In recent years, as demand for computing power has surged—particularly in applications such as artificial intelligence, large-model training, and high-performance computing (HPC)—the limitations of conventional air-cooling systems in terms of heat dissipation and energy efficiency have become increasingly apparent. Liquid cooling, with its significantly higher cooling efficiency compared with air cooling, has demonstrated substantial advantages in data-center energy conservation and the optimization of compute density.

Meanwhile, as global efforts to conserve energy, reduce emissions, and achieve carbon neutrality gain momentum, many countries are continuously tightening energy-efficiency requirements for data centers, with numerous policies explicitly setting Power Usage Effectiveness Objective. Against this backdrop, liquid cooling has evolved from a “optional solution” to an “essential foundational requirement for computing infrastructure.”

Liquid cooling systems utilize cold plates or Immersion coolant Direct contact with the heating element enables highly efficient heat removal. Compared with air, liquids have significantly higher thermal conductivity and specific heat capacity, resulting in greater heat-transfer efficiency and reduced heat accumulation for a given amount of heat input, thereby lowering cooling energy consumption.

Practice has demonstrated that liquid cooling can significantly reduce the PUE of data centers. Traditional air-cooled data centers typically have a PUE in the range of 1.5 to 2.0, whereas optimized liquid-cooling systems can achieve a PUE as low as 1.1 to 1.05. This means that a substantial portion of the energy consumed for cooling and data-center infrastructure is eliminated. Over the long term, electricity costs, operating expenses for thermal-management systems, and the power consumption of mechanical equipment such as air conditioners and fans will all decrease markedly.

A deployment case from NVIDIA shows that when liquid cooling handles 74.9% of a data center’s workload, facility power consumption drops by 18.1% compared with a fully air-cooled configuration, and the data center’s total power draw is reduced by 10.2%. This reduction in energy consumption translates into significant decreases in operating costs and electricity expenses over the course of years of operation.

Meanwhile, liquid cooling enables significantly higher cabinet power densities. For compute-intensive workloads such as high-performance computing and AI training, liquid-cooled cabinets can support deployments with power densities of tens of kilowatts or even higher. This not only boosts computational density per unit of floor space but also spreads infrastructure depreciation and management costs over a larger workload, thereby reducing the overall cost per unit of compute.

According to the “In-Depth Research Report on China’s Liquid-Cooled Data Center Market” published by Kezhi Consulting, lower PUE can reduce data center operating costs; although the initial capital expenditure for liquid-cooled data centers is slightly higher, investment payback typically occurs within about two years. In other words, despite the higher upfront investment in liquid cooling, the cost advantages stemming from energy savings and higher server density result in more competitive long-term operating costs.

Therefore, liquid cooling’s advantages—such as low PUE, superior heat dissipation, and reduced noise—are well aligned with the data center industry’s trends toward high energy consumption and high density. For large-scale data centers, cloud service providers, and AI training centers that face high electricity costs, substantial compute demands, and constrained space, the long-term economic benefits and strategic value of liquid cooling are highly significant.

However, liquid cooling is not a “cost-effective solution” for all data centers.

First, the initial deployment cost of liquid-cooling systems is significantly higher than that of traditional air-cooling systems. Liquid-cooling systems require additional equipment and engineering work, such as quick-connect fittings, Christian Democratic Union Components and products such as solenoid valves and coolant systems require customized design and installation, which all increase the initial capital expenditure. According to industry statistics, the upfront investment for liquid cooling is roughly two to three times that of air cooling; moreover, when using immersion-type liquid-cooled cabinets, the cost per cabinet is even higher.

Secondly, liquid-cooling systems have higher operational and maintenance requirements, necessitating regular inspections, maintenance, and replacement of the cooling fluid. Poor management of the liquid coolant can lead to issues such as corrosion, deposit formation, and microbial growth; moreover, leaks or system failures can cause severe damage to servers and even result in downtime. In contrast, air-cooling technology is well-established, with standardized O&M procedures and lower training costs for maintenance personnel.

Thirdly, when existing data centers are equipped with conventional air-cooling systems, retrofitting them for liquid cooling typically entails extensive modifications, including reconfiguring server racks, floor layouts and piping, as well as the pump and coolant distribution systems. Such upgrades may also require downtime, thereby compromising business continuity. This high level of complexity deters many data center operators from making a full-scale transition to liquid cooling.

For small and medium-sized enterprises with limited budgets, modest computing power requirements, low server density, or highly variable workloads, the combination of “high initial investment, stringent O&M requirements, and complex retrofitting” can make it difficult for liquid cooling to deliver on its cost advantages. In other words, the payback period for liquid cooling may be excessively long, and financial constraints coupled with operational risks pose significant barriers to its wider adoption.

Therefore, from an industrial-structure perspective, liquid cooling is more suitable for large-scale, high-density, high-load organizations that prioritize long-term energy savings and economies of scale, whereas it may not be economically viable for small and medium-sized enterprises with traditional workloads and budget constraints.  

Liquid cooling technology offers clear advantages in reducing PUE, enhancing thermal dissipation efficiency, minimizing noise, and increasing cabinet power density, making it particularly well-suited for large-scale, long-term, compute-intensive data centers. Its long-term energy-saving benefits, improved space utilization, and reduced electricity costs typically more than offset the higher initial capital expenditure and operational complexity.

However, this “cost-saving” premise hinges on three critical conditions: first, the deployment scale and computing density must be sufficiently large; second, there must be the capacity to absorb the initial investment and support specialized operations and maintenance; and third, the system must operate over a long period, with stable workloads and high resource utilization. For small and medium-sized enterprises, lightweight service providers, or use cases where immediate computing demand is relatively low, liquid cooling may not deliver significant cost benefits.

Therefore, whether liquid cooling is “more cost-effective” does not have an absolute answer; rather, it is an economic and technical trade-off that depends on the specific application scenario, system scale, and intended use.

For future large-scale AI computing centers, supercomputing infrastructure providers, and cloud service giants, liquid cooling is virtually the only viable option; for some small and medium-sized enterprises and lightweight data center operators, air cooling may still be the preferred choice, or Air-liquid mixture A hybrid solution—part air cooling and part liquid cooling—is adopted to mitigate high costs and complex operations and maintenance.

In summary, liquid cooling has opened up new possibilities for the data center industry and provided a practical pathway toward greener operations, enhanced energy efficiency, and greater computing power. When deployed in the right use cases at the right time, it can indeed reduce costs; however, treating it as a one-size-fits-all solution and promoting it indiscriminately without careful consideration may introduce significant financial and operational risks.