This photo taken on March 11, 2026 shows the screen of a mobile phone running the open-source AI agent OpenClaw at Wuxing District of Huzhou City, east China's Zhejiang Province. (Photo by Yi Fan/Xinhua)
SHANGHAI, April 3 (Xinhua) -- In a downtown office building in Shanghai, nearly 200 AI engineers are busily fine-tuning one of the world's most popular large language models.
By deploying this homegrown product across data centers worldwide, tech firm MiniMax is gearing up to capture a slice of this year's explosive boom in token consumption.
The latest data from OpenRouter, a leading global AI platform, shows that MiniMax's new edition ranks among the top 10 AI models by token calls worldwide, alongside DeepSeek, Zhipu AI, and three other homegrown models.
Such dominance reflects a broader transformation in how AI is consumed globally. The surge of "OpenClaw," an agent-based AI system that functions as an orchestration layer for AI models, is dramatically increasing token throughput as users delegate complex multi-step tasks to AI assistants rather than engaging in simple chat interactions.
The emerging commercial potential of what has been dubbed the "token economy" is already evident in the encouraging financial reports from some Chinese AI companies.
On Tuesday, Beijing-based Zhipu AI released its first annual earnings report since going public, revealing that total revenue for 2025 surged 131.9 percent year on year, with an overall gross margin of 41 percent -- well above industry standards. Its standout performer was its token sales business, which skyrocketed 292.6 percent from one year earlier.
Last month, Hong Kong-listed MiniMax released its performance announcement, reporting that total revenue for 2025 jumped 158.9 percent year on year, with approximately 70 percent of that income coming from international markets.
"This explosion is fueled by the Chinese model's high intelligence and exceptional cost-performance," said Yan Yijun, vice president of MiniMax, whose team of just over 400 people also maintains a Beijing unit working on research and development.
"While matching the capabilities of top-tier overseas models, domestic alternatives are priced at perhaps one-tenth the cost or even less," explained Yan. This ecosystem empowers global users to deploy AI into their respective application scenarios more rapidly and affordably.
In another notable development, Kimi K2.5, developed by another Chinese AI model competitor, Moonshot AI, was recently adopted as the foundational AI engine by Cursor, a U.S.-based AI coding platform, marking overseas licensing partnerships as a new commercial avenue for Chinese model providers.
Liu Jie, vice president of cloud service provider UCloud, said this expansion allows users worldwide to access a more diverse range of model options, enabling them to select the most suitable services tailored to their specific needs.
"Whether domestically or globally, AI's actual penetration rate remains low. This means there is an enormous growth runway ahead of us," added Yan.
Chinese tech firms' open-source large AI models rank first globally in downloads, significantly lowering barriers to AI adoption, reducing usage costs, and enhancing AI accessibility, said Li Lecheng, Minister of Industry and Information Technology.
However, Zhipu AI CEO Zhang Peng emphasized that high cost-performance does not mean cutthroat price competition. After raising API prices by 83 percent in the first quarter of 2026, Zhipu AI saw call volumes rise rather than fall. MiniMax's Yan attributes the Chinese models' cost advantages to engineering efficiency.
Wei Wei, an official from the Ministry of Industry and Information Technology, said that China will accelerate the development of its open-source ecosystem through three pillars: application-powered growth, tech innovation and ecosystem refinement, thereby contributing Chinese strength to global AI advancement.
In the domestic market, token demand has also skyrocketed, partly driven by China's "AI plus" initiative and the explosive growth of agent-based deployment. Data from the National Data Administration shows China's average daily token calls surged from 100 billion in early 2024 to 140 trillion by March 2026.
To underpin AI growth, China's government work report this year unveiled plans to launch new infrastructure projects focused on hyperscale intelligent computing clusters, alongside the coordinated development of computing capacity and power supply.
China's robust clean energy infrastructure and aggressive expansion plans for the upcoming five-year cycle are poised to secure the critical power supply needed to meet the energy-intensive demands of AI computing.
Xia Lixue, CEO of computing infrastructure provider Infinigence AI, sees China becoming the world's "token factory" by capitalizing on its energy strengths. His vision involves a unique "token economics" model that creates a sustainable pipeline from raw power to computing might, and finally to growth.
A single purchase
