While the global artificial intelligence sector is currently embroiled in a brutal 'race to the bottom' regarding token pricing, Hong Kong-listed XunCe (03317.HK) is charting a radically different course. As general-purpose models from the likes of DeepSeek slash prices to fractions of a cent, XunCe has successfully positioned its specialized 'Scenario Tokens' at a premium range of $10 to $100 per million tokens. This pricing structure is notably more than ten times higher than the rates offered by industry leaders like Anthropic, signaling a shift toward value-based vertical AI.
The strategy appears to be paying off in immediate financial terms. For the quarter ending April 2026, XunCe reported a staggering 300% quarter-on-quarter surge in Annual Recurring Revenue (ARR) specifically from its token invocation services. This growth trajectory has prompted the company to revise its revenue targets upward, now aiming for token calls to account for 20% to 30% of total annual revenue, up from a previous modest goal of just 5%.
This divergence in pricing models highlights a maturing AI market in China where 'commodity' intelligence and 'domain' intelligence are separating. While general-purpose LLMs are becoming a utility with vanishing margins, XunCe’s high pricing power suggests that enterprise clients are willing to pay a massive premium for models that are deeply integrated into specific workflows and complex industrial or financial scenarios. The company's ability to maintain these margins in a deflationary environment indicates a significant competitive moat built on specialized data and operational relevance.
The broader context of this surge is a domestic market where DeepSeek has recently driven the cost of general-purpose tokens down to near-zero levels. By successfully decoupling its pricing from the general market trend, XunCe is demonstrating that the future of profitable AI may lie not in the scale of the model, but in the specificity and high-stakes nature of the application. Investors are now watching closely to see if this high-margin 'Scenario Token' model can be sustained as more players attempt to pivot from general to vertical AI services.