With the significant growth of global computing power, the demand for AI servers has risen sharply, and the power consumption of AI servers has continued to increase due to the large number of high-power CPU and GPU chips.
Chip heat dissipation efficiency has become a key issue, and liquid cooling products are becoming the core focus of data centers.
There are two main types of cooling systems in existing data centers: air-cooled data centers and liquid-cooled data centers.
Currently, air cooling technology dominates data center cooling. According to data from Dawning Digital Innovation, more than 90% of existing data centers use air cooling, and the penetration rate of liquid cooling in domestic data centers is less than 10%.
The power limit of air-cooled technology is 15kW, and the computing power density of AI clusters will generally exceed 20kW per cabinet in the future, so the upgrade demand of liquid cooling technology has become particularly urgent, and it is expected that by 2025, the penetration rate of liquid cooling will exceed 30%.
Pay attention to [Leqing Industry Observation] and gain insight into the industrial pattern!
Panorama of the liquid cooling data center industry chain:
Source: White Paper on Liquid Cooling Technology for Telecom Operations
An overview of the liquid-cooled server industry
Air cooling is the mainstream temperature control solution in data centers, which mainly uses air as the cooling medium. With the advent of the post-Moore's Law era, the increase in server power consumption has led to an increase in temperature, and liquid cooling technology has begun to gain traction.
At present, there are two main cooling technologies in data centers: air cooling and liquid cooling, which achieve cooling effects through heat transfer between air and liquid, respectively.
The critical point of power density of the cabinet from air-cooled to cold-plate heat dissipation is 15-20kW, and the critical point of cold-plate to immersion heat dissipation is 60kW. When server power densities generally exceed 20kW, it will drive the shift from air-cooled to liquid-cooled technology in data centers.
In addition, the average annual PUE value of air-cooled data centers is about 1.75, which cannot meet the minimum requirements stipulated in the policy. The average annual PUE value of data centers with liquid cooling technology is about 1.2, which fully meets the requirements of the policy. In addition, liquid cooling has the advantages of low energy consumption, high heat dissipation efficiency and low noise.
There are two main ways to cool servers:
Source: CCID Consulting's White Paper on Liquid-Cooled Data Centers in China
Cold technology is an efficient cooling method with liquid as the heat dissipation medium, which is mainly divided into three types: cold plate type, immersion type and spray type.
At present, cold plate liquid cooling has the highest maturity in the industry, accounting for up to 90% of the share. The cold plate liquid cooling technology is not only mature and does not need to change the form of the existing server, but also fully meets the needs of AI servers, and is expected to be the first to be promoted and applied.
The development of two-phase immersion liquid cooling is slow due to the fact that two-phase immersion liquid cooling systems can provide better cooling results with higher heat transfer coefficients and heat dissipation limits, but they are challenging to design and relatively expensive.
At present, the global technology giant NVIDIA is actively promoting the hybrid cooling solution combining cold plate and single-phase immersion technology, which is expected to further accelerate the application of immersion technology in high-performance data centers.
In addition, on the outside of the liquid-cooled server, the main adaptation work is concentrated on the liquid-cooled infrastructure such as Manifold, CDU, and cold source.
Competitive landscape of liquid-cooled server market
At present, liquid cooling solutions are being accelerated in the process of promotion, and the improvement of cost and cost performance is an important factor to promote the wide application of liquid cooling. At present, the world's leading manufacturers are vigorously promoting liquid cooling technology, with the main goal of reducing PUE values.
In the early days, the global liquid cooling market was mainly participated by top overseas chip and server manufacturers. In recent years, data center suppliers in Chinese mainland and upstream server manufacturers represented by Taiwan have accelerated the layout of liquid cooling technology.
Among Taiwanese manufacturers, Shuanghong has a wide range of businesses, and the company maintains close cooperation with the world's top chip and server customers.
Chinese mainland manufacturers also started early in liquid cooling technology, mainly leading domestic substitution and accelerating the iteration of liquid cooling products with Inspur Information, Sugon and Xinhua III manufacturers.
According to IDC data, in 2023, the top three of Inspur Information, Ningchang and Super Fusion will occupy about 70% of the market share.
In addition, the fully liquid-cooled server node jointly proposed by Inspur Information and Intel is composed of multiple parts, including node shells, motherboards, CPU chips, memory modules, memory cold plates, etc.
H3C liquid cooling technology solutions:
Source: Xinhua III
From an industry perspective, the Internet, telecommunications, government, and education industries are the main purchasers of liquid-cooled servers. Internet giants such as Alibaba, Baidu and JD.com have taken the lead in liquid cooling technology. The rapid development of AIGC is expected to promote Internet companies to accelerate the large-scale application of liquid cooling technology, thereby further expanding the space of the liquid cooling market.
The barriers to liquid cooling technology itself are not high, but the difficulty of computing power liquid cooling lies in the transformation of servers. Since the server usually carries the core business of the customer, the requirements for stability are very high. Server manufacturers have inherent advantages, and with the gradual opening of the market space in recent years, third-party manufacturers have also accelerated their entry into this market. #服务器##ai服务器##人工智能##算力##科技##财经#
Overall, there are many excellent manufacturers in the field of liquid cooling to provide solutions and related products. In addition to the above-mentioned manufacturers, the main domestic layout manufacturers also include Invic, Feirongda, Gaolan Co., Ltd., Shuguang Digital Innovation, Shenling Environment, Kehua Data, Yimikang, Tongfei Co., Ltd., ZTE, Unisplendour Co., Ltd., Baoxin Software, Runze Technology, Aofei Data, Wangsu Technology, Tuowei Information and Industrial Fortune Union, etc.
The space of the liquid cooling market can be divided into two categories: incremental and stock. The incremental space is mainly driven by the new demand for computing power, and the demand for GPU-based heterogeneous servers for AI large model training continues to grow, resulting in a sharp increase in the demand for liquid cooling for new AI servers. With the maturity of liquid cooling technology and the reduction of costs, the industry is expected to usher in rapid growth.
Pay attention to [Leqing Industry Observation] and gain insight into the industrial pattern!