In June 2026, the "7th China Data Center Green Energy Conference" hosted by CDCC was successfully held in Shanghai. Given the two major development trends of high-density computing power and the construction of a new power system, computing-power and electricity synergy became the core topic of this conference. Zhang Xianglong, Deputy Director of the Energy Technology Research Center, Building Energy Branch of China Information Technology Designing & Consulting Institute (Shanghai), was invited to attend and delivered a speech titled "Technical Pathways and Core Challenges of Computing-Power and Electricity Synergy." His speech won the "Most Popular Speaker" award. The content of the speech is now compiled as follows (with some deletions) for reference by data center practitioners.
I. Policy Background: Why Promote "Computing-Power and Electricity Synergy"?
This year, computing-power and electricity synergy was written into the national government work report for the first time, marking the formal elevation of this model from industry exploration to a national top-level strategy. Its development necessity can be interpreted from the current status of the computing power and power industries respectively.
01. Computing Power Side: Surging Energy Consumption
Given the rapid iteration of GPU chips, the traditional 10kW power standard for cabinets can no longer meet current intelligent computing demands. The power of mainstream intelligent computing cabinets has generally reached 40-60kW, ultra-high-density computing nodes have exceeded 100kW, and the power per cabinet may move towards the megawatt level in the future.
The scale of computing clusters continues to expand. Intelligent computing centers are gradually iterating from 10,000-card clusters to ultra-large-scale computing clusters of 100,000 and 1,000,000 cards; supporting power supply capacity has also been upgraded from tens to twenty megawatts to hundreds of megawatts or even gigawatt (GW) levels. In this context, the growth rate of electricity consumption in the computing power industry continues to be higher than the average electricity consumption growth rate of the whole society, and the proportion of computing power electricity consumption in the total electricity consumption of the whole society is increasing year by year.
Industry institutions have completed forecasts for computing power electricity consumption in 2030 under three scenarios:
(1) High-growth scenario: The annual electricity consumption of computing centers will exceed 700 billion kWh, accounting for 5% of the total electricity consumption of the whole society;
(2) Medium-growth scenario: The annual electricity consumption will be 400 billion kWh, accounting for 3% of the total electricity consumption of the whole society;
(3) Low-growth scenario: The annual electricity consumption will be 300 billion kWh, accounting for 2.3% of the total electricity consumption of the whole society.
02. Power Side: Enhanced Renewable Energy Integration
In 2021, the state proposed building a new power system with renewable energy as the main body. The installed capacity of domestic wind and solar renewable energy has experienced explosive growth. By the end of 2025, the total installed capacity of renewable energy nationwide will account for 60% of the total power installed capacity, an increase of 20 percentage points compared to the early stage of the "14th Five-Year Plan".
However, wind and solar power have strong regional characteristics and volatility, "relying on weather conditions for power generation." Domestic high-quality wind power resources are concentrated in the Northeast, Northwest, and Southeast coastal regions, where the annual utilization hours of wind power can reach 3,000-5,000 hours, while the new energy resource endowment in the central and eastern inland areas is relatively weak.
There is a serious spatial and temporal mismatch between new energy output and the electricity load of the whole society, directly leading to the continuous existence of wind and solar curtailment. Statistical data shows that from January to March this year, the curtailment rate of wind and solar power in many domestic provinces remained in the range of 10%-40%. Although the state has laid out UHV (Ultra-High Voltage) transmission lines such as Longdong-Shandong and Hami-Chongqing to open up new energy export channels, the construction investment scale of UHV lines is large and the construction cycle is long. Coupled with multiple constraints such as land requisition, ecological environmental assessment, and cross-provincial interest coordination, relying solely on UHV is difficult to fully integrate remote new energy.
On one hand, there is a surge in computing power electricity demand; on the other hand, there is the dilemma of new energy integration. Therefore, promoting the aggregation of computing facilities in areas rich in green electricity resources to achieve efficient synergy between "computing" and "electricity" is not only a deepening of the national "East Data and West Computing" project, but also an inevitable choice for achieving green and low-carbon development.
Since 2021, the state has continuously issued multi-level policies to guide the aggregation and layout of computing infrastructure in areas rich in green electricity, and continuously improved the implementation details of computing-power and electricity synergy. A series of policies have advanced layer by layer, continuously improving the top-level design of computing-power and electricity synergy, providing clear policy support for industry implementation.
