The three major operators ramp up their computing power businesses, with the liquid cooling leader's market value surpassing 100 billion yuan.

What are the AI demand drivers behind the recent surge in computing power investment?

Text by Wu Xinzhu

Edited by Li Zhuang

At the recent earnings presentation, the three major carriers all emphasized the importance of computing power business and future deployment. For related industry chain companies, those with a geographical advantage in intelligent computing centers and companies with advanced liquid cooling technology are expected to benefit from the active investment of operators in computing power business.

Recently, the three major operators held earnings presentations one after another. It is understood that by 2025, China Unicom, China Telecom, and China Mobile will have intelligent computing scales of 45 EFLOPS, 46 EFLOPS, and 92.5 EFLOPS respectively. Since 2026, leading internet companies and large model enterprises at home and abroad have seen a rapid increase in token (word unit) consumption, with downstream intelligent computing demand exploding. As the core foundation of AI industry development, the market demand for computing infrastructure continues to expand, opening incremental space for the data center industry chain. Some leading companies in niche fields, such as liquid cooling leader Runze Technology, saw their stock prices start rising at the end of 2025, and by March 25, the stock price was nearly doubled, with a market value exceeding 140 billion yuan.

This image may be AI-generated

Computing services become one of the main directions for operators

Data center market remains buoyant

China Mobile expects its capital expenditure on computing networks to reach 37.8 billion yuan in 2026, a year-on-year increase of 62.4%. In 2025, the company’s revenue from computing services reached 89.8 billion yuan, up 11.1% year-on-year, with data center revenue growing 8.7% year-on-year, including AIDC revenue increasing by 35.4%. Thanks to the rapid growth in intelligent computing demand, intelligent computing services have become the primary engine, with a growth rate of 279%, driving cloud computing service revenue up 13.9% year-on-year. By the end of 2025, the company’s total intelligent computing scale reached 92.5 EFLOPS (trillions of floating-point operations per second, FP16), achieving full specifications from a hundred cards to over ten thousand cards, with over 1.5 million standard IDC racks serving externally.

China Mobile states that computing services are a growth pole for the company’s development. By strengthening the layout of computing networks and enhancing computing supply capacity, it aims to double its revenue by the end of the “14th Five-Year Plan.” Intelligent services are the company’s high ground for winning the future. By promoting innovation in intelligent networks and deploying native AI applications, it also aims to double its revenue by the end of the “14th Five-Year Plan.”

China Unicom’s capital expenditure in 2026 is expected to be around 50 billion yuan, with more than 35% allocated to computing investments. In 2025, the company achieved operating revenue of 392.2 billion yuan, with computing business revenue (computing services + data centers) accounting for over 15%, an increase of 1.1 percentage points over the previous year; data center revenue was 28.1 billion yuan, up 8.5% year-on-year. The company is strengthening resource deployment in “East Data West Computing” hub nodes and green computing power demonstration provinces, with standard rack scale exceeding 1.1 million racks, and has built seven 100-megawatt AIDC parks, with intelligent computing reaching 45 EFLOPS.

China Unicom states that in 2026, it will promote a strong “computing power” strategy. First, with “computing power” as the core, it will strengthen the integrated operation of the computing power service platform, using new algorithms and technologies to promote the integration of cloud, network, edge, and end. Second, it will enhance the competitiveness of Unicom Cloud, accelerate the construction of a “smart entity + Token + AI Cloud” computing operation model, and speed up comprehensive intelligent upgrades of the cloud. Third, it will solidify the three major AI platforms, providing full-process AI services such as data governance, data annotation, model training and inference, and intelligent agent development, accelerating the development of models-as-a-service and agents-as-a-service modes, and promoting the integration of AI with consumer and industrial scenarios.

China Telecom plans to invest about 73 billion yuan in 2026, with 25.5 billion yuan allocated to computing infrastructure, a year-on-year increase of 26%. In 2025, the three basic telecom companies provided data center racks totaling 938k units, with schedulable intelligent computing scale exceeding 94.4 EFLOPS. In terms of computing, China Telecom built the country’s first commercial super-node cluster at the Guangdong-Hong Kong-Macao Greater Bay Area hub, and established provincial inference pools in hot regions like Beijing and Jiangsu, promoting efficient coordination of computing resources. Its own intelligent computing capacity reached 46 EFLOPS; in AIDC (intelligent data centers), it moderately pre-allocated resources at national hub nodes, creating efficient, green, and intelligent computing foundations, with rack power capacity exceeding 3.2 GW, and actively expanding overseas AIDC deployments.

China Telecom states that during the “14th Five-Year Plan,” the company aims to become a leading AI service provider, with token services as the main business line, strengthening original innovation and key core technology breakthroughs, and building high-quality digital intelligence products and services. It will accelerate the upgrade of comprehensive digital infrastructure.

