey Data Centers, comments, “Partnering with OptiCool allows us to offer a cooling pathway that is both efficient and flexible. "Together, we’re making it easier for customers to deploy advanced liquid cooling while maintaining the operational clarity and reliability they expect.” Lawrence Lee, Chief Channel Officer at OptiCool, adds, “By working with Sabey, we’re able to bring our two-phase refrigerant systems into facilities designed to support the next generation of compute. "This partnership helps customers move forward with confidence as they transition to more advanced cooling architectures.” For more from Sabey Data Centers, click here.

Vertiv launches new MegaMod HDX configurations

Vertiv, a global provider of critical digital infrastructure, has introduced new configurations of its MegaMod HDX prefabricated power and liquid cooling system for high-density computing deployments in North America and EMEA. The units are designed for environments using artificial intelligence and high-performance computing and allow operators to increase power and cooling capacity as requirements rise. Vertiv states the configurations give organisations a way to manage greater thermal loads while maintaining deployment speed and reducing space requirements. The MegaMod HDX integrates direct-to-chip liquid cooling with air-cooled systems to meet the demands of pod-based AI and GPU clusters. The compact configuration supports up to 13 racks with a maximum capacity of 1.25 MW, while the larger combo design supports up to 144 racks and power capacities up to 10 MW. Both are intended for rack densities from 50 kW to above 100 kW. Prefabricated scaling for high-density sites The hybrid architecture combines direct-to-chip cooling with air cooling as part of a prefabricated pod. According to Vertiv, a distributed redundant power design allows the system to continue operating if a module goes offline, and a buffer-tank thermal backup feature helps stabilise GPU clusters during maintenance or changes in load. The company positions the factory-assembled approach as a method of standardising deployment and planning and supporting incremental build-outs as data centre requirements evolve. The MegaMod HDX configurations draw on Vertiv’s existing power, cooling, and management portfolio, including the Liebert APM2 UPS (uninterruptible power supply), CoolChip CDU (cooling distribution unit), PowerBar busway system, and Unify infrastructure monitoring. Vertiv also offers compatible racks and OCP-compliant racks, CoolLoop RDHx rear door heat exchangers, CoolChip in-rack CDUs, rack power distribution units, PowerDirect in-rack DC power systems, and CoolChip Fluid Network Rack Manifolds. Viktor Petik, Senior Vice President, Infrastructure Solutions at Vertiv, says, “Today’s AI workloads demand cooling solutions that go beyond traditional approaches. "With the Vertiv MegaMod HDX available in both compact and combo solution configurations, organisations can match their facility requirements while supporting high-density, liquid-cooled environments at scale." For more from Vertiv, click here.

Data centre cooling options

Modern data centres require advanced cooling methods to maintain performance as power densities rise and workloads intensify. In light of this, BAC (Baltimore Aircoil Company), a provider of data centre cooling equipment, has shared some tips and tricks from its experts. This comes as the sector continues to expand rapidly, with some analysts estimating an 8.5% annual growth rate over the next five years, pushing the market beyond $600 billion (£445 billion) by 2029. AI and machine learning are accelerating this trajectory. Goldman Sachs Research forecasts a near 200TWh increase in annual power demand from 2024 to 2030, with AI projected to represent almost a fifth of global data centre load by 2028. This growth places exceptional pressure on cooling infrastructure. Higher rack densities and more compact layouts generate significant heat, making reliable heat rejection essential to prevent equipment damage, downtime, and performance degradation. The choice of cooling system directly influences efficiency and Total-power Usage Effectiveness (TUE). Cooling technologies inside the facility Two primary approaches dominate internal cooling: air-based systems and liquid-based systems. Air-cooled racks have long been the standard, especially in traditional enterprise environments or facilities with lower compute loads. However, rising heat output, hotspots, and increased energy consumption are testing the limits of air-only designs, contributing to higher TUE and emissions. Liquid cooling offers substantially greater heat-removal capacity. Different approaches to this include: • Immersion cooling, which submerges IT hardware in non-conductive dielectric fluid, enabling efficient heat rejection without reliance on ambient airflow. Immersion tanks are commonly paired with evaporative or dry coolers outdoors, maximising output while reducing energy use. The method also enables denser layouts by limiting thermal constraints. • Direct-to-chip cooling, which channels coolant through cold plates on high-load components such as CPUs and GPUs. While effective, it is less efficient than immersion and can introduce additional complexity. Rear door heat exchangers offer a hybrid path for legacy sites, removing heat at rack level without overhauling the entire cooling architecture. Heat rejection outside the white space Once captured inside the building, heat must be expelled efficiently. A spectrum of outdoor systems support differing site priorities, including energy, water, and climate considerations. Approaches include: • Dry coolers — These are increasingly used in water-sensitive regions. By using ambient air, they eliminate evaporative loss and offer strong Water Usage Effectiveness (WUE), though typically with higher power draw than evaporative systems. In cooler climates, they benefit from free cooling, reducing operational energy. • Hybrid and adiabatic systems — These offer variable modes, balancing energy and water use. They switch between dry operation and wet operation as conditions change, helping operators reduce water consumption while still tapping evaporative efficiencies during peaks. • Evaporative cooling — Through cooling towers or closed-circuit fluid coolers, this remains one of the most energy-efficient options where water is available. Towers evaporate water to remove heat, while fluid coolers maintain cleaner internal circuits, protecting equipment from contaminants. With data centre deployment expanding across diverse climates, operators increasingly weigh water scarcity, power constraints, and sustainability targets. Selecting the appropriate external cooling approach requires evaluating both consumption profiles and regulatory pressures. For more from BAC, click here.

