In June 2026, the "7th China Data Center Green Energy Conference" was successfully held in Shanghai. Tang Rulin, Dean of the AI Research Institute at Huaxin Consulting and Design Research Institute Co., Ltd., delivered a speech titled "Latest Trends and Practices in Computing-Power and Electricity Synergy in China" and was awarded the Most Popular Speaker. The speech content is now compiled as follows (with some deletions) for reference by data center practitioners.
01. Policy Interpretation
Let's look at a set of data first. Opinions vary on the proportion of electricity consumed by domestic computing centers to total social electricity consumption, with figures ranging from 5.0% to 3.5%. Compared with traditional high-energy-consuming industries such as electrolytic aluminum and photovoltaic modules, this proportion is not large, but it is one of the fastest-growing sectors in total social electricity consumption. For this reason, the government has listed it alongside the eight major high-energy-consuming industries and has been strengthening energy efficiency control in recent years.
Looking at global trends, by 2025, the proportion of electricity consumed by AI-related businesses in global data centers will surge from 2% to 10%; by 2030, the annual power consumption of intelligent computing will account for 5% of global power generation. Therefore, there is a saying in the industry—the end of AI is energy. Let us add another sentence: the ultimate solution for energy is new energy.
The synergy of computing power and electricity is the result of the resonance of the transformation of the two major industries. Reviewing the evolution of the computing power industry: 20 years ago, data centers were network machine rooms of operators, which later evolved into IDC (Internet Data Centers), and then emphasized energy conservation, efficiency improvement, and PUE reduction. Now, the focus is on green computing power, emphasizing the substitution of energy consumption structure.
The same applies to the power side. Initially, it was the separation of power plants and the grid, with green electricity mainly supplied to industrial scenarios. Many large private enterprises and chemical companies built their own gas power generation and thermal power equipment within their parks. Later, the government promoted power marketization reform, leading to increased electricity price fluctuations, and energy storage became very popular for a time. During the "15th Five-Year Plan" period, the construction of a new power system was heavily emphasized.
The core of the new power system is new energy, and the combination of computing power and electricity, especially the integrated development with new energy, is the synergy of computing power and electricity. The two have a two-way spatial and temporal synergy relationship—computing power requires green, low-cost power supply, while power requires stable load consumption. Surveys of "Five Major and Six Small" energy enterprises show that the curtailment rate of wind and solar power in some regions has exceeded 30%, and for many medium-sized new energy enterprises, the curtailment rate in local areas has even reached 50%. Computing power can precisely provide a relatively stable consumption load.
In January 2025, the National Data Administration explicitly proposed promoting the proportion of green electricity consumption in computing power facilities; the central enterprises' "AI+" policy has promoted large energy and infrastructure central enterprises to shift towards computing centers and new energy fields, and now multiple central enterprises internally clearly support investments in computing-power and electricity synergy projects.
In May 2025, the "Notice on Matters Concerning the Orderly Promotion of Green Electricity Direct Connection Development" was issued, proposing two methods: grid-connected and off-grid, without distinguishing between single/multi-user direct connection at that time; in July, regarding the coordinated layout of energy bases and computing power clusters, the state issued relevant instructions for the eight major clusters; in September, the release of Document No. 1192, "Notice on Improving the Price Mechanism to Promote the Nearby Consumption of New Energy Power Generation," had a significant impact on the pricing logic of green electricity direct connection; it is particularly worth mentioning that in June 2025, the National Energy Administration listed the synergy of computing power and electricity as the first batch of pilot directions for the new power system.
After the Two Sessions, relevant policies further accelerated their implementation: the synergy of computing power and electricity was included in the national new infrastructure project; the rigid requirement of 80% green electricity was elevated from an industry initiative to national will; multi-user green electricity direct connection prioritizes support for computing power facilities, meaning a GW-level energy base can supply energy to multiple users within an entire computing power cluster, which has become a direction explicitly encouraged by the state.
02. Typical Cases
Let's first look at the technical characteristics of GPU power load.
Schematic Diagram of Typical GPU Load Curve
Training and inference are completely different.
Training: High base load, with GPU utilization continuously at a high level. However, note that it has pulse fluctuations—for example, GPUs alternate between the computing phase and the inter-card communication phase; when zoomed in, the load shows microsecond-level sawtooth fluctuations. Macroscopically, when switching from one task to another, it presents a step-like leap.
Inference: Short cycle, strong fluctuation, directly related to business request volume and modalities.
These characteristics determine the stringent requirements of intelligent computing on power supply—it requires both continuous high-reliability power supply and the flexibility to handle instantaneous fluctuations. This differs greatly from the smooth power load of general computing.
Let's look at commercial demands again.
First, the cost per kWh must be lower than 0.5 RMB, otherwise it will be uncompetitive nationwide. Currently, the cost of green electricity in the western region is 0.3-0.4 RMB, but going through the traditional grid, adding dispatch fees and fund subsidies, it is hard to control. In the past, it relied on government subsidies, but now fiscal pressure is high, and cost closure must be achieved through model reconstruction.
Second, timeliness of delivery. It takes 2-3 years from land acquisition to operation for a computing center, while photovoltaic and wind power projects can be built within half a year after approval. Many new energy projects cannot find incremental computing power loads after operation, resulting in a situation of "having electricity but no computing."