Since 2026, competition among internet cloud vendors’ computing power has intensified, with vendors like DeepSeek, Kimi, Alibaba Qianwen releasing model upgrade versions, expanding training and inference computing demands. Industry competition has shifted from single-model performance to overall system engineering efficiency and computing supply capacity.

Huachuang Securities points out that the price increases by cloud vendors and the expansion of AI expenditure are jointly reshaping the AIDC industry logic, with the market size of AIDC expected to grow at a compound annual rate of 31.5%, and industry barriers shifting from capital investment to technological integration and operational efficiency; the price hikes improve the return expectations for computing assets, and demand explosions raise the industry ceiling, upgrading AIDC from a heavy asset industry to a core infrastructure track with high barriers and high certainty.

Competitive landscape optimization

Valuations of leading data center companies rise

Reviewing the stock price trends of the data center industry chain, the Dongcai Data Center Index and Dongcai Liquid Cooling Concept Index surged in the second half of 2025. From July 1 to August 28, the data center index increased by 26.80%, then stabilized until December 18, and started a new rally on December 19, with trends closely aligned with cloud vendors’ capital expenditure rhythm; the liquid cooling concept index increased by 62.70% from July 1 to October 30, reflecting the concentrated realization of liquid cooling business performance in the semi-annual report.

Runze Technology’s stock price outperformed the index, with a maximum increase of 137.08% from July 1, 2025, to March 25, 2026. In the first half of 2025, the company’s AIDC business focus shifted fully to intelligent computing services. The 2025 earnings forecast shows the company expects a net profit attributable to parent of 1.88 to 1.98 billion yuan, up 5.71%–11.33% year-on-year, mainly driven by the rapid development of the AI industry and continuous breakthroughs in business deployment, with AIDC business achieving leapfrog growth, boosting revenue and profit simultaneously.

From the company’s perspective, in 2025, Runze Technology’s strategic layout underwent a leap, evolving from a regional computing service provider to a nationwide one. By the end of the reporting period, the delivery scale of computing centers in the Beijing-Tianjin-Hebei, Yangtze River Delta, and Greater Bay Area regions was significant, with a continuous release of nationwide deployment results; at the same time, the company has achieved deep coverage of several leading internet and AI companies, with a strategic optimization of terminal customer structure.

Additionally, the company’s delivery scale of computing centers reached a new high. In 2025, about 220 MW of new computing centers were delivered, laying a solid foundation for performance growth; the core capabilities of AIDC business continued to deepen, with successful delivery of industry-leading 100 MW intelligent computing centers. Finally, the scale of liquid cooling technology applications expanded rapidly, and the pace of industrialization accelerated, further strengthening the company’s core competitiveness in the AIDC track.

In 2025, asset securitization in the IDC industry made breakthroughs. Real estate investment trusts (REITs) platforms helped related companies establish a full-cycle capital operation system of “fundraising, investment, management, and exit,” activating existing assets and optimizing capital structure, aiding enterprises in separating heavy and light assets. For example, in August of that year, Southern Runze Technology’s data center REIT (also called “infrastructure fund”) was officially listed on the Shenzhen Stock Exchange, and in February 2026, the company’s infrastructure fund completed a secondary fundraising and established the second phase of infrastructure asset-backed special plan.

On the other hand, stricter energy consumption indicators have increased the threshold for new data center supply, especially in core areas where supply is limited. For example, approvals for energy consumption indicators in the Beijing-Tianjin-Hebei region have significantly slowed, with phased halts and power restrictions, coupled with local energy efficiency constraints, extending project landing cycles. Existing IDC resources are showing scarcity.

Under policy and resource constraints, cloud vendors’ deployment of computing power in “first-tier/around first-tier” and “emerging nodes” is evolving in a complementary manner. In recent years, ultra-large-scale and park-level data centers have gathered significantly in Inner Mongolia and Ningxia, leveraging advantages in land, electricity, green power, and PUE, along with policy support, mainly supporting large-scale training clusters, data storage, and offline analysis—demand insensitive to latency.

Shenwan Hongyuan believes that the long-term AI inference weight is expected to rise and deepen industry coupling, with the importance of low latency and localized capabilities also increasing. These demands are difficult to simply spill over to central and western regions. Beijing-Tianjin-Hebei, Yangtze River Delta, and Guangdong-Hong Kong-Macao Greater Bay Area will drive the development of inference-focused computing industries through industrial clusters and talent factors, potentially further strengthening the momentum for computing infrastructure demand; geographically, around first-tier nodes like Zhangjiakou, Kunshan/Spring City, and Shaoguan will absorb spillover from first-tier cities and form a collaborative echelon, constructing a dynamic pattern of low latency in the east and large-scale computing in the west.

