Token Overseas Expansion Surges, Green Energy Sector Strongly Leads Gains, Computational Power Synergy Stands at New Opportunity

On March 11, the green energy sector defied the trend and saw multiple stocks hitting the daily limit. In recent days, market attention on the green energy sector has continued to heat up, attracting significant capital. The core driver behind this shift comes from the elevation of “computing and electricity collaboration” from local pilot projects and departmental policies to a top-level national design: this year’s government work report proposed implementing large-scale intelligent computing clusters and computing-electrical synergy as part of new infrastructure projects.

In the AI era, computing and electricity collaboration is becoming a key factor and engine in global technological competition. “Who has cheap electricity, stable electricity, and quick grid adjustments, holds the cost advantage in the AI era,” commented People’s Daily. The essence of Token going overseas is to turn electricity into computing power, and then turn computing power into intelligence. “The end of AI is electricity, and the end of electricity is China.”

The rise of computing and electricity collaboration from industry buzzword to top-level design indicates that China is systematically transforming the vast advantages of its power system into core competitiveness and strategic moat of the digital economy. It is a deep “genetic integration” of two major national infrastructures. Roland Berger partner Qiu Zeyu analyzed to The Paper that with the explosion of AI large models, demand for computing power is skyrocketing, and data centers have become high-energy-consuming loads. The industry consensus is “the end of computing power is electricity,” but China is not short of electricity; what is lacking is precise matching of computing power and electricity. The core challenge for computing-electrical synergy is to solve the “structural and temporal mismatch” of electricity.

Several details worth noting: According to OpenRouter, a global major model aggregation platform, from February 16 to 22, China’s weekly token calls for AI large models reached 5.16 trillion, more than doubling in three weeks, surpassing the US large models in weekly scale. Developers worldwide have found that running tasks with Chinese large models is significantly cheaper than with American ones. By 2025, China’s total installed renewable energy capacity will reach 2.34 billion kilowatts, with nearly 40% of electricity coming from green sources for every 10 degrees of electricity used. Power costs at computing centers in western hubs like Ulanqab, Inner Mongolia, can be as low as below 0.3 yuan per kWh, creating a notable “green electricity lowland” effect. Under the “dual carbon” goals, national policies require that by 2025, over 80% of new data centers in national hubs use green electricity.

Lin Boqiang, director of the China Energy Policy Research Institute and chair professor at Xiamen University School of Management, believes that as the fundamental energy source for computing power, the stability, economy, and sustainability of electricity supply directly impact the development prospects of the computing industry.

Besides “the end of computing power is electricity,” another industry saying is: “the end of electricity is green electricity.” Both electricity and computing power have peaks and troughs; aligning their rhythms creates a win-win situation.

Qiu Zeyu stated that high proportions of green electricity require smarter scheduling mechanisms to meet computing demands. The essence of computing-electrical synergy is to dynamically match and intelligently dispatch computing and electricity resources, enabling efficient consumption of green power and low-carbon operation of data centers. The signal from this upgrade is: “China is turning the ‘physical advantage’ of its power system into the ‘neural system advantage’ of the digital economy era.”

“Training large AI models requires massive computing power, and data centers are major power consumers. China has leading renewable energy generation capacity worldwide, and through energy storage to smooth fluctuations, it can provide stable green power for computing centers. Overseas companies accessing Chinese large models essentially achieve computing power exports or token exports, transforming energy advantages into digital service advantages and creating a global energy value loop,” said Tian Qingjun, senior vice president of Envision Group.

The market generally believes that Token exports will drive domestic computing capacity expansion, thereby boosting electricity demand.

Qiu Zeyu told The Paper that computing and electricity collaboration has moved from concept to pilot, entering a “pre-breakthrough night,” but progress is hindered by three major issues: mismatched planning across different departments, supply-demand mismatches, and data silos.

The challenges include: first, the planning system is not yet coordinated, with computing centers and power infrastructure planned by different departments; some data centers still face the “last mile” problem of power supply; second, the natural mismatch of supply and demand—computing is concentrated in the east while green electricity is abundant in the west, and AI training requires highly stable power, while renewable energy is intermittent; third, dispatching mechanisms and data are not yet integrated—data centers could serve as “dispatchable loads” for grid peak shaving, but data barriers between computing and grid sides hinder unified, efficient coordination.

To strengthen the bidirectional integration of computing and electricity, the key lies in technological empowerment and breaking down barriers. On the technical side, breakthroughs in multi-source data fusion and real-time scheduling optimization centered on energy large models are needed, along with promoting “mapping models” for precise conversion between computing and electricity. Energy storage technology, as a key “stabilizer,” can participate in peak regulation and generate revenue from peak-valley electricity price differences. Market mechanisms should prioritize expanding pilot projects for integrated dispatching and trading of existing computing centers, forming a quantifiable value loop, and linking interests across power generation, grid, sales, and computing enterprises, exploring new power supply models like “green electricity direct supply + main grid backup.”

“A unified power grid has already been built. Currently, we are constructing a unified national computing network. In the future, both must collaborate to improve utilization and reduce costs,” suggested Zhang Yunquan, member of the National People’s Congress and researcher at the Institute of Computing Technology, Chinese Academy of Sciences, during the Two Sessions. He recommends precise matching based on computing types: large model training tolerates higher latency and is less geographically sensitive, so it can be scheduled to western regions rich in renewable energy; inference computing requires strict low latency and should be deployed near eastern load centers; supercomputing tasks are sensitive to electricity prices and should be located in low-cost power areas. Additionally, non-real-time computing can be shifted to nighttime off-peak hours through temporal optimization to match “computing peaks” with “electricity valleys.”

Qiu Zeyu believes that there is no one-size-fits-all template for computing-electrical synergy; it must be “adapted to local conditions and diverse in models.” The core logic is to let computing centers “follow energy, build around the grid,” forming a precise match between resource endowments and industrial needs.

Based on China’s energy and computing layout mismatches, three typical collaboration models may emerge in the future: the western “green electricity local consumption” model, the eastern “flexible load dispatch” model, and the “multi-dimensional advantage stacking” model in special regions like Guizhou. For regions like the Yangtze River Delta and Beijing-Tianjin-Hebei, where computing demand is concentrated but electricity is tight, the focus is on unlocking the regulation potential of existing data centers, involving computing loads in grid peak shaving, and optimizing intelligent operations and cooling sources to achieve optimal matching of computing output and energy utilization. The key is to enable computing to “understand electricity price signals and be guided by market mechanisms to flexibly shift loads.”

Yang Jianyu, deputy of the National People’s Congress and chairman of China Mobile Zhejiang, pointed out that the Yangtze River Delta, Beijing-Tianjin-Hebei, and Pearl River Delta regions account for over 60% of national computing demand, but less than 20% of energy supply, with high external dependence on electricity. As AI applications rapidly expand, the power supply supporting intelligent computing centers in the east needs strengthening. He suggests: on one hand, continue shifting high-load, high-energy-consuming AI training demands to the west; on the other hand, promote “power following computing” by accelerating “west-to-east power transmission,” implementing cross-province and cross-region transmission channels, energy storage capacity, and infrastructure. Support eastern regions to increase power infrastructure investments based on local conditions to meet the immediate needs of AI applications.