II. Technical System: Building a Five-Dimensional Synergy Framework
The core logic of computing-power and electricity synergy is to connect computing power, power, network, and park systems to form a five-layer synergy framework of space, time, scheduling, market, and source-grid-load-storage, supported by four major technical tracks to ensure overall operation.
01. Five Core Synergy Frameworks
(1) Spatial synergy: Adhere to "East Data and West Computing," guide the eastern computing power demand to transfer to the western green electricity-rich areas, and realize the spatial complementarity of computing power and energy.
(2) Temporal synergy: Computing tasks can be adjusted in real time according to electricity prices or new energy output, and combined with energy storage technology for peak shaving and valley filling to achieve temporal matching.
(3) Scheduling synergy: Promote the deep integration of computing power scheduling and power scheduling, achieve "source follows load, computing follows optimal electricity," and realize global optimization.
(4) Market synergy: Establish market mechanisms such as green electricity trading, computing power trading, and virtual power plants to achieve a value closed loop through market driving.
(5) Source-grid-load-storage synergy: Realize two-way mutual assistance between computing power and electricity, ensuring more reliable, greener, and more economical energy use for computing centers.
02. Detailed Explanation of Four Subdivided Technical Tracks
(1) Computing Power Side
The underlying foundation of computing power side synergy technology is the modeling of computing load characteristics, clarifying the mapping relationship between different computing tasks, the operating status of hardware clusters, and real-time power consumption, and building a computing load prediction model. On this basis, two core technologies are developed: first, unified computing power orchestration and scheduling technology, which standardizes labels and unified scheduling rules for different GPU architectures, hardware models, and cluster resources to realize unified allocation of computing resources across computer rooms and regions; second, multi-dimensional computing power optimization scheduling strategy, where scheduling decisions simultaneously take into account the idle rate of computing resources, regional real-time electricity prices, and local green electricity output, prioritizing the scheduling of elastic computing tasks to computing nodes with lower electricity prices and sufficient green electricity resources.
(2) Power Side
The power side has two core objectives: selecting the optimal green electricity access mode and achieving efficient green electricity integration within the park.
The underlying support is a new energy output prediction model, which accurately predicts wind and solar power output combined with meteorological and historical power generation data. Superimposed with electricity market price prediction, it forms a green electricity application auxiliary decision-making system, automatically comparing the comprehensive benefits of solutions such as direct green electricity purchase, distributed photovoltaics, park microgrids, and direct green electricity connection to select the optimal path. After green electricity is connected to the park, relying on new power supply architectures such as 800V full DC power supply and integrated energy storage and distribution, it improves the green electricity integration efficiency in high-density computing scenarios, which is also a key power technology direction currently promoted in the industry.
(3) Network Side
Focusing on the two core needs of network security and deterministic low-latency transmission. Both computing power and power grid scheduling data belong to highly sensitive production data. Once cross-system interaction suffers a network attack, it is extremely easy to cause major safety accidents such as computing power interruption and power grid load out of control. It is necessary to strictly follow the national standards related to the security of power monitoring systems and build a complete security protection system with hierarchical isolation and encrypted transmission. At the same time, real-time computing power scheduling and second-level load response of virtual power plants have strict requirements for network latency and jitter indicators. It is necessary to build an industry-specific deterministic transmission dedicated network to ensure that scheduling instructions are issued at the millisecond level.
(4) System Synergy Side
System synergy is divided into two levels: internal small synergy within the park and large-scale cross-regional synergy.
Internally within the park, relying on multi-energy complementarity and green microgrids, it integrates photovoltaics, energy storage, diesel generators, and computing loads. At the same time, it optimizes the matching between cooling and computing loads through intelligent cooling-power linkage. In terms of energy storage configuration, a cascaded layered application scheme is adopted, relying on supercapacitors on the server side to smooth out millisecond-level instantaneous fluctuations, and then using lithium battery energy storage in the power distribution room to complete minute-level peak shaving and valley filling. In addition, various loads in the park can also be layered and aggregated to connect to the power grid to participate in virtual power plant response to obtain peak shaving subsidy revenue.
