Anxiety over Energy Consumption of AI: Growth Limits and the Green Dilemma

This round of rapid development of AI has triggered two forms of anxiety related to energy consumption. First, whether energy consumption will become a bottleneck for AI industry development. Second, whether the energy consumption of AI will hinder the progress of the green transition—and in the context of China, whether it will affect the achievement of the country’s dual carbon goals. In response to these concerns, we analyze the issue from three dimensions: The energy consumption of AI, the application of AI on both the demand and supply sides of electricity, and the impact of AI on the energy consumption of China’s economic system. The AI industry’s energy consumption mainly stems from two processes: Model training and inference. In the near term, as the AI industry remains in a phase of rapid early-stage growth, constraints mainly come from the supply side. We estimate that by 2030, the electricity consumption of China’s AI-driven intelligent computing centers could reach up to 430bn kWh, accounting for 4.7% of the country’s total electricity consumption. In the long term, AI energy consumption will likely be closely related to the scale of user demand and task complexity. If the task complexity remains constant, a rise in the scale of demand will bring about a linear rise in AI energy consumption. However, growing task complexity could lead to super-linear growth in AI energy consumption. Two forces might temper this spike in energy consumption: Users’ demand for cost-effectiveness may moderate the industry’s pursuit of maximum AI performance, and the push for greater energy efficiency in chips, servers, and data centers will likely intensify as the industry scales up. Can the application of AI technologies across industries play a meaningful role in energy savings? On the power consumption side, AI can improve energy efficiency by enhancing the operational efficiency of individual devices and optimizing the workflows of technical systems. This is particularly relevant in manufacturing, energy use in buildings, and transportation. On the power supply side, AI can also facilitate the development of green electricity by optimizing green power supply and consumption systems. Overall, how will AI’s development impact the green transition process in China? Using a computable general equilibrium (CGE) model, we conducted a systemic analysis of AI’s effects on energy consumption and carbon emissions. The results show that in the short term, the large-scale application of AI in China will bring about a rise in total energy consumption. However, it is still uncertain how AI will impact total carbon emissions and whether AI will help to reduce the intensity of energy consumption and carbon emissions. If China proactively promotes AI applications in energy-intensive industries and accelerates AI’s role in supporting green power development, these efforts may offset the negative effects of rising AI energy consumption on China’s green transition.