This is the third year that China’s annual Government Work Report has mentioned artificial intelligence, with the language used becoming more concrete and specialised.
In the energy sector, there is particular focus on AI for electricity grids. The National Development and Reform Commission has said that by 2027 at least five specialised large language models will be deeply embedded in the power grid, power generation and other fields.
However, AI could be a double-edged sword for China’s transition to clean energy.
On one hand, the country has high hopes for the technology: in Shanghai, Xinjiang and Beijing AI models and algorithms are being used to predict renewable energy output and to optimise and secure grids. Data centres, meanwhile, could in theory help use up green electricity that would otherwise be wasted and add flexibility to the grid.
But the data centres on which AI relies are energy hungry. They are predicted to use 3-5 per cent of all China’s electricity generation by 2030 – against residential consumption of 15 per cent. Generative AI and other advanced services can cause short-term spikes in electricity consumption and there are already examples of that affecting grid security.
So, what can AI and the data centres that power it do to make China’s grids more efficient, keep those grids stable, and make use of green electricity? Dialogue Earth spoke to the experts.
“
The integration of AI into virtual power plants is at a very early stage. In a grid like China’s, which prioritises reliability, you can’t make full use of AI for dispatch. Mostly it’s used to assist in decision-making.
Gao Hongchao, chief scientist assistant, National Key Project on Virtual Power Plants
The rise of AI in grid applications
The International Energy Agency predicts that AI-powered algorithms could optimise grid operations, improve the integration of wind and solar output and cut unexpected outages by up to 50 per cent. “If scaled up, existing AI-led interventions could lead to global electricity savings of around 300 terawatt-hours”. That’s more than twice the amount of power Beijing uses in a year.
Many of China’s electricity suppliers and research organisations are working on AI tools for their own specific needs. The China Southern Power Grid, for example, says it has developed a precise power forecasting system which can support the trading of electricity and improve both the security and cost efficiency of the grid.
AI is beginning to be involved with virtual power plants (VPPs) which are aggregations of distributed energy resources managed collectively to behave like a single power plant.
Early this year, Shanghai Securities News reported on how AI had helped a local VPP cope with a cold snap. On receiving the weather forecast, the VPP told a manufacturer to reduce consumption. The article describes AI as the “brain” of VPPs, able to reduce inaccuracy in real-time consumption estimates to under 3 per cent, and make 85 per cent accurate prediction of mid- and long-term spot market prices.
Shanghai, Jiangsu and Guangdong are using data centre energy storage facilities in provincial VPPs, according to state media outlets, with AI predicting electricity demand, dispatching power and steering consumers to switch demand from peak times to quieter periods, so reducing pressure on the grid. This is expected to shave “3.5 gigawatts” off peak demand in 2026, according to business outlet Qianjia.
According to a list compiled by Deben Consulting, China’s grids will use AI to check for faults, predict risks and support “demand response”. This means encouraging consumers to shift their electricity use to times when it is more plentiful or general demand is lower, something which can also support greater use of renewable power.
Although, as the International Energy Agency states, AI is already being deployed to “transform and optimise energy and mineral supply, electricity generation and transmission, and energy consumption,” some experts are dubious that it can make a useful contribution to China’s energy grids as they exist today.
How far can AI improve China’s grids?
Gao Hongchao, chief scientist assistant for the National Key Project on Virtual Power Plants, said that “in engineering terms, the integration of AI into virtual power plants is at a very early stage. You could even say that current projects haven’t seen very good or real applications and outcomes.”
Gao told Dialogue Earth a key issue is that the current legislation isn’t able to assign responsibility for any errors made by the tools. “In a grid like China’s, which prioritises reliability, you can’t make full use of AI for dispatch. Mostly it’s used to assist in decision-making.”
Zhang Shuwei, chief economist at the Draworld Centre think-tank, told Dialogue Earth that AI can only have a limited impact, even in a supporting role.
He said that China’s grids are focused on maintaining stable supplies, and so AI is mainly used to that end, rather than for improving efficiency, as in the west.
