China is strategically relocating its AI computing infrastructure, with a significant push towards western and inland provinces, moving away from congested eastern cities. This shift is primarily driven by the availability of cheaper land, cooler climates which reduce cooling costs, and access to abundant renewable energy sources. This move aligns with Beijing's "Eastern Data Western Computing" strategy, launched in 2021, to distribute computing power across the country.
Key areas benefiting from this westward expansion include Ulanqab in Inner Mongolia, which has emerged as a major AI computing hub, attracting investments from tech giants like Huawei, Apple, Alibaba, Kuaishou, and Tencent. DeepSeek, an AI pioneer, is planning to build a large data center in Ulanqab with a potential 1 gigawatt of computing capacity, with some capacity expected by late 2027 or early 2028. UCloud's Ulanqab Intelligent Computing Center, covering about 140,000 square meters, is designed for approximately 12,000 server racks, experiencing extremely high utilization rates even before new construction is complete. Similarly, Qinghai province, with its clean energy and cool climate, has become a pilot region for green electricity and computing power, deploying 28,116 petaflops by the first half of 2026. China Mobile's Highland Big Data Center in Haidong, Qinghai, with its initial phase opened in 2015, is expanding with a second phase investment of $346.5 million (2.5 billion yuan), adding 14,000 racks and 32,400 petaflops of computing capacity.
The economic drivers for this decentralization are substantial. Electricity in Ulanqab, for instance, costs about $0.04 to $0.05 (0.32-0.35 yuan) per kilowatt-hour, significantly lower than the $0.08 to $0.12 (0.60-0.90 yuan) in eastern and southern China. This cost difference can amount to as much as $690 million (5 billion yuan) annually for a 1-GW data center. This cheap electricity, combined with cooler temperatures and proximity to green energy sources (critical for meeting requirements of international customers like Apple), makes these western regions highly attractive. While AI model training facilities are moving west of the Hu-Huanyong Line, inference capacity — for answering questions and generating images — is moving eastward towards major AI demand centers in the Yangtze River Delta, Greater Bay Area, and Beijing-Tianjin-Hebei region. For now, computing scarcity is a greater constraint than latency, making large, centrally scheduled clusters in the west more efficient.
China plans to invest approximately $295 billion (2 trillion yuan) over the next five years to build a nationwide network of interconnected computing hubs. State-owned firms like China Mobile and China Telecom will operate these data centers, with an ambition to rely on local suppliers, such as Huawei Technologies Co., for at least 80% of technology like AI chips, aiming to reduce dependence on foreign companies like Nvidia Corp. and Advanced Micro Devices Inc. DeepSeek is reportedly planning a large order of at least 160,000 of Huawei’s next-generation Ascend 950DT accelerators for its new data center in Inner Mongolia, which could establish one of the largest clusters of Huawei AI chips globally.
Despite the significant investment and infrastructure development, researchers have raised questions about the broader local economic benefits. While data centers bring substantial investment in land, servers, and infrastructure, they do not consistently create a large number of permanent jobs, raising concerns about the immediate economic spillover. For example, Guizhou province, despite its high GDP growth, recorded near-bottom wage growth, suggesting that infrastructure investment alone may not guarantee equally strong growth in local incomes. Furthermore, provinces like Guizhou have accumulated significant debt due to heavy spending and incentives to attract data center projects, ranking among the highest in debt-to-revenue and debt-to-GDP ratios nationally, posing a challenge regarding the long-term financial sustainability of these regional government investments.