Third, regional coordination. The synergy of computing power and electricity involves multiple elements such as computing power, land, power grid, government, and new energy indicators. It is necessary to coordinate multiple resources to truly implement it.
Typical Cases
Zhongwei Zero-Carbon Green Power Park. This is currently a relatively mature large-scale multi-user green electricity direct connection grid-connected engineering practice. The approach is: the government gives new energy enterprises 2GW and 2.6GW indicators; new energy enterprises and computing power service providers establish a joint venture; a part of the green electricity is directly connected to achieve low-cost power supply, and a part of the green electricity trading feeds back the electricity price difference of existing data centers, bringing the existing electricity price down to 0.36 RMB. Ultimately, it achieves a win-win closed loop, a very interesting model.
Ulanqab: Computing power service providers bind strategic cooperation with capital parties, and capital parties invest in energy. This belongs to point-to-point single-user green electricity direct connection, grid-connected method.
A project in Qinghai: The future development potential in the Qinghai region is very large and deserves close attention. This case is off-grid, with dedicated line direct supply and a high energy storage ratio, but the overall economy is average, and sustainability remains to be considered.
Overseas: Google, Meta, etc., either build their own new energy, acquire it, or take equity stakes. They will more frequently use capital binding methods.
03. Mainstream Models
As shown in the figure, it is a panoramic matrix view of computing-power and electricity synergy drawn by Huaxin based on extensive practical experience. We divide the synergy models into two dimensions: divided into centralized and distributed according to the spatial layout level, and divided into market-led and technology-led according to the leading force. This chart is also very interesting.
Currently, there are five mainstream models
1. Green Electricity Direct Connection (Centralized, Technology-Driven): It has the greatest potential, mainly using centralized wind power/photovoltaic, and of course, biomass, such as waste incineration power generation, is also beginning to be adopted. It requires high technical standards.
2. Microgrid (Distributed, Technology-Driven): It has very high requirements for intelligent dispatching.
3. Virtual Power Plant (Both Distributed and Centralized Energy): You can understand it as a regional aggregation of N microgrids, achieving broader intelligent dispatching through a large platform.
4. Green Electricity Trading (Market-Oriented, but with Certain Technical Content): It is the current mainstream model, but green electricity trading often cannot reduce electricity costs.
5. Green Certificate Trading (Purely Market-Oriented): Green certificates still have issues such as carbon emission reduction certification domestically and internationally, and the integration of electricity and carbon has not been fully realized.
In addition, there is distributed nuclear power, but China is relatively cautious about the development of nuclear power, while overseas may be more aggressive.
From the perspective of development potential (circle size), grid-connected green electricity direct connection is the most popular and has the lowest cost per kWh.
04. Reflections and Outlook
Combining the above demand characteristics of computing-power and electricity synergy and domestic and foreign practical situations, we propose a practical architecture—a three-tier power supply system of "green electricity direct connection main supply layer + source-grid-load-storage support layer + large power grid bottom-line layer":
Main Supply Layer: Grid-connected green electricity direct connection. Relying on centralized new energy bases, green electricity is directionally transmitted to intelligent computing clusters through point-to-point physical dedicated lines, forming a clear and traceable power supply channel, while reducing energy storage pressure and improving stability through "large grid backup capacity."
Support Layer: "Source-Grid-Load-Storage" integrated model. Adopting the coupling model of "wind and solar + energy storage + intelligent computing load," it provides flexible regulation and dynamic balance for the system in scenarios where new energy fluctuates, load fluctuates, and local power grid regulation pressure increases.
Bottom-Line Layer: Public power grid double-circuit backup. It provides high-level guarantee for the continuity of intelligent computing businesses under extreme conditions.
Secondly, select an adapted business model to promote collaborative cost reduction across the entire industry chain of electricity prices.
• Long-term Agreement Trading (PPA): Computing power enterprises and power enterprises sign long-term agreements, investing and constructing respectively. Computing power locks in low-cost green electricity, and power obtains stable consumption. Advantage: Simple, flexible, and independent risk.
• Joint Venture: Parties establish a project company according to the equity ratio, sharing risks and benefits. Advantage: Deep binding, which can coordinate the incremental load of computing power and the rhythm of power construction.
• Cross-Shareholding: Long-term positioning, risk isolation, without affecting main business operations. It will become one of the trends in the future.
Ensuring power supply and price is the result of joint efforts across the entire industry chain. It requires government promotion, coordination with the power grid, comprehensive consideration of multiple scenarios such as green electricity trading and direct connection, and joint efforts of multiple entities including energy suppliers, electricity sellers, and the power grid to promote technological innovations such as nearby energy supply, voltage level improvement, and power supply architecture perfection, achieving collaborative cost reduction across multiple links in the whole chain.
Finally, summarizing in three sentences:
First, 2026 is the first year for the true large-scale implementation of the synergy of computing power and electricity and green electricity direct connection.
Second, grid-connected multi-user green electricity direct connection will become the mainstream model.
Third, creating a business model that connects the interests of multiple entities is the key to success or failure—the synergy of computing power and electricity is ultimately an "economic account."