In China’s data center market, compliance now picks the site, not cost.
by Miranda Dai, Tractus China
For most of the past decade, a foreign company siting a data center in China asked the same questions it would ask anywhere. Where is power cheapest? Where does the climate keep cooling bills down? Where is land available?
Those questions still matter, but they no longer come first. Since late 2024, the binding constraint has been ownership. Where a foreign investor can legally own and control the asset now decides the map, and cost optimization happens inside that boundary.
China runs the world’s second-largest pool of computing capacity, based on both aggregate and intelligent computing measures. The buildout sits under one national framework: the National Integrated Computing Network paired with the “East Data, West Computing” (EDWC) project, which routes data generated in the crowded, power-constrained east to eight national computing hubs for processing. Those hubs span the Beijing-Tianjin-Hebei cluster, the Yangtze River Delta, the Guangdong-Hong Kong-Macao Greater Bay Area, Chengdu-Chongqing, Inner Mongolia, Guizhou, Gansu and Ningxia, with 10 more state-level clusters planned.
The scale behind it is large. According to the Tianjin Cyberspace Administration, EDWC has drawn more than RMB 1 trillion (US$147.7 billion) in investment since the start of the 14th Five-Year Plan and spurred over 30 new computing-focused towns. National computing capacity has grown at close to 30% a year, more than 70% of new capacity has gone west and about 80% of the country’s intelligent computing output now comes from the designated hubs.
CAICT reports an average in-service PUE of 1.42 and more than 140 facilities holding Grade 4A or higher green low-carbon certification. By the end of June 2025, CAICT’s computing power index recorded in-service stock of 10.85 million standard racks and 788 EFLOPS of intelligent computing capacity at FP16 precision, with about 72.6% of racks inside the hub provinces.
Why Ownership Now Leads
The shift comes down to a policy change at the end of 2024. China designated computing assets as critical information infrastructure, which moves them out of ordinary commercial investment and under national security oversight. At the same time, it opened a narrow door. In four pilot areas — Beijing, Lingang in Shanghai, Shenzhen and Hainan — foreign investors can now wholly own and operate internet data centers (IDCs) once they clear the statutory compliance review. Everywhere else, including across the eight national hubs, foreign equity is capped at 50%, so the asset has to run as a Chinese-majority joint venture.
Ownership is only the first gate. Every foreign-backed IDC project, whether greenfield, an acquisition or an existing operation, goes through a three-part review before investment: a foreign investment security review, a cybersecurity review and a data compliance review. The vetting covers equity control, data sovereignty, supply chain integrity and the risk of unauthorized cross-border data flows. Both wholly owned and joint-venture operations then have to abide by ongoing rules on data localization, data classification, regular security audits and prompt incident reporting.
Based on how the approval process runs today, a wholly foreign-owned approval typically takes 8 to 12 months, against 5 to 7 months for a JV with a local partner. Timelines tend to run long because the application materials often need reworking across several review stages. The sticking point is cybersecurity audit trails for cross-border data flows. Foreign firms usually follow global audit standards, and those rarely satisfy China’s requirements for complete cross-border data logs and long-term record retention, which is where most applications stall.
The Cost Layer Sits Inside the Constraint
Once the ownership route is settled, the familiar cost questions come back, and here the west has clear advantages. Data centers need stable, low temperatures to hold down cooling costs, and the western hubs have the climate for it. Inner Mongolia, Ningxia and Gansu have low annual temperatures and short frost-free periods, which allows 6 to 10 months of free-air cooling a year in most areas and cuts mechanical cooling hours sharply. Gui’an in Guizhou has mild, steady temperatures and stable geology, with good cooling conditions year-round. As a result, western sites typically run at a PUE of 1.15 to 1.25, against an eastern average of 1.4 to 1.6, according to CAICT’s green computing power report.
Power is the bigger lever. Electricity is 50% to 70% of operating cost at a conventional data center, and more at power-hungry AI clusters, according to CAICT. Eastern coastal hubs face high demand and tight energy quotas, with industrial power at CNY 0.6 to 0.8 per kWh. Western hubs, drawing on abundant wind, solar and hydropower through mechanisms such as power-computing integration and direct renewable supply, bring that down to CNY 0.28 to 0.40 per kWh, with renewable penetration above 80% on average.
“Foreign firms usually follow global audit standards, and those rarely satisfy China’s requirements for complete cross-border data logs and long-term record retention, which is where most applications stall.”
Cheaper power and cheaper cooling together cut large-scale operating costs by more than 30%. The trade-off is latency. Western sites are too far from users for real-time, interactive workloads.
Two Routes, Determined by Workload
That leaves foreign investors with two workable routes, and the right one depends on the workload.
For high-value, latency-sensitive work, financial trading, proprietary enterprise cloud, cross-border office access and real-time model inference, the four pilot areas are the better fit through a wholly foreign-owned structure. Power and maintenance cost more there, but full ownership gives direct control over data governance and the low latency these services need, while meeting global regulatory and security standards.
For power-hungry, latency-tolerant workloads, large AI model training, cold-data archiving, offline batch processing and disaster recovery, the western hubs make more sense, particularly the newly built clusters in Inner Mongolia, Guizhou and Gansu. Inner Mongolia and Ningxia still rely heavily on coal across their grids, but the data center clusters inside the national hubs are targeted to reach over 80% renewable supply through dedicated wind and solar and power-computing integration. With low power costs and free-air cooling on top, that cuts long-run operating expense by more than 30%.
For most multinationals, the sensible approach is a hybrid. Keep latency-sensitive, data-sovereign workloads in a wholly-owned facility in one of the four pilot areas for full operational control, and place training, archiving and non-core workloads in a compliant domestic JV to speed up licensing. The most common mistake when picking a JV partner in a hub province is to value government connections over a proven operating track record and clean governance. Relationship-led partners often cannot deliver technically, or they bring hidden compliance risk.
The direction of travel is fairly clear. Expect tighter standardized oversight, a gradual widening of market access and stricter green energy benchmarks. As the pilot matures, access rules should become more transparent and the east-west split of computing work more defined. For foreign investors, the era of choosing a site on cost alone is over. The lasting principle is to match the workload to the region’s rules, and to put compliance ahead of cost.
Miranda Dai is a consultant at Tractus China. Analysis is based on public policies, industry reports and Tractus’ client advisory experience as of June 2026.