China Southern Power Grid packs computing power into containers, and the "power-hungry behemoth" that consumes 196 billion kWh of electricity has finally found its "taming switch" from the national team.
The power grid enters the electricity-carbon calculation synergy, reconstructing the competition logic of the computing power industry.
Yujian Energy learned that recently, China Southern Power Grid completed the test of a green and low-carbon mobile computing power shelter in Guiyang, Guizhou. This 40-foot container is equipped with high-density computing power, intelligent temperature control, and a dispatching system that can communicate with the power grid in real time.
What does this mean?
In 2025, the electricity consumption of domestic computing power centers reached 196 billion kWh, a year-on-year increase of 18.1%, and the proportion of the total electricity consumption of the whole society exceeded 1.9% for the first time. At this rate, by 2030, this figure will climb to 500 to 700 billion kWh, which means that one out of every 20 kWh of electricity is consumed by data centers.
While the entire industry is struggling with "how to save electricity", the mobile computing power shelter tested by China Southern Power Grid in Guiyang gives a completely different answer - move the computing power into a container and plug it directly beside the substation. The dispatching instructions of the power grid can directly tell the computing power when to calculate and when to rest in real time. This is not just a container of servers; it is a list of control rights transfer issued by the power grid to the computing power industry.
No one can withstand the "power-hungry monster"
How much power does computing power consume? A few sets of data are enough to illustrate.
Statistics from the China Academy of Information and Communications Technology show that in 2025, the electricity consumption of domestic computing power centers was 196 billion kWh, a year-on-year increase of 18.1%, while the growth rate of the total electricity consumption of the whole society during the same period was only 5.2%. The growth rate of computing power electricity consumption is nearly three times the average level of the whole society. Data from the International Energy Agency also confirms this trend: in 2025, the electricity consumption of global data centers was about 485 billion kWh, a year-on-year increase of 17%. By 2030, the electricity consumption of Chinese computing power centers is expected to account for 3.7% to 5.3% of the total electricity consumption of the whole society. In some more radical predictions, it may even exceed 7% to 9%. According to the current trend, the power load of a gigawatt-level artificial intelligence data center is equivalent to that of a medium-sized city with a population of one million.
This is not just a simple numerical problem; there are three contradictions breaking out simultaneously behind it.
Spatially, the demand for computing power in the east is strong, but the green power resources are scarce. In the west, there is an abundance of green power, but the demand for computing power is insufficient. The "East Data West Computing" initiative has been advocated for several years, but the circulation of computing power between the east and the west is still difficult. Temporally, wind and solar power generation are greatly affected by the weather, while the training of large AI models requires stable power supply 24/7. There is more green power than needed when it is available, and not enough when it is gone. In terms of rhythm, computing power technology doubles every 18 months, but the planning of power infrastructure is on a five-year cycle. The situations of "computing power waiting for electricity" and "electricity waiting for computing power" alternate.
What's more troublesome is that as of the end of 2025, the total installed capacity of wind and photovoltaic power in the country accounted for 47.3% of the total installed capacity, but the power generation only accounted for 22%. The rate of abandoned wind and solar power dropped below 95% for the first time during the 14th Five-Year Plan period, and the utilization rate of photovoltaic power in Qinghai, Tibet, Xinjiang, and Gansu dropped below 90%.
The computing power industry itself is also under great pressure. Electricity bills are no longer just a string of numbers but a real mountain. Electricity bills already account for a large proportion of the operating costs of data centers. For those most power-hungry AI intelligent computing centers, the electricity bills can easily reach hundreds of millions. In the domestic market, cloud service providers such as Alibaba Cloud, Tencent Cloud, and Huawei Cloud are desperately lowering the unit price of computing power in the price war, but the rigid cost of electricity cannot be reduced, and the profit margin is continuously squeezed. Since last year, some cloud service providers have started to raise the usage price of large models, with the highest increase even exceeding 400%.
That is to say, the key to the cost of computing power lies in electricity and carbon emissions. Whoever solves these two problems can stand at the forefront in the second half of the computing power industry.
The mobile computing power shelter is not a product but a "dispatching card" of the power grid company
All the problems mentioned above are what the computing power industry is complaining about, but the one who can really prescribe the medicine is not the computing power company itself but the power grid company. Because this is essentially an energy supply problem, not a technical problem.
The mobile computing power shelter tested by China Southern Power Grid in Guiyang is essentially the power grid company playing its three cards - the power system, the carbon market, and computing power dispatching - at the same time. The shelter uses a 40-foot standard container as the carrier and adopts a "high-efficiency liquid cooling + natural cooling" composite refrigeration solution, and its energy-saving effect is significantly better than that of traditional computer rooms. The shelter can be directly connected to substations or zero-carbon parks, revitalizing the existing land and power resources, and shortening the construction cycle of computing power projects from years to months.
But these are just the surface. The real key lies in two other things.
First, the shelter is connected to the electricity-carbon calculation synergy operation platform of China Southern Power Grid. This platform is based on reliable metering of the power grid, load dispatching, the power market, green power certificates, carbon accounting, and computing power dispatching. In this platform, the shelter is not a "user" but a flexible adjustment unit that can be dispatched and controlled by the power grid in real time.
The power grid can use price signals to tell the shelter when to calculate and when not to. The cost of wind and photovoltaic power generation in Guizhou is extremely low. When the power generation is large, the supply in the power market exceeds the demand, and the advantage of off-peak electricity price is prominent. The platform predicts the electricity price through AI. When the electricity price is high, it reduces the operation of computing power, and when the electricity price is low, it increases the operation of computing power, and gives priority to the periods with more wind and photovoltaic power. Therefore, the computing power cost of the shelter is much lower than the market price.