In terms of data center overseas expansion, Shenwan Hongyuan points out that Hong Kong, as a network connection hub between China and the world, has an irreplaceable geographical advantage; data interconnection between Asian regions requires submarine cables, with 11 out of 14 Asian submarine cables connecting to Hong Kong, making it the most important hub. Moreover, under the explosion of inference demand, domestic and foreign CSPs are increasing large-scale computing clusters at core nodes, mainly for latency-sensitive and proximity-based user needs.

GDS Services, an early mover in overseas business, saw its net revenue increase by 10.8% year-on-year to 938k yuan in 2025, with net profit of 959 million yuan, turning around from a loss of 771 million yuan in 2024.

Liquid cooling sector begins to take shape

Heat management companies expand their business

Uptime Institute, a globally recognized data center standards organization and third-party certification body, reports that the average PUE (Power Usage Effectiveness) of data centers worldwide in 2024 was about 1.56. Compared with previous years, this indicates a slowdown in industry energy efficiency improvements. Achieving further efficiency gains requires higher capital expenditure and deeper architectural changes. It is understood that China has set energy efficiency constraints as mandatory targets, requiring that by the end of 2025, new and renovated large and super-large data centers’ energy utilization efficiency be reduced to within 1.25, with national hub data centers not exceeding 1.2 and the national average below 1.5 (as per the “Special Action Plan for Green and Low-Carbon Development of Data Centers” issued by the National Development and Reform Commission, document No. 970 [2024]).

Policy constraints on high-power-consuming intelligent computing centers have elevated liquid cooling from an “optional energy-saving measure” to a “mandatory engineering method” to meet compliance. It is understood that the progress of the “East Data West Computing” project has further reinforced this trend, with some advanced data centers achieving PUE as low as 1.10, setting strict industry benchmarks. Cloud providers are accelerating upgrades to AI data center architectures, pushing liquid cooling technology from early pilot projects toward large-scale deployment.

Huajin Securities points out that the value of liquid cooling extends far beyond reducing PUE. Its higher supply water temperature enables natural cooling and waste heat recovery, elevating the value of data centers from “single-point energy efficiency management” to a strategic level of “campus-level energy and carbon coordination.” Market research firm TrendForce estimates that the penetration rate of liquid cooling technology in AI data centers will jump from 14% in 2024 to 40% in 2026, with continued growth in the coming years.

This application prospect has attracted heat management companies into the liquid cooling industry. For example, InvenSense, after entering the electronic cooling field, has become a mainstream supplier in the communications, computing, and power electronics equipment industries, serving many top clients. In the liquid cooling computing sector, the company has established deep cooperation with server manufacturers, data center integrators, and leading cloud service providers, covering the entire chain from equipment to data centers.

In November 2025, InvenSense announced that the penetration rate of liquid cooling was rapidly increasing, with fierce competition in the domestic market, especially in data center rooms. The company is involved in both data center and server markets, domestically and overseas. The widespread application of liquid cooling technology and solutions offers good development opportunities.

Runze Technology, since delivering China’s first entire building pure liquid-cooled intelligent computing center in 2023, has continuously expanded the deployment scale of liquid-cooled cabinets, accelerating further in the first half of 2025. The company has helped clients deploy single-cabinet kilowatt-level super-node domestic liquid-cooled intelligent computing clusters, upgrading from server-level to cabinet-level and even cross-cabinet resource supply modes.

Runze Technology believes that the core of liquid cooling in the computing center industry lies in the synergy of multiple detailed segments, essentially a process of accumulated experience. In 2023, the company delivered China’s first large-scale entire building liquid-cooled intelligent computing center, gaining long-term operational data and a first-mover advantage. In September 2025, the company disclosed that its Langfang Area B 200 MW intelligent computing center project was designed according to liquid cooling standards and is now progressing with supporting construction, aiming to respond quickly to large-scale liquid cooling demand with ready stock.

Recently, NVIDIA’s first Vera Rubin rack went live on Microsoft Azure. This AI computing platform uses 100% liquid cooling, with water at 45°C, and reduces installation time from two days to two hours, greatly easing data center cooling pressure. Changcheng Securities notes that under power restrictions, higher token throughput at a given power means lower costs. Achieving this requires coordinated hardware and software performance and cooling system configurations. NVIDIA continues to promote liquid cooling solutions in new systems, confirming that increasing AI computing density is driving changes in cooling solutions. Liquid cooling system solution providers and key component suppliers are likely to continue benefiting.

(This article was published in the March 28 edition of “Securities Market Weekly.” The stocks mentioned are for illustrative analysis only and do not constitute investment advice.)