This is the third year that China’s annual government work report has mentioned artificial intelligence, and the language used has become more specific and specialized.
In the energy sector, AI for power grids is of particular interest. The National Development and Reform Commission said that by 2027, at least five specialized large-scale language models will be deeply integrated into power grids, power generation and other fields.
But AI could be a double-edged sword for China’s clean energy transition.
Meanwhile, China has high hopes for the technology, with AI models and algorithms being used in Shanghai, Xinjiang and Beijing to predict renewable energy output and optimize and secure power grids. Data centers, on the other hand, could theoretically help consume green power that would otherwise be wasted, giving flexibility to the power grid.
However, the data centers that AI relies on are energy-intensive. It is projected to use 3 to 5 percent of China’s total electricity generation by 2030, compared to 15 percent for household consumption. There are already examples where generative AI and other advanced services can cause short-term spikes in electricity consumption, impacting grid security.
So what can AI and the data centers that power it do to make China’s power grid more efficient, maintain grid stability, and harness green power? Dialogue Earth spoke to experts.
“
The integration of AI into virtual power plants is in the very early stages. In a power grid like China’s, which emphasizes reliability, it is not possible to fully utilize AI for dispatching vehicles. Primarily used to aid decision making.
Mr. Gao Honchao, National Key Project on Virtual Power Plant, Assistant Principal Scientist
The rise of AI in grid applications
The International Energy Agency predicts that AI-powered algorithms could optimize grid operations, improve the integration of wind and solar power, and reduce unplanned power outages by up to 50%. “Existing AI-driven interventions, if scaled up, could lead to approximately 300 terawatt-hours of electricity savings globally,” which is more than double the amount of electricity used by the Chinese government annually.
Many of China’s electricity providers and research institutes are working on AI tools to suit their own needs. For example, China Southern Power Grid says it has developed an accurate power forecasting system that can support power trading and improve both the safety and cost efficiency of the power grid.
AI is starting to play a role in virtual power plants (VPPs), which are collections of distributed energy resources that are collectively managed to behave like a single power plant.
Earlier this year, Shanghai Securities News reported how AI helped the local VPP cope with the cold wave. In response to the weather forecast, the VPP has instructed manufacturers to reduce consumption. The article explains that AI is the “brain” of VPP and can reduce the inaccuracy of real-time consumption forecasts to less than 3% and predict medium- to long-term spot market prices with 85% accuracy.
According to state media, Shanghai, Jiangsu and Guangdong provinces are using energy storage facilities in data centers in local VPPs, where AI predicts electricity demand, delivers power and guides consumers to switch demand from peak periods to off-peak periods, reducing strain on the grid. This is expected to reduce peak demand by “3.5 gigawatts” in 2026, according to business outlet Qianjia.
China’s power grid will use AI to check for outages, predict risks and support “demand response”, according to a list compiled by Deben Consulting. This means encouraging consumers to shift their electricity usage to times of day when it is more plentiful or when demand is generally lower, which also supports greater use of renewable electricity.
As the International Energy Agency puts it, AI is already being deployed to “transform and optimize energy and mineral supplies, power generation and transmission, and energy consumption,” but some experts doubt whether AI can make a useful contribution to China’s energy grid as it exists today.
How far can AI improve China’s power grid?
Gao Hongqiao, Assistant Chief Scientist of the National Key Project on Virtual Power Plants, said: “From an engineering perspective, the integration of AI into virtual power plants is at a very early stage. You can even say that current projects have not seen very good practical applications and results.”
Gao told Dialogue Earth that a key issue is that current laws do not allow for accountability for errors caused by tools. “Grids like China, which prioritize reliability, cannot take full advantage of AI for ride dispatch. It is mostly used to support decision-making.”
Zhang Shuwei, chief economist at think tank Dora World Center, told Dialogue Earth that AI will only have a limited impact, even in a supporting role.
He said China’s power grid is focused on maintaining a stable supply, so AI is primarily used for that purpose, rather than to improve efficiency as in Western countries.