Second, the shelter turns carbon emissions from a "burden" into an "asset". The energy consumption and carbon emission data of each step of the shelter are quantifiable and traceable. The electricity-carbon calculation synergy system can track the carbon emission information of each kWh of electricity behind each computing power task, forming a "carbon footprint portrait". Therefore, low-carbon computing power can be certified and traded.
This is not a loophole. The Central Economic Work Conference in 2024 clearly stated that "a number of zero-carbon parks will be established", and it entered the substantial implementation stage after the Two Sessions in 2026. The shelter is deployed in zero-carbon parks and uses green power for calculation. The saved carbon emission allowances can be directly cashed in the carbon market. And the carbon market is expanding - in 2025, the cumulative trading volume of the national carbon market reached 865 million tons, with a turnover of 57.633 billion yuan. The controlled greenhouse gas emissions exceeded 8 billion tons at one go, accounting for about 70% of the country's total emissions.
Looking at these two things together, the conclusion is clear: China Southern Power Grid is not selling the servers in the container; it is selling the "computing power optimized by both the power system and the carbon market". Equipment manufacturers make the things inside the container, and the power grid company sells the service chain after the container is plugged in. These are two completely different business models.
In the second half of the computing power battle, what cards do the power grid, operators, and Huawei hold?
China Southern Power Grid is not the only one calculating this account.
At the Mobile Cloud Conference in May 2026, Chen Zhongyue, the chairman of China Mobile, said: Computing power services are clearly defined as the company's main business. It is necessary to explore a four-in-one cooperation model of "computing power planning, power support, energy storage peak shaving, and power grid backup" and build a coordinated dispatching system for the national computing power network and the new power grid. China Telecom's "Xirang" platform has dispatched 87 EFLOPS of computing power. Although Huawei does not generate electricity or dispatch power, it defined the synergy of computing and electricity as a new model for data center construction at the beginning of 2025.
These signals point to the same trend: The computing power industry is transitioning from the first half where "whoever has the chips has the say" to the second half where "whoever can optimize the energy cost structure has the say".
But looking closely, each party has different cards. China Mobile and China Telecom are doing "computing power companies actively adapting to the power system" - optimizing energy consumption behavior and responding to electricity price signals, but the dispatching power is still in the hands of the power grid. China Southern Power Grid's mobile computing power shelter turns the computing power directly into a "programmable load" in the power grid dispatching system. This is not adaptation but incorporation. Huawei's role is simpler: selling equipment and solutions without participating in operations.
Although all the state-owned enterprises have entered the game, the power grid company has electricity and carbon, the operators have networks and computer rooms, and Huawei has technology. It is still hard to say who can get the most profit on this new "electricity-carbon calculation" chessboard. But one thing is certain - The game rules of the computing power industry are being rewritten by these state-owned enterprises.
The state is redefining the boundaries of computing power infrastructure
Looking at the national policy map, this shift is not accidental.
2026 is a watershed for the policy of computing and electricity synergy. The Government Work Report for the first time clearly proposed "implementing new infrastructure projects such as ultra-large-scale intelligent computing clusters and computing and electricity synergy", and "computing and electricity synergy" has officially risen to the national top-level strategy. Immediately afterwards, four departments including the National Development and Reform Commission, the National Energy Administration, the Ministry of Industry and Information Technology, and the National Data Administration jointly issued the Action Plan for Promoting the Two-Way Empowerment of Artificial Intelligence and Energy, which is a top-level design document for the in-depth integration of the energy revolution and the digital revolution.
Before that, the Politburo meeting of the CPC Central Committee in April 2026 listed the computing power network together with the water network, the new power grid, and the new generation communication network as the "six networks" and included them in the top-level design of the national strategic infrastructure. The concept of the "six networks" clearly defined the position of the computing power network at the Politburo level - It is no longer a subdivision of the information industry but a national-level infrastructure on a par with the power system and the water resource network.
The policy has laid the foundation for the industry, and computing power has become the second basic energy form after electricity. A country cannot hand over the pricing and dispatching rights of basic energy to any single enterprise. As the demand for AI computing power continues to soar, the cost structure of computing power increasingly depends on the cost of electricity and carbon. Whoever controls the pricing rights of electricity and carbon has the final say in the computing power industry.
Putting computing power into a container essentially breaks the dependence of computing power on a fixed physical space and makes "plug-and-play computing power" possible. But what really changes the game rules is that the power grid company enters the game with its power system and carbon market. The shelter is just a terminal node, and the real core competitiveness lies in the system behind the shelter - power trading, carbon accounting, and computing power dispatching. These three are combined, and the cost structure of computing power is completely reconstructed.
This means that the competition logic of the computing power industry is being rewritten. In the past, Alibaba Cloud, Tencent Cloud, and Huawei Cloud competed in terms of chip computing power, model effects, and ecological stickiness. In the future, the cost line of computing power will increasingly approach a basic curve determined by the power and carbon markets. Those players closest to the power system and the carbon market - power grid companies, operators, and large power enterprises - are gaining a new cost advantage. For cloud service providers, this may not be a threat, but it is a structural variable that must be faced.
Globally, China's advantage in this field lies in having the world's largest power system, the world's most complete carbon market framework, and the world's largest demand for computing power. Yujian Energy believes that if the integration of electricity, carbon, and computing is successful, the moat for computing power costs will not be TSMC's advanced manufacturing process but the UHV transmission lines, the design ability of the carbon market, and the system ability of power grid dispatching.
This article is from the WeChat official account "Yujian Energy", author: Zhao Jianan. Republished by 36Kr with permission.