In early June, I traveled to Giyang in Githuhou Province, southwestern China, and visited the Gyang Data Center in China Mobile, designated as the “National Green Data Center.” The facility is located in a pack of modern buildings in a new area of Gian. Since 2021, the cluster has become a hub for China's internet infrastructure and hosts data centers run by leading high-tech companies.
Engineer Jian Chonghai walked the center's energy-saving innovations: modular cooling cabinets, or “Maglev” air conditioning systems. Unlike traditional AC, Maglev air conditioners eliminated the friction of the compressor and achieved the same cooling effect at 30-40% power, he said.
Jian's focus on energy efficiency is driven by an ambitious expansion plan. As product manager Li Haiyan told me, AI models training and running are set up to be part of the business.
China Mobile is not alone. From big companies like Alibaba, Tencent and Huawei to startups like Deepseek, Chinese companies are trapped in fierce competition to develop AI services.
This represents a new challenge for China, especially as experts believe that AI will become one of the most energy-intensive industries in the country. China's central challenge is how to become a global leader in AI services without putting climate action goals at risk.
“
Electricity and computing are considered public services. This means it should be accessible and low cost. However, to reach the targets of the country's climate, it must be clean and green. The big question is how can we make it both affordable and sustainable?
The possibilities of AI to improve efficiency and help other sectors decarbonize faster offers some hope, but it has become important to prioritize renewable energy and build new power systems that meet computing needs.
Increases appetite
According to the International Energy Agency (IEA), electricity usage in data centers in China is expected to increase by 170% between 2024 and 2030.
Last year, data centers around the world consumed about 1.5% of the total electricity generated, and recent IEA reports estimates have been estimated. This share is growing rapidly. Power usage in data centers is increasing to around 12% per year, mainly driven by AI-specific servers. This is four times the pace of overall electricity demand.
The report predicts that by 2030, China and the US will account for almost 80% of global data center energy growth.
Wang Yongzhen, an associate professor at Beijing University of Technology, along with steel, cement and petrochemists, told me that data centers are one of China's leading energy-intensive industries.
He estimates that by 2030, Chinese data centers will demand around 105 gigawatts of electricity, use 26.3 billion liters of water, and release 310 million tonnes of CO2.
This amount of electricity is more than half of China's housing demand in 2024.
East Data, West Computing
In 2022, China launched an initiative called East Data and West Computing. Under the plan, western states and regions such as Guizhou, Inner Mongolia, Gansu and Ningxia will be tasked with handling computing jobs such as AI training and data storage. A workload that does not require real-time response.
Meanwhile, clusters in regions such as Beijing Tianjin Hebei, Youngtu River Delta, Chengdu Yong-Hyeong and Greater Bay Area focus on real-time services such as video streaming and the use of AI chatbots.
One of the key goals of this initiative is to reduce energy consumption by leveraging the favorable climate and abundant renewable energy in western China. For example, while Guiyang's average annual temperatures between 18-20°C naturally reduce cooling needs, Inner Mongolia offers abundant wind and solar power.
The plan is expected by the end of this year, newly built data centers at hub nodes across the country will run with over 80% of renewable power. Areas with strong solar, wind, or hydropower resources will build low-carbon data centers that handle computing intensive but time-sensitive tasks.
Wang Yongzhen said that green computing power is not only consistent with national strategy, but also provides concrete economic benefits for businesses. Making the data center more energy efficient means smaller electricity bills, he said.
When AI meets the grid
One important part of the puzzle is the National Integrated Computing Network. It is developed by the government to bring public and private cloud computing resources together into a single platform. It was highlighted in a recent report by research institute RAND entitled “China's Evolving Industrial Policy AI.”
Kyle Chan of Princeton University, the lead author of the report, resembles the networks the “utilities” model of AI computing resources. This is an approach that reflects China's typical infrastructure development strategy aimed at reducing regional inequality.
The network is in line with another major initiative to build “clean, efficient, flexible and intelligent” power systems. From 2024 to 2027, China plans to build this “new power system” that relies on a variety of renewable sources, such as AI, and smart technologies.
Under these two initiatives, electricity and computing are considered public services. This means it should be accessible and low cost. However, to reach the targets of the country's climate, it must be clean and green. The big question is how can we make it both affordable and sustainable?
Wang Yongzhen envisions that computing power and electricity create a synergistic effect. “One of the key goals is to increase the share of green electricity used in data centers. The second is to reduce energy consumption. Not only is it isolated high-tech upgrades, but also improve overall system efficiency.
Simply put, if the area is facing tight power sources, computing tasks can use power spot market prices as guidance to migrate to data centers that meet speed requirements but have low cost. This approach increases the reliability of AI for customers while mitigating local power crunches.
“In the face of intermittent wind and solar power, data centers can also act as buffers by 'turning on' certain computing tasks, such as AI training during oversupply periods. ”
Wang also said the data centers could be tailored to ensure that grids can take advantage of energy storage systems in emergencies. This has many in common with the grid models from vehicles that ensure that electric vehicles return power to the grid at peak demand.
Still, the challenges remain.
First of all, the synergy between computing and power networks described by Wang requires extensive coordination at various parties, government agencies, and even at the staff level. It also needs to be tested in a variety of scenarios for a long time.
Wang Yongzhen said he noticed a communication gap between IT engineers like Li Haiyan and operational staff like Jian Chonghai between staff members in the data center. This gap could become a major bottleneck in the future. “If we want this highly tuned network vision to be a reality, they need to keep both their computing and energy systems constantly.”
Another challenge is market design.
China Academy of Information and Communications Technology points out that green power certificate systems track and verify renewable generation are lagging behind market demand. In 2023, of the 3.8 million mWh of the green power traded, only 1.5 million hours (MWH) was involved in the certificate.
This could hinder data centers that want to buy more green power or track the share of green power with their own energy use.
ai marches
The 2017 state legislative document aims to ensure that China will become the world leader in AI theory, technology and application by 2030. This indicates that despite energy, climate targets and pressures from other countries such as the US, China's AI development will not slow.
For China, the goal is not only to “win the race” with the US, but also to build a “resilient AI industry that will increase productivity across the sector, from manufacturing and healthcare to education and governance.”
Wang Yongzhen said advances in AI are increasing power demand from data centers, but the increase in computing power has also helped support decarbonisation efforts in other sectors. Therefore, considering the energy consumption of AI in the broader context of social decarburization, it is very simple to equate the rising energy use with set-folding to climate targets.
One of the biggest hurdles in the efforts to build a carbon footprint management system across the country is data sensitivity, Xu Ming, a professor at Tsinghua University, told me on a podcast. Full carbon accounting requires tracking everything from suppliers to delivery. Many companies fear revealing trade secrets.
Acting as Neutral Data Steward, AI-powered agents can safely store emissions data and grant access to regulatory authorities while protecting confidentiality. In this way, AI becomes an important tool in reducing industrial emissions.
However, these ideas are still in the experimental stage and it takes time to see specific results on how AI can actually limit or reduce emissions.
Back at Gui'an Data Center, Li Haiyan compared the mainstream AI model to an undergraduate student. In the future, engineers will train more specialized, domain-centric AI models similar to graduate students, doctoral degrees and postdocs, which can provide more targeted and practical solutions. She believes that this shift will reduce the demand for chips used in both training and operational AI models, which could reduce demand for running and cooling.
This article was originally published on Dialogue Earth under a Creative Commons license.