Southco develops blind-mate mechanism for liquid cooling

Southco, a US manufacturer of engineered access hardware including latches, hinges, and fasteners, has developed a high-tolerance blind-mate floating mechanism designed for next-generation liquid-cooled data centres. The company says the design is intended to address mechanical tolerance challenges that affect cooling system efficiency and operational stability. It notes that demand for liquid cooling is increasing as traditional air-cooling methods struggle to manage higher power densities associated with AI workloads and high-performance computing. Adoption is accelerating further as operators pursue sustainability and targeted PUE reductions. Liquid cooling, however, requires reliable physical connections, with Southco highlighting that even small alignment deviations at manifold and cold-plate interfaces can disrupt coolant flow, increase pump energy consumption, and heighten the risk of leaks. Managing mechanical deviation in liquid cooling systems Citing guidance in the Open Compute Project’s rack-mounted manifold requirements, Southco notes that a 1mm deviation can raise flow resistance by 15%, leading to around a 7% increase in pump energy. In large facilities, these effects scale alongside thousands of connection points. The company identifies several contributors to misalignment in operational environments: • Accumulated tolerances between rack formats, including EIA-310-D and ORV3, which may reach ±3.2mm • Displacement caused by vibration during transport and operation • Thermal expansion of materials, including copper manifolds expanding more than 1mm over typical temperature ranges Rigid, low-tolerance couplings can leave systems vulnerable to leaks, rising operational costs, and downtime risk, and the newly introduced blind-mate floating mechanism is designed to absorb movement and compensate for these deviations. The product offers floating tolerance of ±4mm radially, axial displacement absorption up to 6mm, and automatic self-centring when disconnected. The design is intended to support long-term leak prevention and meet standards applicable to OCP and ORV3 liquid cooling deployments. Southco adds that the mechanism includes sealing rated to withstand high-pressure testing in line with ASME B31.3 requirements and is intended to support more than ten years of continuous operation. It uses universal quick-disconnect interfaces to enable “blind” maintenance without precise alignment or tooling. The company positions the technology as a step towards enabling rapid maintenance, reducing equipment handling time, and lowering the risk of service interruption. It also points to reduced energy used by pumps through lower flow resistance. Southco sees future development in integrating sensing for temperature, flow, and pressure; exploring lighter materials; and working towards greater standardisation across suppliers and data centre ecosystems.

LiquidStack secures 300MW CDU order from major US operator

LiquidStack, a global company specialising in liquid cooling for data centres, has announced a 300-megawatt order of coolant distribution unit (CDU) capacity from a major US-based data centre operator. The multi-site order will support AI-ready data centre deployments and highlights accelerating demand for scalable liquid cooling solutions for high-density AI workloads. The order comprises LiquidStack’s high-capacity CDU-1MW, designed to support rapid deployment, high-performance, operational efficiency, and future scalability for the next-generation of data centre environments. Liquid cooling for AI infrastructure The customer, a long-established operator with a growing portfolio of AI-ready facilities across the United States, selected LiquidStack as its liquid cooling partner to support the expansion of AI-ready, high-density infrastructure. LiquidStack says its manufacturing and delivery capabilities enable the accelerated fulfilment of the 300-megawatt order, supporting aggressive build-out timelines across the multiple sites. “Orders of this size signal a clear inflection point for liquid cooling,” says Joe Capes, CEO of LiquidStack. “Operators are committing to liquid cooling as core infrastructure for AI, and LiquidStack is uniquely positioned to support that transition at scale.” This announcement follows continued momentum for LiquidStack, including its inclusion on NVIDIA’s Recommended Vendor List for CDUs, the expansion of its manufacturing capacity in Carrollton, Texas, and increasing adoption of its CDU platforms to support AI and accelerated computing workloads. For more from LiquidStack, click here.

