Artificial intelligence (AI) in China is entering a new phase, shifting from catching up on technological capabilities to capturing value, with the focus moving from training to inference, from technology to application, and from potential to actual profitability. This assessment aligns with broader policy direction. The People's Bank of China recently stated in its Q1 2026 Monetary Policy Execution Report that AI will inject new momentum into high-quality economic development through accelerated technological iteration, deepening industrial integration, and expansion into international markets. A genuine transformation in the path to AI commercialization is quietly unfolding at Xunce (03317), known as the "Token First Stock."
Business Model Upgrade: Token Payment Revenue Share Poised to Rise from 5% to 30%
Financial reports show Xunce Technology achieved operating revenue of RMB 12.85 billion in 2025, a year-on-year increase of 103.28%. Revenue in the first half of the year was only RMB 1.98 billion, surging to RMB 10.87 billion in the second half, a quarter-over-quarter jump of 449%, signaling a clear and concentrated surge in demand. The average revenue per user (ARPU) rose from RMB 2.72 million in 2024 to RMB 5.59 million in 2025, up 105.4% year-on-year. Revenue per employee increased 135.25% year-on-year to RMB 2.87 million. The annual gross profit margin remained stable at a high level of 61.66%, significantly above that of computing power providers and model developers during the same period.
Entering April 2026, Xunce's Token call ARR surged 300% quarter-over-quarter, with the share of Token payment revenue exceeding 5%. The annual target is to increase this share to the 20%-30% range. The significance of these figures lies not in absolute scale but in signaling a qualitative transformation of the business model: the company is systematically transitioning from a traditional subscription model to a Token payment and value-sharing model.
Management recently stated, "The core logic behind this target is that customer usage patterns are already changing. Previously, customers paid more for modules, solutions, or projects. But as AI truly integrates into business processes, customers are more concerned with how much is actually called, what results are generated, and the business value created."
According to Xunce's pricing logic, Token payment is not simply "charging per word" but is based on three dimensions: the price per call (depending on data scarcity, real-time requirements, and industry complexity), the number of calls (the customer's actual usage in real business operations), and module depth (the more modules integrated and the deeper the embedding into processes, the higher the overall value). Essentially, this is a mechanism that deeply ties the charging method to the real value created for the customer—"the deeper the customer uses, the more calls made, and the higher the requirements for results, the greater the value of the data capabilities we provide."
Management anticipates that with the explosion in AI inference demand, the Token payment model will gradually become a component of the company's revenue structure over the next two to three years, with its share expected to increase significantly.
AI Inference Phase: Premium for Vertical Tokens
The requirements for data quality in the AI inference phase are fundamentally different from the training phase: it demands real-time, precise, and measurable data supply, not massive but coarse-grained historical datasets. Xunce's core competitiveness is precisely built on this layer.
Xunce Technology has been deeply involved in real-time data infrastructure for a decade. Centered on AI Data Agent, it has built a full-chain technical system covering data acquisition, cleaning, standardization, real-time computing, and model optimization, spanning ten high-growth sectors including finance, robotics, and commercial aerospace. Leveraging private data operations accumulated in high-barrier industries, it serves as "high-quality data fuel" consumable at the AI application layer, model layer, and even operating system level, enhancing the business value per unit Token and forming a solid competitive barrier.
More critically, China's daily Token call volume has exceeded 140 trillion, growing over a thousandfold in two years. JPMorgan predicts AI inference Token consumption will grow 370 times by 2030. IDC forecasts global AI Agent growth of 139% over five years. Every intelligent agent running 24/7 is a new engine for vertical Tokens.
Currently, Xunce's vertical Token pricing reaches $10 to $100 per million Tokens, over ten times that of similar products from Anthropic. This pricing power is backed by data barriers formed through nearly ten high-value industry scenarios domestically. Management stated, "We will further expand into the robotics AI training data platform, commercial satellite computing platform, and aviation industry computing platform—each is a new multi-billion-yuan track, and the application scenarios in the Chinese market for these fields are also more abundant."
The core characteristic of the "value capture" phase for Chinese AI, as described in the Morgan Stanley report, is that AI capabilities are no longer just about technological prowess but are truly penetrating core business processes and forming sustainable commercial monetization. The upgrade of Xunce's Token payment model is a microcosm of this characteristic. As management summarized the company's value proposition: "Large models determine what AI can do; what Xunce determines is whether AI can create value sustainably, reliably, and scalably in real enterprise scenarios."
Transitioning from "selling data" to "selling Token calls," Xunce's performance and stock price are poised for a "Davis Double Play," targeting a market capitalization of RMB 200 billion!
Comments