A research report released by Guotai Haitong Securities Co., Ltd. highlights that liquid cooling offers superior heat dissipation, higher rack density, lower fan energy consumption, and better waste heat recovery conditions. It is evolving from an "optional solution" for high-performance computing into essential infrastructure for high-density AI racks. The domestic liquid cooling industry is transitioning from policy-driven initiatives toward standardized, large-scale applications, with regulatory requirements and refined standards jointly lowering barriers in design, testing, and acceptance, thereby accelerating the shift from demonstration projects to widespread deployment.
Why liquid cooling outperforms air cooling
Liquid cooling uses liquid to absorb heat directly near the chip, shortening the heat transfer path from chip to environment while isolating the server cooling loop from the data center facility water loop via a CDU. Compared to air cooling, it provides stronger heat dissipation, higher rack density, lower fan energy consumption, and better waste heat recovery. While air cooling remains the dominant approach for existing data centers, cold plate liquid cooling is emerging as the primary technical direction for new high-power AI rack requirements.
High-power rack ramp-up drives liquid cooling demand
As AI model scale and computational complexity grow, GPU single-chip power consumption is rising sharply, and when combined with increasing GPU counts, cooling pressure expands rapidly from the chip level to the rack level. Liquid cooling is thus transitioning from an optional approach for high-performance computing to a necessary component of high-density AI racks. High-power racks are scaling quickly: NVIDIA has announced that its Rubin NVL72 is now running at CoreWeave, Google Cloud, Microsoft Azure, OCI, and Nebius, with explicit language stating "production is ramping up." Additionally, Uptime Institute surveys show that the share of operators deploying racks above 30kW continues to climb, rising from 19% in 2025 to 24% in 2026, with growth primarily driven by racks exceeding 50kW.
Domestic standards take shape, policy accelerates liquid cooling penetration
China's liquid cooling sector is moving from policy guidance toward standardized, scaled application. The Ministry of Industry and Information Technology requires new large data centers to achieve a PUE below 1.3 and explicitly encourages liquid cooling technology R&D. Shanghai's "Computing Power Pujiang" initiative further mandates that liquid cooling racks account for over 50% of deployments and supports the installation of high-density racks above 100kW. On the standards front, industry specifications for cold plate, immersion, coolant, and energy efficiency testing have been rolled out since 2021, with the cold plate liquid cooling whole-rack standard landing in 2025 and the national cold plate liquid cooling system standard set for release in 2026. Together, policy constraints and standard refinements are reducing the threshold for design, testing, and acceptance, driving liquid cooling from pilot projects toward faster, large-scale adoption.