SPAL targets data centre cooling needs

SPAL Automotive, an Italian manufacturer of electric cooling fans and blowers, traditionally for automotive and industrial applications, is preparing to showcase its cooling technology at Data Centre World in London in March 2026, with a particular focus on brushless drive water pumps used in data centre thermal management. The pumps are designed for stationary applications where cooling demand is continuous and high. They feature software control compatibility - including CAN, PWM, and LIN - supporting precise regulation of coolant flow and temperature. The company says the pumps consume less power than mechanically driven units and use IP6K9K-rated brushless systems intended to mitigate issues such as overload, reverse polarity, and overvoltage. The role of cooling components in data centres Alongside its pumps, SPAL will display its wider cooling portfolio, which includes fans and blowers designed for controlled airflow and heat dissipation. The company plans to highlight the use of matched replacement components, particularly for systems that rely on coordinated assemblies of fans, pumps, and related controls. James Bowett, General Manager at SPAL UK, says, “In a world where costs are constantly under pressure, it’s false economy to opt for cheaper parts as this will not only affect the performance of the component itself, but the entire suite of parts within a system. "The only way to ensure effective, reliable, long-life operation is to replicate the set up installed at the point of manufacture. That means choosing the best calibre parts throughout.” SPAL states that its products are supplied with a four-year manufacturer’s warranty and are used to help maintain stable conditions for sensitive electronics. The company highlights that the growth of data centres linked to AI and cloud services is increasing demand for equipment designed specifically for energy efficiency, water use, and controlled cooling. SPAL will exhibit at Data Centre World on Stand F15, held at ExCeL London on 4–5 March 2026.

Ireland’s first liquid-cooled AI supercomputer

CloudCIX, an Irish provider of open-source cloud computing platforms and data centre services, and AlloComp, an Irish provider of AI infrastructure and sustainable computing solutions, have announced the deployment of Ireland’s first liquid-cooled NVIDIA HGX-based supercomputer at CloudCIX’s facility in Cork, marking a development in the country’s AI and high performance computing (HPC) infrastructure. Delivered recently and scheduled to go live in the coming weeks, the system is based on NVIDIA’s Blackwell architecture and supplied through Dell Technologies. It represents an upgrade to the Boole Supercomputer and is among the first liquid-cooled installations of this class in Europe. The upgraded system is intended to support industry users, startups, applied research teams, and academic spin-outs that require high performance, sovereign compute capacity within Ireland. Installation and infrastructure requirements The system stands more than 2.5 metres tall and weighs close to one tonne, requiring structural modifications during installation. Costellos Engineering carried out the building works and precision placement, including the creation of a new access point to accommodate the liquid-cooled rack. Jerry Sweeney, Managing Director of CloudCIX, says, “More and more Irish companies are working with AI models that demand extreme performance and tight control over data. This upgrade gives industry, startups, and applied researchers a world-class compute platform here in Ireland, close to their teams, their systems, and their customers.” The project was led by AlloComp, CloudCIX’s AI infrastructure partner, which supported system selection, supply coordination, and technical deployment. AlloComp co-founder Niall Smith comments, “[The] Boole Supercomputer upgrade represents a major step forward in what’s possible with AI, but it also marks a fundamental shift in the infrastructure required to power it. "Traditional data centres average roughly 8 kW per rack; today’s advanced AI systems are already at [an] unprecedented 120 kW per rack, and the next generation is forecast to reach 600 kW. Liquid cooling is no longer optional; it is the only way to deliver the density, efficiency, and performance for demanding AI workloads.” Kasia Zabinska, co-founder of AlloComp, adds, “Supporting CloudCIX in delivering Ireland’s first liquid-cooled system of this type is an important milestone. The result is a high-density platform designed to give Irish teams the performance, control, and sustainability they need to develop and deploy AI.” The system is expected to support larger model training, advanced simulation, and other compute-intensive workloads across sectors including medtech, pharmaceuticals, manufacturing, robotics, and computer vision. CloudCIX says it will begin onboarding customers as part of its sovereign AI infrastructure offering in the near term. To mark the deployment, CloudCIX and AlloComp plan to host a national event in January 2026, focused on supercomputing and next-generation AI infrastructure. The event will bring together industry, research, and policy stakeholders to discuss the role of sovereign and energy-efficient compute in Ireland’s AI development.