At the cross-regional synergy level, the focus is on solving the connection difficulties caused by the misalignment of planning cycles between computing centers and the power grid, and their belonging to different competent departments. On the one hand, it coordinates in advance the docking of computing site selection, new energy power station layout, and power grid expansion planning. On the other hand, it builds a multi-objective optimization scheduling model that takes into account carbon emissions, electricity costs, and power supply reliability, ultimately opening up a national unified computing and power trading market, and relying on a market-oriented long-term mechanism to continuously drive the stable operation of computing-power and electricity synergy.
III. Main Pathways: Implementation Practice from Cabinets to Systems
In terms of specific technical pathways, it can be promoted layer by layer from servers and data centers to the system level.
Cabinet Level: Smoothing Instantaneous Fluctuations
Measured data shows that the computing operation load has an instantaneous fluctuation of 1.3-2.5 times. In response, supercapacitors can be integrated inside the server, relying on its millisecond-level fast charge and discharge characteristics to directly smooth the instantaneous power impact generated by a single server. At the same time, lithium battery energy storage is matched on the UPS and power distribution side, which can not only provide minute-level emergency power supply guarantee and complete second-level load smoothing, but also use the peak-valley electricity price difference to engage in arbitrage and increase revenue.
Data Center Level: Building a Green Microgrid
Promoting the 800V full DC power supply system to adapt to high-density computing loads, and it is also convenient to directly connect distributed new energy and energy storage equipment on the DC side, comprehensively improving the flexible regulation capability of the data center itself. Relying on this flexible power supply foundation, the data center can also package its own adjustable loads to form a virtual power plant and carry out real-time load interaction with the power grid.
In addition, building distributed photovoltaics within the park is the most mature and conventional solution for implementing green microgrids, but there are clear regulatory requirements for implementation: photovoltaic-related buildings and supporting sites must belong to the same land use red line, and the overall project must be uniformly planned and filed to avoid the problem of splitting plots and fragmenting projects to evade regulation.
Regional Level: Exploring Direct Green Electricity Connection and Backup Power Sharing
Direct green electricity connection is an important pathway to promote the nearby integration of new energy. It can be divided into two types: grid-connected and off-grid. The actual revenue of project implementation is mainly constrained by two core factors: the standard of transmission and distribution fees and the investment cost of supporting energy storage. On the basis of direct green electricity connection, the industry can also innovatively explore the regional backup power sharing model, replacing the diesel generator sets individually configured by each data center through the unified construction of centralized energy storage stations. During daily operation, new energy power generation and centralized energy storage supply power collaboratively. Once the mains power is interrupted, the energy storage will first provide hour-level stable backup power. If a long-term power outage scenario is encountered, mobile diesel generator sets will be dispatched to supplement the power supply. This new model can not only realize clear traceability of the whole process of green electricity and ensure the effect of local integration, but also fundamentally resolve the industry pain points of high idle rate of traditional fixed diesel generators and excessive early construction investment costs, achieving a dual improvement in economy and reliability.
System Level: Realizing Refined Computing Power Scheduling
Not all computing tasks are suitable for scheduling. According to latency sensitivity and interruption tolerance, they can be divided into four categories:
(1) Real-time response type (e.g., financial transactions): High latency requirements, uninterruptible, not suitable for scheduling.
(2) High-concurrency online type (e.g., short videos): High latency requirements, not suitable for scheduling, but power capping can be done inside the server.
(3) Long-cycle high-intensity training type (e.g., large model training): Relatively low latency requirements, can be planned in advance and migrated across regions.
(4) Opportunistic elastic tasks (e.g., data backup): Insensitive to latency and interruption, most suitable for computing power scheduling.
IV. Core Challenges: Four Checkpoints on the Way Forward
Despite the broad prospects, the implementation of computing-power and electricity synergy still faces many challenges:
(1) Business model level: The investment cost of direct green electricity connection projects is high; computing power scheduling across cloud providers is difficult; the market mechanism for computing-power and electricity synergy is not yet clear, lacking economic driving force.
(2) Planning synergy level: Computing power and power planning belong to different systems, lacking an effective synergy mechanism, leading to difficulties in matching resources and site selection.
(3) Resource and demand matching level: Direct green electricity connection requires sufficient and close-proximity new energy resources around computing facilities, making location matching difficult; moreover, computing loads grow gradually, bringing challenges to project economic evaluation.
(4) Network operation level: New models such as direct green electricity connection and virtual power plants bring new operational risks and uncertainties, placing higher demands on operation personnel.
My sharing ends here, thank you all!