Shangzheng Shen Yi Du | How to Reshape the Global AI Industry Landscape with China's Large Model Usage Reaching New Heights?

Since the beginning of this year, a “Lobster (OpenClaw) storm” has swept the globe. While operating, OpenClaw acts as the “scheduling brain” of large models, with its complexity leading to a dramatic surge in global calls for large models.

In this generational leap of AI large model application experience and scale, domestic large models have achieved a historic turnaround—by March 22, the call volume of domestic large models had unprecedentedly exceeded that of overseas large models for three consecutive weeks. The token consumption pressure brought by OpenClaw has led many overseas users to switch to Chinese large models, which are also leading in performance but offer better cost-effectiveness.

According to interviews conducted by Shanghai Securities Journal with companies and experts in the large model industry chain, the wave of globalization for domestic large models is still in its infancy. This not only helps the industry chain build new competitive advantages but also promotes the global large model ecosystem to continuously evolve towards a diverse symbiosis, accelerating the realization of global AI inclusiveness.

Insight: What Drives the Competitiveness of Chinese Large Models?

According to data from the global large model aggregation platform OpenRouter, last week (March 16 to March 22), the total call volume for global AI large models was 20.4 trillion Tokens, with a week-on-week growth of 20.7%. Among them, the call volume for Chinese large models was 7.359 trillion Tokens, up 56.9% week-on-week; the call volume for American large models was 3.536 trillion Tokens, an increase of 7.35% week-on-week.

Among the top nine large models by call volume, five are Chinese large models, namely Xiaomi Group’s MiMo V2 Pro, Jieyue Star’s Step 3.5 Flash, MiniMax M2.5, DeepSeek V3.2, and Zhipu GLM 5 Turbo.

It is noteworthy that the users of the OpenRouter platform mainly come from overseas, meaning the above rankings are primarily determined by votes from overseas users.

When discussing the reasons for breaking into the global market, MiniMax Vice President Yan Yijun told reporters: “This explosion is purely driven by reputation and communication brought about by high intelligence and high cost-effectiveness.”

In recent years, many Chinese large model companies have nearly “infinitely narrowed” the gap with the world’s top models through multidimensional innovation in model capabilities. Based on this, cost-effectiveness has become the core driving force for domestic large models to explode in the overseas market. “While the capabilities are on par with overseas leading models, the price of domestic models may only be a fraction of theirs, so users will naturally ‘vote with their feet,’” said Yan Yijun.

This cost-effectiveness is closely related to China’s advantages in electricity supply and pricing. Professor Lu Jing of Zhejiang University told reporters that electricity is an important resource endowment for AI, and the advantages of large-scale investment in electricity infrastructure in our country are beginning to manifest. The cross-border consumption of computing resources by large models is fundamentally driven by the export of electricity.

Song Yu, Vice President of Tiandis Zhixin, stated: “What truly determines competitiveness is not just the electricity price itself, but how many effective Tokens can be stably produced per unit of energy consumed.” This relies on the formation of a full-chain, globally coordinated system in China, from chip design to data center operation, from electricity scheduling to cluster management.

Cloud vendors represented by UCloud have already completed the global infrastructure layout overseas. According to Liu Jie, Vice President of UCloud, the company has now deployed 30 availability zones in 22 regions worldwide, covering key markets such as Southeast Asia, Europe, North America, South America, and Africa, providing Chinese AI companies with globally accessible low-latency, high-stability computing and network services, significantly reducing the physical threshold for Tokens to go abroad.

Entering the global market amplifies the advantages of China’s industrial system: it not only expands market space but also signifies an improvement in resource utilization efficiency. Liu Jie gave an example, stating that China’s nighttime computing resources are relatively idle, while Europe and America are experiencing peak usage during the day; this cross-timezone consumption of computing resources can effectively enhance GPU utilization and further leverage the advantage of relatively low domestic electricity prices.

Opportunity: Chinese Tokens Going Abroad Reshape the Global AI Landscape

Looking ahead, the industry consensus is that the scale of Chinese Tokens going abroad will continue to grow. Yan Yijun stated: “Whether domestically or globally, the current actual penetration rate of AI is still very low. This means we have an extremely large growth space ahead of us.”

This growth will strengthen the positive cycle of “application-driven innovation” across the entire industrial chain of domestic large models.

For large model companies, Lu Jing stated: On one hand, the active engagement of overseas users with our open-source large models is beneficial in forming a consumption path dependency, while a large number of users using them will also help drive rapid iteration and upgrading of domestic large models; on the other hand, although the current Token economy has not yet formed a mature profit model, the active engagement of overseas users contributes to achieving global revenue growth.

For cloud service providers, Liu Jie told reporters that the export of Tokens has brought a significant demand for services including computing power, network acceleration, and security protection at overseas nodes.

For chip companies, Song Yu believes that domestic general-purpose GPU companies are facing dual opportunities for large-scale commercialization and technological iteration: on one hand, the explosive demand for inference computing power brought by Tokens going abroad pushes inference chips from being “usable” to being “high cost-effective and scalable,” and leading chip companies with system-level energy efficiency and full-stack adaptation capabilities will gain stronger pricing power and stable gross margins; on the other hand, the diverse language, multi-modal, and low-latency demands of overseas scenarios are forcing chip companies to continuously innovate in aspects such as memory bandwidth and dynamic scheduling, accelerating technological upgrades.

Chinese Tokens going abroad will also promote the global large model ecosystem to continuously evolve towards a diversified symbiosis.

Liu Jie believes: On one hand, the export of Chinese large models will break regional technological monopolies, reduce global AI call costs through healthy competition, and accelerate the inclusivity of AI applications; on the other hand, companies in different regions will have more diverse model choices and will be able to select the most suitable model services based on their own needs.

The industry has also noted that this year’s government work report first mentioned “synchronized electricity and computing,” listing it alongside ultra-large-scale intelligent computing clusters as part of new infrastructure projects. The industry expects that market behaviors driven by the opportunities of Token going abroad will likely form a bi-directional empowerment with national strategies.