Motivair by Schneider Electric introduces new CDUs

Motivair, a US provider of liquid cooling solutions for data centres and AI computing, owned by Schneider Electric, has introduced a new range of coolant distribution units (CDUs) designed to address the increasing thermal requirements of high performance computing and AI workloads. The new units are designed for installation in utility corridors rather than within the white space, reflecting changes in how liquid cooling infrastructure is being deployed in modern data centres. According to the company, this approach is intended to provide operators with greater flexibility when integrating cooling systems into different facility layouts. The CDUs will be available globally, with manufacturing scheduled to increase from early 2026. Motivair states that the range supports a broader set of operating conditions, allowing data centre operators to use a wider range of chilled water temperatures when planning and operating liquid cooled environments. The additions expand the company’s existing liquid cooling portfolio, which includes floor-mounted and in-rack units for use across hyperscale, colocation, edge, and retrofit sites. Cooling design flexibility for AI infrastructure Motivair says the new CDUs reflect changes in infrastructure design as compute densities increase and AI workloads become more prevalent. The company notes that operators are increasingly placing CDUs outside traditional IT spaces to improve layout flexibility and maintenance access, as having multiple CDU deployment options allows cooling approaches to be aligned more closely with specific data centre designs and workload requirements. The company highlights space efficiency, broader operating ranges, easier access for maintenance, and closer integration with chiller plant infrastructure as key considerations for operators planning liquid cooling systems. Andrew Bradner, Senior Vice President, Cooling Business at Schneider Electric, says, “When it comes to data centre liquid cooling, flexibility is the key with customers demanding a more diverse and larger portfolio of end-to-end solutions. "Our new CDUs allow customers to match deployment strategies to a wider range of accelerated computing applications while leveraging decades of specialised cooling experience to ensure optimal performance, reliability, and future-readiness.” The launch marks the first new product range from Motivair since Schneider Electric acquired the company in February 2025. Rich Whitmore, CEO of Motivair, comments, “Motivair is a trusted partner for advanced liquid cooling solutions and our new range of technologies enables data centre operators to navigate the AI era with confidence. "Together with Schneider Electric, our goal is to deliver next-generation cooling solutions that adapt to any HPC, AI, or advanced data centre deployment to deliver seamless scalability, performance, and reliability when it matters most.” For more from Schneider Electric, click here.

Supermicro launches liquid-cooled NVIDIA HGX B300 systems

Supermicro, a provider of application-optimised IT systems, has announced the expansion of its NVIDIA Blackwell architecture portfolio with new 4U and 2-OU liquid-cooled NVIDIA HGX B300 systems, now available for high-volume shipment. The systems form part of Supermicro's Data Centre Building Block approach, delivering GPU density and power efficiency for hyperscale data centres and AI factory deployments. Charles Liang, President and CEO of Supermicro, says, "With AI infrastructure demand accelerating globally, our new liquid-cooled NVIDIA HGX B300 systems deliver the performance density and energy efficiency that hyperscalers and AI factories need today. "We're now offering the industry's most compact NVIDIA HGX B300 options - achieving up to 144 GPUs in a single rack - whilst reducing power consumption and cooling costs through our proven direct liquid-cooling technology." System specifications and architecture The 2-OU liquid-cooled NVIDIA HGX B300 system, built to the 21-inch OCP Open Rack V3 specification, enables up to 144 GPUs per rack. The rack-scale design features blind-mate manifold connections, modular GPU and CPU tray architecture, and component liquid cooling. The system supports eight NVIDIA Blackwell Ultra GPUs at up to 1,100 watts thermal design power each. A single ORV3 rack supports up to 18 nodes with 144 GPUs total, scaling with NVIDIA Quantum-X800 InfiniBand switches and Supermicro's 1.8-megawatt in-row coolant distribution units. The 4U Front I/O HGX B300 Liquid-Cooled System offers the same compute performance in a traditional 19-inch EIA rack form factor for large-scale AI factory deployments. The 4U system uses Supermicro's DLC-2 technology to capture up to 98% of heat generated by the system through liquid cooling. Supermicro NVIDIA HGX B300 systems feature 2.1 terabytes of HBM3e GPU memory per system. Both the 2-OU and 4U platforms deliver performance gains at cluster level by doubling compute fabric network throughput up to 800 gigabits per second via integrated NVIDIA ConnectX-8 SuperNICs when used with NVIDIA Quantum-X800 InfiniBand or NVIDIA Spectrum-4 Ethernet. With the DLC-2 technology stack, data centres can reportedly achieve up to 40% power savings, reduce water consumption through 45°C warm water operation, and eliminate chilled water and compressors. Supermicro says it delivers the new systems as fully validated, tested racks before shipment. The systems expand Supermicro's portfolio of NVIDIA Blackwell platforms, including the NVIDIA GB300 NVL72, NVIDIA HGX B200, and NVIDIA RTX PRO 6000 Blackwell Server Edition. Each system is also NVIDIA-certified. For more from Supermicro, click here.