For example, in 2019 the National Energy Administration required that dispatch operations should consider seasonal conditions and set “stable operation quotas” and “calculate and analyse stability of trunk lines and localities.”
Zhang explained that Chinese generators all stand ready to respond to dispatch orders, bringing capacity on or offline as required. “In a system like that, you don’t get big price fluctuations across time,” he said.
“If all dispatch-ready generators are running steadily, there’s no scope for price changes. But price changes are essential for advanced grid-control technologies like AI to play their role, as they profit from price differences across generators and consumers.” With no financial incentive to provide more accurate predictions, it will be hard to apply AI for greater efficiency.
However, AI tools are being used to shave peak demand. In Shenzhen, the technology was used over 150 times between 2023 and 2025 to reduce load at peak times, according to Science and Technology Daily.
But Zhang Shuwei thinks it’s hard to use data to decide when peak shaving is necessary and fair – and AI won’t help.
“When demand is 1,000 kilowatt-hours and you have supply of 800, you need to drop 200. But deciding where is a value judgement. That’s not going to change just because you use AI. AI can’t bring about fairer or more efficient allocations – all it does is amplify the (existing) system’s imbalances.”
Anders Hove, a senior research fellow at the Oxford Institute for Energy Studies, agrees that flexibility is a problem, saying the use of AI for peak shaving in Shanghai is not a typical example. The biggest bottleneck for China’s grids, he says, and for the use of AI in grid operations, remains a lack of flexibility.
He said that overall, China’s peaks in demands are “much more modest” than those seen in the west. “China’s power system has a higher proportion of industrial demand than in North America or Europe, hence it has somewhat more stable load.
However, peak load is growing rapidly, and this is a serious concern for grid officials, along with the problem of the ‘duck curve’ due to high solar output at midday, high peak demand in late afternoon/early evening, leading to fast ramping requirements. Flexibility is needed to handle such peaks.”
However, in China “most electricity trading is on a monthly or annual basis. In particular, intra-province trading and inter-province transfers are almost always done on long-term contracts concluded far in advance.” Although the government often “talks about how important inter-provincial trades are” and offers examples of flexibility, those cases are the exceptions.
In fact, China’s transmission lines “do not, like those in North America and Europe, send electricity back and forth according to demand,” he said. “Government policy makes clear that the goal is for 90 per cent of electricity transfers to be done under long-term contracts – which runs contrary to the idea of flexibility.”
This means that even if data centres and AI can predict gaps in supply and demand, they cannot flexibly and promptly deliver power to where it is needed, Hove said.
‘Eastern Data, Western Compute’ and ‘Compute/Electricity Coordination’
In 2022, China launched a project to meet computational demands arising in the east of the country in new data centres to the west, where there are ample supplies of wind, solar and other clean sources of energy.
Anders Hove said that while there is research showing data centres can be a source of flexibility in the system 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 place is only practical for a very small part of their operations,” he said. “More important are economic factors. Customers don’t tell them which tasks can wait till later, nor do they care where and when the tasks are done. They just pay for speed and reliability.” The need to provide customers with immediate responses means that while data centres in the east may send some tasks to facilities in the west of China, they will not relocate there.
However, data centres can also encourage the development of clean energy. Last month the government introduced the idea of coordinating the power and data centre sectors, with the centres to shift from being major power consumers to aiding in grid management.
The policy will see data centres integrated into power allocation, so they use as much green electricity as possible where it is available. For example, bright and sunny Ningxia’s “compute and power integration” project has hooked its 500-megawatt solar power farm up to the grid.
According to Xinhua, the National Data Administration has said new computing facilities built at eight national computing hubs must draw 80 per cent of their power from green electricity.
Zhang Shuwei said this is a positive signal for “additionality”: if data centres can, through power purchase agreements and contractual arrangements, promote new investment in wind and solar power and help build their own power supplies, “that’s good news for the climate”.
This article was originally published on Dialogue Earth under a Creative Commons licence.