For example, in 2019, the National Energy Administration mandated that dispatch operations consider seasonal conditions, set “stable operational quotas” and “calculate and analyze mainline and regional stability.”
Zhang explained that all of China’s generators are ready to respond to dispatch orders and can bring capacity on or offline as needed. “Such a system does not have large price fluctuations over time,” he said.
“If all dispatch-ready generators are operating stably, there is no room for price changes. But for advanced grid control technologies such as AI to do their job, price changes are essential, as they profit from the price differential between generators and consumers.” Applying AI to increase efficiency is difficult because there is no financial incentive to provide more accurate forecasts.
However, AI tools are being used to reduce peak demand. According to Science and Technology Daily, the technology was used in Shenzhen more than 150 times between 2023 and 2025 to reduce load during peak hours.
However, Zhang Shuwei believes that it is difficult to use data to determine when peak cuts are necessary and fair, and that AI is useless.
“If the demand is 1,000 kilowatt-hours and the supply is 800, you need to drop 200 kilowatt-hours. But where you decide is a value judgment. Using AI doesn’t change that. AI can’t bring about fairer or more efficient allocation. All it does is amplify the imbalances in the (existing) system.”
Anders Hove, a senior research fellow at the Oxford Energy Institute, agreed that flexibility was an issue and said the use of AI to cut peaks in Shanghai was not a typical example. He said the lack of flexibility remains the biggest bottleneck for China’s power grid and the use of AI in grid operations.
Overall, China’s demand peaks are “much more gradual” than in Western countries, he said. “China’s power system has a higher share of industrial demand than North America or Europe, so the load is a little more stable.
However, peak loads are increasing rapidly, which, along with the problem of “duck curves” due to high midday solar output and high late afternoon/evening peak demand, has become a serious concern for grid managers, leading to rapidly increasing requirements. Flexibility is required to deal with such peaks. ”
However, in China, “most electricity trade is done on a monthly or annual basis. Intra-provincial trade and inter-provincial remittances in particular are mostly done on the basis of long-term contracts signed long ago.” Although the government often “talks about how important inter-provincial trade is” and provides examples of flexibility, those cases are the exception.
In fact, China’s power lines “are not sending power back on demand, unlike North American or European power lines,” he said. “Government policy is clear that the aim is for 90 per cent of electricity transmission to be under long-term contracts, which runs counter to the idea of flexibility.”
This means that even if data centers and AI can predict the gap between supply and demand, they won’t be able to flexibly and quickly deliver power where it’s needed, Hove said.
“Data in the East, Computing in the West” and “Linking Computing and Power”
China has launched a project in 2022 to meet computing demand in the country’s east, where wind, solar and other clean energy sources are plentiful, with new data centers in the west.
Anders Hove said that while there is research showing that data centers can be a source of system flexibility rather than a constant source of load, the reality is that most such facilities in China are still located in the east.
“Flexibility across time and space is practical for only a small portion of operations,” he said. “More important are the economics. Customers don’t say which tasks can wait until later, and they don’t care when or where the tasks are completed. They just pay for speed and reliability.” The need to provide instant responses to customers means data centers in the east may send some tasks to facilities in western China, but not relocate them there.
However, data centers can also facilitate the development of clean energy. Last month, the government introduced the idea of aligning the power and data center sectors, moving the centers from being primary consumers of electricity to supporting grid management.
Under this policy, data centers will be integrated into power allocation and will use green power as much as possible when available. For example, the “Computing and Power Integration” project in the bright and sunny Ningxia Hui Autonomous Region has connected a 500-megawatt solar power plant to the power grid.
The National Data Agency said new computing facilities to be built in eight national computing hubs will need to source 80 percent of their electricity from green power, according to Xinhua.
Zhang Shuwei said this is a positive signal of “additionality,” and if data centers can encourage new investment in wind and solar power and help build their own power sources through power purchase agreements and contractual arrangements, “that’s good news for the climate.”
This article was originally published on Dialogue Earth under a Creative Commons license.
