China's AI Infrastructure and Energy Strategy
Grid transmission infrastructure in Ningxia province, with solar farm in the distance. Image captured by Anne-Amélie Campant.
Analyzing the interplay of artificial intelligence and electricity demand in China is crucial as the nation with the world’s second-largest data center energy load.
intro
Artificial intelligence is reshaping global energy systems, with global electricity demand from data centers projected to grow five times faster than overall electricity demand in the next decade. This dynamic is particularly consequential in China, which has the world’s second-largest data center energy load. China is also responsible for roughly a third of global carbon dioxide (CO2) emissions, while its government is committed to peaking CO2 emissions before 2030 and reaching net zero emissions before 2060 (the “dual carbon goals.”). This work examines how far the expansion of China’s AI infrastructure aligns with these dual carbon goals as well as other sustainability metrics. It provides a comprehensive assessment of the compute demand system, grid integration, cooling and water use, and supply chain embodied emissions.
This research integrates an applied analytical framework combining energy systems analysis, lifecycle emissions assessments, and policy evaluation. The study draws on Chinese government planning outputs, international sustainability and energy datasets, technical literature, and learnings from individuals and firms who are stakeholders in the industry. Case study examples of the Eastern Data, Western Computing initiative, Chinese hyperscale deployments, and AI-enabled grid applications support the analysis with empirical evidence.
The analysis finds that China’s AI ecosystem interacts in complex ways with the nation’s climate commitments: it neither entirely undermines them, nor is it yet comfortably integrated with them.
Current gains in operational data center efficiency reduce electricity consumption at the facility level but transfer environmental pressure to other parts of the system, including water consumption and upstream emissions. Liquid cooling technologies illustrate this pattern: they lower electricity intensity at times at the cost of higher water intensity, a cost felt more acutely in China’s arid provinces. Record renewable deployment across the country does just enough to meet new electricity demand but is insufficient to displace the industry’s reliance on coal. Metrics used to evaluate facility-level efficiency, in turn, fail to capture significant embodied emissions in upstream processes. This finding is particularly important because China has a meaningful share of the industries that AI infrastructure relies upon, like steel and cement. The environmental impact of data centers’ embodied emissions in China is therefore substantial.
The study identifies a central challenge of systems coordination and contributes six recommendations centered on lifecycle carbon accounting, grid-aware compute siting, dual resource-efficiency standards, upstream clean energy mandates, and strengthened emissions disclosure.
If adopted, these recommendations would reposition AI as a structural enabler rather than a pressure on China's green transition.
Adapted from Work Submitted in Partial Fulfilment of the Requirements for the Degree of Master’s in Global Affairs at Tsinghua University. Advised by Dr. David Sandalow.
