Cathay Pacific Haitong: High-power platform starts mass production and increases liquid cooling penetration rate
The Zhitong Finance App learned that Cathay Pacific Haitong released a research report saying that liquid cooling has advantages such as strong heat dissipation capacity, high cabinet density, low fan energy consumption, and good waste heat recovery conditions. Liquid cooling is gradually changing from an “optional solution” for high-performance computing to a necessary infrastructure for high-density AI cabinets. The domestic liquid cooling industry is transitioning from policy guidance to standardized and large-scale application. Policy restrictions and standard improvements jointly lower design, testing and acceptance thresholds, and accelerate the penetration of liquid cooling from demonstration projects to large-scale deployment.
Cathay Pacific Haitong's main views are as follows:
Compared with air cooling, liquid cooling efficiency advantages are outstanding
Liquid cooling shortens the “chip-to-environment” heat transfer path by using liquid near the chip to directly absorb heat, and isolates the server cooling loop from the data center facility water loop through the CDU. Compared with air cooling, liquid cooling has advantages such as strong heat dissipation capacity, high cabinet density, low fan energy consumption, and good waste heat recovery conditions. Currently, air cooling is still the mainstream solution for existing data centers, and cold plate liquid cooling is the main technical direction for new demand for high-power AI cabinets.
High power cabinet volume drives demand for liquid cooling
As the scale and computational complexity of AI models increased, the power consumption of a single GPU chip increased rapidly, and the number of GPUs increased by increasing the power consumption of a single chip, rapidly increasing cooling pressure from “chip level” to “cabinet level”, and liquid cooling gradually changed from an “optional solution” for high-performance computing to a necessary infrastructure for high-density AI cabinets.
High-power cabinets are quickly discharged. Nvidia announced that Rubin NVL72 is already running on CoreWeave, Google Cloud, Microsoft Azure, OCI, and Nebius. NVIDIA clearly uses the expression “production is ramping up”. Furthermore, according to Uptime Institute research, operators deploying cabinets with peak power above 30 kW will continue to increase. The proportion of operators with 30 kW or more cabinets will rise from 19% in 2025 to 2026 Of 24%, the increase mainly comes from 50kW or more cabinets.
Domestic industry standards have been implemented, and policies have driven the increase in liquid cooling penetration rate
The domestic liquid cooling industry is transitioning from policy guidance to standardized and large-scale application. The Ministry of Industry and Information Technology requested that the PUE of new large-scale data centers be reduced to less than 1.3 and clearly encouraged research and development of liquid cooling technology; the Shanghai “Computing Power Pujiang” initiative further proposed that liquid cooling cabinets account for more than 50%, supporting the deployment of high-density cabinets above 100 kW.
In terms of standards, communications industry standards such as cold plate, immersion, coolant and energy efficiency tests have been implemented one after another since 2021, standards for cold plate liquid cooling cabinets were implemented in 2025, and national standards for cold plate liquid cooling systems were released in 2026. Policy restrictions and standard improvements jointly lower design, testing and acceptance thresholds, and accelerate the penetration of liquid cooling from demonstration projects to large-scale deployment.