Energy-guzzling AI industry threatens China’s green transition
China’s rebalancing policy will power the country’s data centres with more renewable energy, says EAI deputy director Chen Gang, but this is far from sufficient to reduce China’s AI’s carbon footprint due to the soaring energy demand from the AI sector and grid bottlenecks.
China has been installing record amounts of green energy in recent years, with more than half of the world’s new wind and solar capacities built in the country in 2023. Nevertheless, the world’s top greenhouse gas emitter is still at risk of missing its key climate targets due to its coal reliance.
Recently a new challenge from the energy-guzzling artificial intelligence (AI) industry loomed large in China’s green transition. Although the government has made some preparations for the upcoming energy spike from AI development and training, any underestimation of such risk may lead to a derailing of China’s low-carbon transition due to the collision of the country’s AI and climate ambitions.
Chinese companies quickly joined the race without thinking much about chatbots’ sizeable carbon footprints.
AI’s footprint to expand
The development of AI requires massive data processing power in data centres, which consume significant amounts of electricity. Tech giants’ goal of reducing their carbon footprints is in jeopardy as they rely on more and more energy-hungry data centres to power their AI products and services.
The birth of ChatGPT has pushed global tech giants to enter the race of developing their own chatbots. Google’s emissions have risen by nearly 50% in recent years due to electricity consumption by data centres and supply chain emissions. The International Energy Agency estimates that data centres’ total electricity consumption could double from 2022 levels to 1,000TWh (terawatt hours) in 2026, approximately Japan’s level of electricity demand.
The rise of ChatGPT has intensified the competition between China and the US in AI. Chinese companies quickly joined the race without thinking much about chatbots’ sizeable carbon footprints. Tech giants, as well as startups including Baidu, Alibaba, Tencent, JD.com, Huawei, SenseTime, Project AI 2.0 etc, have released similar products to ChatGPT. Most of these companies, however, have yet to publicise the emissions data from these new AI services in their ESG (environmental, social, and governance) reports.
Data centres play a crucial role in training and operating AI models. Thanks to China’s huge investment in AI research and development, data centres and related supply chains, the energy consumption and carbon emissions in this area are expected to grow substantially in the next few years. The share of data centres’ electricity consumption in China’s total will rise from 3% in 2022 to above 5% in 2025.
In the future, more and more large-scale AI computing centres will be in the western region to tap the abundant renewable resources there.
China has a plan, but it is not enough
In 2022, China launched the “East Data, West Computing” (东数西算) plan, a mega project to relocate national data centre clusters in less developed yet renewable energy-rich western areas.
Western regions like Inner Mongolia, Guizhou and Gansu province have built national computing hubs and data centres to meet the computing power needs of the economically vibrant eastern region. Much of the energy-intensive processes involving training large AI models and the production of AI hardware are also located in western provinces.
For a long time, a substantial part of the potential energy generated by wind turbines and solar panels in China’s resource-rich western and mid-western areas could not be transmitted to the power grids connecting the coastal areas. Amid excessive investment powered by massive government subsidies, renewable power generators in China have been facing the worst curtailment rates in the world.
In the future, more and more large-scale AI computing centres will be in the western region to tap the abundant renewable resources there. Based on China’s climatic and geographic characteristics, northwestern Gansu, Xinjiang, Inner Mongolia and Tibet are primary provinces for wind and solar power development, while southwestern Sichuan, Yunnan and Guizhou provinces have bountiful hydropower resources.
China plans to relocate the training of many big AI models developed in the eastern region to the western region. This “East Datam West Training” (东数西训) strategy will help China alleviate the energy and emission pressures in the east by increasing the efficiency of green electricity capacities in the west.
Although China is building smart grids and adding green energy storage capacity, the process is not fast enough to catch up with the surging power demand in the AI sector.
Such a rebalancing policy will power China’s data centres with more renewable energy, but it is far from being sufficient to reduce AI’s carbon footprint in the country due to the soaring energy demand from the AI sector and grid bottlenecks.
After the Chinese government released the New Generation of AI Development Plan in 2017, many provincial governments made ambitious plans to develop the AI industry and to build data centres. The energy consumption in China’s data centres has maintained double digit growth since then. The number and size of local data centres under construction or in planning are still growing due to the recent AI frenzy driven by ChatGPT.
About two thirds of China’s total electricity generation still comes from coal-fired thermal plants, so much of the newly added energy needs in data centres every year will still be met by the polluting coal burning. China is building more coal plants, with 50 gigawatts added in 2023, equivalent to two-thirds of the increase in global coal power capacity.
The overdevelopment of renewable energy in the vast western region always faces grid and intermittency bottlenecks. It is difficult for China’s electricity grids connecting local data centres to absorb all the wind or solar energy, with large peaks and troughs depending on the wind speed and solar radiation. Although China is building smart grids and adding green energy storage capacity, the process is not fast enough to catch up with the surging power demand in the AI sector.
AI: a problem and a solution
Paradoxically, the development of AI can provide solutions to these energy bottlenecks. Smart Grids rely on AI technology to optimise the management of grids, use renewable energy sources more efficiently and reduce wastage. AI can be used to enhance energy management systems in buildings and factories, leading to reduced carbon footprints and improved energy efficiency.
In China, AI has been used to improve traffic management and logistics, reducing fuel consumption and emissions from transportation. As China is the world’s largest producer of electric vehicles (EVs), AI technology installed in EVs can be optimised to reduce carbon emissions in the transportation sector.
The Chinese government needs to implement effective policies and regulations to ensure that the growth of the AI industry aligns with its mitigation goals. AI development poses challenges as well as provides opportunities to the country’s low carbon transition.
