中国数据中心：偏远地区的拐点与变革-China’s Emerging Frontiers – Data Centers Inflection and Disruption in Remote Areas

Summary of Key Points from the Conference Call on China's Data Centers Industry Overview - The focus is on the data center (DC) industry in China, particularly in remote areas, driven by the demand for artificial intelligence (AI) and the "Eastern Data, Western Computing" policy [1][30][40]. Core Insights and Arguments 1. Market Dynamics: The demand for data centers in remote areas is expected to grow significantly, with a projected 66% year-over-year increase in total data center orders, reaching 3.4GW in 2026. Remote areas are forecasted to experience a compound annual growth rate (CAGR) of 54% from 2025 to 2028, compared to 28% in Tier 1 cities [31][88]. 2. Capacity Growth: China's installed data center capacity is expected to reach approximately 44GW by 2028, which will be about 43% of the US capacity at that time. This represents a CAGR of 18% from 2025 to 2028 [2][89]. 3. Shift in Demand: The previous cloud cycle saw a 50:50 split between Tier 1 markets and remote areas. In the AI era, remote areas are anticipated to capture 70-80% of the volume due to scale needs and resource availability [3][88]. 4. Policy Influence: The NDRC's policy is directing new data center deployments to specific hubs in West China, which aligns with the current AI model training requirements. This policy is expected to foster a healthier market by controlling oversupply [3][40][41]. 5. Cost Advantages: Power costs in remote areas like Ulanqab are significantly lower (56% lower than nearby Hebei), making them more attractive for data center operations. The total cost of ownership is also more favorable in these regions [51][48]. Key Players and Recommendations 1. Preferred Companies: GDS Holdings and VNET are highlighted as top picks due to their strategic pivots and resource buildup in remote areas. Other notable players include Centrin and Chindata [4][35]. 2. Emerging Hubs: Ulanqab in Inner Mongolia and Zhongwei in Ningxia are identified as key beneficiaries of the AI era, with Ulanqab currently having a capacity of 490k racks and Zhongwei aiming for 300k racks by the end of 2025 [61][62][114]. 3. Investment Returns: The yield on cost (YoC) for data centers in both Tier 1 markets and remote areas is expected to stabilize around 11%, indicating that the economic viability will depend on scale rather than location [117][118]. Additional Important Insights 1. Construction and Delivery: The construction cycle for data centers has shortened from 1-2 years to 6-9 months, improving responsiveness to market demand and reducing the risk of oversupply [37][106]. 2. Hyperscaler Strategies: Hyperscalers are shifting towards outsourcing data center needs rather than relying solely on in-house facilities, which is a significant change from previous strategies [103]. 3. Latency Considerations: AI workloads are less sensitive to latency, allowing for more flexibility in data center locations compared to the cloud era, where proximity to Tier 1 cities was crucial [56][102]. This summary encapsulates the critical insights and projections regarding the data center industry in China, emphasizing the shift towards remote areas driven by AI demand and supportive government policies.