China Data Center Industry 2025: AI Driving Unprecedented Growth
China's data center industry has entered a period of explosive growth in 2025, fueled primarily by the massive computational demands of AI model training and inference. The country is building data centers at an unprecedented pace, with total capacity expected to surpass 15 million standard racks by end of 2025. From the traditional internet giants expanding GPU clusters for their proprietary large language models to new entrants building dedicated AI supercomputing facilities, the data center landscape is being fundamentally reshaped.
TL;DR
China's data center market reached 350 billion RMB in 2025 with capacity exceeding 15 million racks. AI-driven demand accounts for over 40% of new capacity additions. GPU cluster deployment grew 200% YoY. Energy consumption from data centers exceeded 300 TWh, prompting the government to mandate green energy targets. Major operators include China Telecom, China Unicom, Alibaba Cloud, and ByteDance.
Key Insights
Total Market Revenue
China's data center market revenue reached approximately 350 billion RMB in 2025, growing 30% year-over-year, driven by surging AI workloads, 5G edge computing expansion, and enterprise cloud migration accelerating across all industries.
Total Rack Capacity
China's total data center rack capacity exceeded 15 million standard racks by end of 2025, with new additions of approximately 3 million racks in the single year, primarily concentrated in the Beijing-Tianjin-Hebei, Yangtze Delta, and Guizhou big data clusters.
AI Workload Share of New Capacity
Over 40% of newly commissioned data center capacity in 2025 was specifically designed and optimized for AI workloads, featuring high-density GPU computing, advanced liquid cooling systems, and ultra-high-speed interconnect networks for large-scale model training.
GPU Cluster Deployment Growth
GPU computing cluster deployment in Chinese data centers grew approximately 200% year-over-year in 2025, as tech giants and AI startups raced to build infrastructure supporting large language models, with individual clusters ranging from 10,000 to over 100,000 GPUs.
Side-by-Side Comparison
| Operator | Total Capacity | AI-Specific | Key Regions | Energy Source |
|---|---|---|---|---|
| China Telecom | 4.5M racks | 800K GPU racks | National coverage | 30% renewable |
| China Unicom | 3.8M racks | 600K GPU racks | National coverage | 25% renewable |
| China Mobile | 3.2M racks | 500K GPU racks | National coverage | 28% renewable |
| Alibaba Cloud | 1.5M racks | 400K GPU racks | Zhangbei, Hangzhou | 45% renewable |
| Tencent Cloud | 1.2M racks | 300K GPU racks | Guizhou, Tianjin | 40% renewable |
| ByteDance | 1.0M racks | 350K GPU racks | Hohhot, Datong | 35% renewable |
| BDT/Nebula AI | 200K GPU racks | 200K GPU racks | Hohhot | 50% renewable |
| Chindata Group | 800K racks | 150K GPU racks | Beijing, Hebei | 42% renewable |
Frequently Asked Questions
The acceleration is driven by three converging forces: the AI arms race requiring massive GPU computing capacity, the 5G rollout creating demand for edge computing infrastructure, and the government's East-Data-West-Compute policy mandating that eastern data processing be distributed to western regions with abundant clean energy. China aims to build the world's largest AI computing infrastructure by 2027 to support its ambition of global AI leadership.
The East-Data-West-Compute (Dongshu Xisuan) policy, launched in 2022, directs data processing workloads from energy-hungry eastern coastal cities to western regions like Guizhou, Inner Mongolia, Gansu, and Ningxia where renewable energy (hydropower, wind, solar) is abundant and cheap. The initiative has established 10 national computing hubs and is China's flagship strategy for aligning data center growth with carbon neutrality goals.
Chinese operators are deploying multiple solutions: liquid cooling (both cold plate and immersion) reducing PUE below 1.2 in advanced facilities, renewable energy procurement through power purchase agreements, waste heat recovery systems for district heating, AI-optimized power management systems, and building data centers in naturally cool regions. The government mandates that new data centers achieve PUE below 1.25 by 2026.
The three telecom operators (China Telecom, China Unicom, China Mobile) collectively operate the largest data center capacity due to their nationwide networks and government-backed infrastructure buildout. Among tech companies, Alibaba Cloud, Tencent Cloud, and ByteDance are the largest operators. Specialized data center REITs and operators like Chindata Group and VNET (21Vianet) also hold significant capacity, primarily serving enterprise and hyperscale clients.
US export controls on advanced AI chips (NVIDIA A100/H100, later A800/H800, and subsequent restrictions) have created significant challenges. Chinese data centers increasingly rely on domestic alternatives like Huawei Ascend 910B, Cambricon MLU370, and Moore Threads GPUs, though these lag NVIDIA in performance. Companies are also stockpiling chips, optimizing model architectures for available hardware, and some are establishing overseas computing facilities to access restricted chips.