As artificial intelligence models continue to grow in scale, next-generation GPU servers are delivering significantly higher computing density and power consumption. For data center operators, efficiently managing the heat generated by AI infrastructure has become an increasingly important factor in deploying computing capacity.
In August 2026, liquid cooling once again emerged as a major focus across the global data center industry.
From direct-to-chip cooling and immersion cooling to integrated power and thermal management architectures, more companies are investing in liquid cooling technologies.
The trend suggests that liquid cooling is moving beyond its traditional role as a specialized technology for high-performance computing and becoming a critical infrastructure component for next-generation AI data centers.
Traditional data centers primarily rely on air cooling. However, the rapid expansion of AI GPU clusters is significantly increasing rack power density.
Next-generation AI computing platforms such as NVIDIA GB300 NVL72 feature high-density GPU architectures that place substantially greater demands on rack power delivery and thermal management. NVIDIA has also continued to highlight liquid cooling as an important component of high-density AI infrastructure.
Air cooling remains viable for many conventional data center applications, but its scalability becomes increasingly challenging as rack power density rises.
Simply increasing airflow cannot indefinitely improve cooling performance. It can also increase fan power consumption, cooling energy requirements and space constraints.
As a result, AI infrastructure development is forcing the industry to reconsider how heat is managed:
The question is no longer only how to deliver power to AI servers, but also how to efficiently remove the heat generated by that power.
This is one of the key drivers behind the rapid adoption of liquid cooling.
Liquid cooling is not a single technology, but an integrated system involving multiple components.
Direct-to-chip cooling has become one of the most widely discussed approaches for AI servers. Cold plates are positioned directly on high-heat components such as GPUs and CPUs, allowing coolant to remove heat more efficiently.
Other important components include Coolant Distribution Units (CDUs), liquid cooling piping, quick disconnects, manifolds and heat exchangers.
Compared with traditional air cooling, liquid cooling provides greater thermal management capabilities for high-density computing and creates more room for future increases in rack power density.
This is also creating a new data center equipment supply chain.
Traditional procurement has focused largely on servers, switches, storage, UPS systems and HVAC equipment. As AI infrastructure evolves, cold plates, CDUs, liquid-cooled racks and cooling fluids are becoming increasingly important.
Cooling equipment is moving from a supporting role to a core component of data center infrastructure.
On August 17, global electronics and connectivity company Molex announced a strategic investment in liquid cooling startup CAEPlus.
CAEPlus is developing its BoundaryCool active liquid cooling platform to address increasing thermal loads in AI and high-performance computing data centers, including the thermal challenges of next-generation GPUs and ASICs.
Molex said the investment would accelerate the development, validation and commercialization of the platform.
The development is significant because the liquid cooling market is no longer being driven solely by specialized thermal management companies.
Increasingly, electronics, connectivity and infrastructure companies are entering the market.
This suggests that liquid cooling is moving from an early technology-development stage toward commercialization, productization and large-scale deployment.
The competitive focus is also changing.
Liquid cooling providers will need to address not only thermal performance, but also standardization, manufacturing capacity, global supply chains and long-term maintenance.
For large AI data centers, liquid cooling systems will need to become as standardized and scalable as servers and networking equipment.
Another important development in August is the closer integration of data center power and cooling systems.
On August 17, Trane Technologies and Eaton announced a collaboration to develop an integrated power and thermal management reference architecture for AI data centers.
The companies said the solution could improve energy efficiency by up to approximately 15%, reduce installation costs by up to 30% and cut copper usage by up to 80%.
The logic behind this development is straightforward.
Higher GPU power means higher heat generation. At the same time, pumps, CDUs and chillers used in liquid cooling systems also consume electricity.
As a result, power efficiency and cooling efficiency can no longer be considered independently.
Future data centers will increasingly need to optimize the entire infrastructure—from power delivery and server racks to liquid cooling and heat rejection.
The goal is not simply to reduce the energy consumption of an individual component, but to improve the efficiency of the entire facility.
Beyond improving thermal performance, liquid cooling is also being explored as a way to reduce water and energy consumption.
In August, the Malaysian Palm Oil Board announced the development of Sawit EcoTherm, a palm-based immersion cooling fluid.
According to MPOB, the fluid is designed for immersion cooling and could reduce water and energy consumption in data centers. The technology is reportedly approaching commercialization.
The development highlights another emerging direction in liquid cooling:
The competition is no longer only about how efficiently a cold plate transfers heat. Cooling fluids themselves are becoming an area of innovation.
This could be particularly important in regions where water resources are constrained while data center capacity continues to grow.
Future liquid cooling technologies are therefore likely to focus on three objectives:
higher thermal efficiency, lower energy consumption and lower resource requirements.
The developments of August demonstrate that liquid cooling is entering a new phase of industrial development.
AI GPUs continue to deliver higher computing performance, but this also results in higher power consumption and thermal density. Data center operators therefore require more efficient cooling solutions, while equipment manufacturers are increasing investment in liquid cooling technologies and manufacturing capacity.
From Molex's investment in CAEPlus, to Trane and Eaton's integrated power-and-cooling architecture, and Malaysia's exploration of palm-based immersion cooling fluids, these developments point to one common trend:
Liquid cooling is no longer simply an optional technology for AI data centers. It is becoming an essential component of high-density computing infrastructure.
As AI servers continue to evolve, the liquid cooling supply chain is expected to expand further—from GPU cold plates and CDUs to liquid-cooled racks, piping, quick disconnects, cooling fluids and thermal management systems.
For the global data center industry, competition in the AI era is therefore becoming more than a competition for computing power.
It is increasingly a competition for power efficiency, thermal management and infrastructure efficiency.
As higher-power AI servers continue to enter data centers, liquid cooling is likely to expand from large-scale HPC facilities into a broader range of data center environments.
The transition from the “air-cooling era” to the “liquid-cooling era” may become one of the defining infrastructure shifts of the AI data center industry.
