China’s energy system is becoming a major test bed for artificial intelligence at the same time that it is becoming the world’s largest solar-power base by installed capacity. Xinhua reported on September 4 that the National Energy Administration had announced photovoltaic capacity of 1,286 gigawatts at the end of July 2026, just above the 1,285 gigawatts of coal-fired capacity.
The milestone concerns installed capacity, not actual electricity generation. Solar output changes with sunlight, weather, and time of day, while coal plants can run on demand. But the crossover creates a powerful reason to use software that can forecast supply, schedule storage, manage variable loads, and coordinate electricity consumption with renewable generation.
Xinhua’s report identifies AI as one part of that response. It describes applications in research and development, operations and maintenance, safety monitoring, energy storage, and electricity-market trading. The examples come from specific Chinese companies and projects. They show where AI is being tested, but their claimed performance results should be treated as company or project statements rather than independent audits.
Solar Capacity Changes the Operating Challenge
The scale of China’s photovoltaic build-out creates a different operating problem from the one faced by a coal-heavy system. Solar power can add large volumes of energy at certain hours while falling quickly at others. Energy-storage operators must decide when to charge and discharge batteries. Grid managers need to anticipate changes in supply and demand. Factories and data centers may also have an incentive to shift flexible activity to cheaper periods.
Xinhua reports that Weheng Intelligent Technology developed WHES OS, an AI-native energy operating system. In the example described by the news agency, a plant manager gives the system a natural-language instruction to prioritize production and charge batteries when electricity is cheaper. The system generates a scheduling strategy without the manager writing code or setting detailed rules.
That example captures why AI agents are attracting attention in the sector. Traditional energy-management software follows predefined rules. An agent can interpret a higher-level goal and combine operational data with market conditions. The value of such a system depends on whether it makes reliable decisions when prices, weather, equipment status, and customer demand change quickly.
The story is not only about power plants. China’s data centers are becoming a major electricity consumer because AI training and inference require large computing loads. Xinhua said China’s data centers used 170 billion kilowatt-hours in 2025, representing 1.6% of national electricity consumption. It cited a China Galaxy Securities projection that data-center use could exceed 700 billion kilowatt-hours by 2030, or more than 5% of the total.
That 2030 figure is a securities-firm projection, not an official forecast. It nevertheless illustrates why policymakers are pairing computing policy with energy planning. An AI system needs reliable power, while a power system with growing renewable capacity needs flexible loads and better forecasting. The two challenges are increasingly linked.
EastFrontier explored this tension in June in an article on China’s AI data centers and the green-energy paradox. The September capacity crossover adds a concrete number to that earlier debate. China may have an exceptional supply of renewable capacity, but turning that capacity into dependable clean power for computing requires transmission, storage, pricing systems, and operational software.
Storage Companies Are Applying AI to Warnings and Trading
Xinhua provides several examples of AI inside battery-storage operations. Contemporary Amperex Technology Co., or CATL, says its TENER Smart Storage platform can provide fault warnings up to seven days in advance and identify fault causes within five minutes. Those are company capability claims, but they point to a valuable use case: detecting risks before a thermal problem turns into a safety incident or an unplanned outage.
The difference between a conventional alarm and an AI-based system is not simply speed. Traditional systems may trigger when a sensor crosses a fixed threshold. A model can look for combinations of measurements that indicate a problem before any one value becomes obviously abnormal. The quality of the result depends on sensor coverage, data quality, the range of failures included in training data, and the operating environment.
Xinhua also reports that an AI trading agent at an Envision Group smart storage station in Binzhou, Shandong, achieved 95% accuracy in predicting peak-valley electricity-price spreads. The source does not provide the sample period, benchmark, or method behind that figure. It should therefore be read as an example of an operating project, not proof that AI can consistently predict power prices across China.
Still, the use case is clear. Storage operators earn more when they charge electricity at low prices and discharge when prices are higher, subject to battery limits and grid rules. Forecasting renewable output, weather, congestion, and market quotes can make those decisions more complex. An AI system may help, but it also needs risk controls because an incorrect forecast can reduce revenue or create operational problems.
EastFrontier reported in August that Guizhou was using AI to balance hydropower, wind, and solar. The new Xinhua report broadens the picture from one provincial use case to storage safety, market trading, and factory scheduling. Together, the examples show how AI is moving from planning presentations into daily energy operations.
Policy Is Linking Clean Power and Computing Infrastructure
Xinhua said the 2026 government work report included new-infrastructure projects around hyperscale intelligent-computing clusters and a requirement to coordinate computing capacity with electricity supply. The news agency said that provision appeared in the report for the first time. Its account of China’s next five-year planning outline likewise calls for green power and computing facilities to be developed together.
The policy direction is practical. New computing facilities need to be located where the grid can support them, and renewable-energy bases need demand that can absorb variable output. In May, four Chinese authorities released an action plan that envisioned a substantial increase in clean-energy supply for AI-computing infrastructure by 2030, Xinhua reported. The plan also calls for more AI application in the energy sector.
The report points to an operating project in Ulanqab, Inner Mongolia, where a data center combining renewable generation, storage, and computing load began operating in July 2025. It uses self-generated wind and solar power, stores surplus in batteries, and relies on the grid as a backup. Such projects are useful laboratories, but their performance will depend on local resource patterns, equipment reliability, and the economics of storage.
China’s long-term advantage is not guaranteed merely because solar capacity exceeds coal capacity by one gigawatt. Grid integration is difficult, and Xinhua identifies remaining obstacles, including a shortage of industry-specific AI models, limited high-quality industry data, and weak mechanisms for cross-enterprise and cross-industry sharing.
The September milestone nevertheless shows why AI and energy are converging in China. Renewable capacity creates new opportunities and new variability. Data centers create growing demand and potentially flexible loads. Energy companies are testing agents that can read conditions, warn of faults, schedule batteries, and trade electricity. The next measure of success will be whether those systems improve safety, lower operating costs, and help the grid use more clean power without creating new reliability risks.