- The project will deploy Alibaba’s domestic PPU chips in a 10,000-GPU-equivalent computing pool serving Alibaba Cloud’s AI training operations
- The investment comes as liquid cooling becomes increasingly critical for data centers running power-hungry AI chips
Merscooling Technology (巨冷科技), a Zhejiang-based liquid cooling technology provider, will invest 2.69 billion yuan ($400 million) to build a liquid-cooled AI computing center in Quzhou, Zhejiang Province.
The facility has a planned computing capacity of 1,200 petaflops and a 10,000-chip computing pool based on Alibaba’s domestically developed pulse processing unit (PPU) processors.
The project was officially registered on August 21 and will be located in the Baijiatang area of Quzhou Smart New City.
It will comprise two computing-center buildings and primarily support Alibaba Cloud’s AI model training operations.
From national R&D project to commercial deployment
Merscooling was established in 2024 by Quzhou-based Juhua Group, China’s leading fluorochemical producer, together with Beijing University of Chemical Technology and Hangzhou-based IT service provider Immertech.
The three partners began working together in 2021 under a national key R&D program focused on materials and technologies for liquid cooling in data centers.
Their work has led to advances in single-phase immersion cooling, covering coolant, thermal-management equipment and specialized servers in an integrated solution.
Merscooling says it can provide turnkey services spanning the design, construction and delivery of liquid-cooled data centers.
The company has already delivered liquid-cooling projects for Alibaba Cloud in Shanghai, China Telecom’s Tianyi Cloud and China Mobile’s liquid-cooled computing demonstration project in Gansu’s Qingyang, part of the national east-to-west computing initiative.
Its immersion-cooling technology can reduce power usage effectiveness, or PUE, to below 1.07 and cut energy consumption by more than 40% compared with conventional air-cooled data centers, Merscooling said.
Liquid cooling moves toward the mainstream
The Quzhou project comes as the rapid growth of AI computing is pushing traditional air cooling closer to its physical limits.
Market consultancy TrendForce estimates that liquid cooling will be used in 53% of AI chips in 2026, with penetration approaching 60% in 2027.
Nvidia’s next-generation Rubin platform is designed around full liquid cooling, while Google has deployed liquid cooling across more than 80% of its AI servers, according to the company.
The shift is being driven by rising chip power consumption. As AI accelerators pack more computing capacity into increasingly dense systems, removing heat efficiently has become a critical constraint on data-center performance and energy use.
Quzhou builds an AI infrastructure hub
Quzhou is positioning itself as a major cloud-computing data-center cluster in Zhejiang, supported by access to renewable electricity and its role in the province’s broader computing-network strategy.
The Merscooling project fits directly into that ambition, bringing a domestically developed liquid-cooling solution together with Alibaba’s PPU-based computing infrastructure.
Merscooling’s revenue reached tens of millions of yuan in 2025 and is expected to rise to hundreds of millions of yuan this year.
The company is also drawing on Juhua Group’s decades of chemical-industry expertise to develop an integrated supply chain covering coolants, thermal-management systems and immersion-cooled servers.
For the global AI infrastructure industry, the Quzhou project points to a broader shift in China’s approach to data-center construction: liquid cooling is moving beyond pilot projects and toward large-scale commercial deployment as the country expands domestic AI computing capacity.
