According to TrendForce's latest liquid cooling industry research, with NVIDIA GB200 NVL72 rack servers set for volume shipments in 2025, cloud service providers are accelerating upgrades to AI data center infrastructure. This is driving liquid cooling technology from early pilot programs toward large-scale deployment, with penetration rates in AI data centers expected to rise significantly from 14% in 2024 to 33% in 2025, with continued growth in subsequent years.
TrendForce notes that GPU and ASIC chips used in AI servers have dramatically increased power consumption. Taking NVIDIA's GB200/GB300 NVL72 systems as an example, single rack thermal design power (TDP) reaches 130kW-140kW, far exceeding the processing limits of traditional air cooling systems, leading to the adoption of Liquid-to-Air (L2A) cooling technology. Due to constraints from existing data center building structures and water circulation facilities, L2A is expected to become the mainstream transitional cooling solution in the near term.
As new-generation data centers are completed starting in 2025, combined with continuously increasing AI chip power consumption and system density, Liquid-to-Liquid (L2L) architecture is expected to accelerate adoption from 2027 onwards, providing more efficient and stable thermal management capabilities and gradually replacing current L2A technology to become the mainstream cooling solution for AI server rooms.
Currently, the four major North American CSPs continue to increase AI infrastructure investment, launching new waves of data center expansion domestically and in Europe and Asia. These operators are simultaneously building liquid cooling-compatible facilities. Google and AWS have already activated modular buildings with liquid cooling wiring capabilities in the Netherlands, Germany, Ireland, and other locations. Microsoft is conducting liquid cooling pilot deployments in the U.S. Midwest and multiple Asian locations, planning to make liquid cooling systems standard infrastructure starting in 2025.
TrendForce points out that rising liquid cooling penetration rates are driving demand expansion for cooling modules, heat exchange systems, and peripheral components. Cold plates, serving as core components for contact heat exchange, have major suppliers including Cooler Master, AVC, BOYD, and Auras Technology. Except for BOYD, the other three companies have expanded liquid cooling capacity in Southeast Asia to meet high-intensity demand from U.S. CSP clients.
Coolant Distribution Units (CDU) are key modules responsible for heat transfer and coolant distribution in liquid cooling circulation systems, divided into two main categories by deployment method: In-row and Sidecar types. Sidecar CDUs are currently the market mainstream, with Delta Electronics as the leading manufacturer. Vertiv and BOYD are primary suppliers of In-row CDUs, whose products offer stronger cooling capabilities suitable for high-density AI rack deployments.
Quick disconnects (QD) are key components connecting cooling fluid pipelines in liquid cooling systems, with their airtightness, pressure resistance, and reliability being crucial for the safe and stable operation of cooling architectures. Currently, the NVIDIA GB200 project is dominated by international major manufacturers, including CPC, Parker Hannifin, Danfoss, and Staubli, who have gained advantages through existing certification systems and high-end application experience.