On April 28, the Ministry of Industry and Information Technology and the National Data Administration jointly issued a notice to implement the "Model-Data Resonance" initiative by 2026. The program targets over twenty key sectors, including steel, petrochemicals, non-ferrous metals, power equipment, information and communications, and cybersecurity, aiming to systematically foster a positive cycle of interaction between data, models, and practical applications. The notice requires each industry to identify no fewer than five high-quality general-purpose datasets, construct at least one specialized dataset per application scenario, and develop a minimum of one dedicated model or specialized intelligent agent per scenario. This indicates that more than twenty critical Chinese industries will, under policy guidance, collectively undergo a structural build-out of vertical industry data assets. Thousands of specialized datasets and unique intelligent agents are expected to be deployed in volume within the next year. The ultimate measure of success for this wave of AI industrialization will be the return on investment, determined by how effectively the daily Token consumption translates into tangible business decision-making value. The policy directly addresses the core issue of "Token conversion efficiency."
XUNCE's (03317) deep, over-a-decade-long expertise in high-barrier industries has created a formidable data moat, which is now emerging as a critical variable affecting the ROI of large-scale model industrialization.
1. After AI Democratization, What Are Companies Really Competing On? A paradox is emerging regarding the returns from deploying large models: while the cost of accessing general-purpose large models is approaching zero, allowing nearly all enterprises to utilize AI equally, a growing number of companies are finding that their AI investments are not yielding the expected returns. The reason is straightforward. General-purpose models rely on public data, but when AI is applied to core business scenarios like healthcare, manufacturing, and energy, the signal-to-noise ratio of generic data drops sharply. Using general models to process data results in far lower accuracy compared to vertical models trained on a decade of industry-specific data. For instance, a manufacturing firm using a general model for equipment failure analysis would get significantly inferior results compared to a specialized system integrated with real industrial control logs. Huatai Securities notes that AI pricing power is shifting from the compute layer to the application layer, with the premium for high-value vertical Tokens strengthening continuously. Data from Tencent Research Institute shows a vast disparity: Tokens for casual chat cost just $0.01 per million, while those for legal document review can reach $1,000—a 100,000-fold difference in value. This signifies that "Token conversion efficiency" is becoming the core variable for ROI in corporate AI deployment. Companies unable to build high-quality vertical data assets will see their AI investments degenerate into inefficient battles of Token consumption.
2. XUNCE's Moat: Not Faster, but More Accurate in "Conversion" In the race for Token conversion efficiency, XUNCE's advantage lies not in marginal algorithmic speed gains but in the vertical industry data assets accumulated over more than ten years—core infrastructure essential for large models operating in real business environments. This advantage manifests in three key areas: First, a barrier created by data scarcity. XUNCE's vertical Tokens are priced between $10 and $100 per million Tokens, more than ten times the rate of providers like Anthropic. Its deep integration into client business processes creates a very strong moat. Second, the shortest conversion path. A general Token might require 100 API calls to produce a correct result, whereas a XUNCE vertical Token often achieves the desired outcome in just one call. Clients evaluate the total cost—while the unit price is higher, it eliminates 99% of ineffective calls, leading to an overall reduction in expense. Third, continuously strengthening network effects. Daily Token usage in China has surpassed 140 trillion, growing over a thousandfold in two years; J.P. Morgan predicts AI inference Token consumption will grow 370-fold by 2030; IDC forecasts a 139% increase in global AI Agents over five years. Every operation of an intelligent agent expands the application scale for vertical Tokens.
3. The "Oil Refinery" Analogy: Tokens Cost Ten Times More but Save 99% of Calls The primary challenge for companies using AI is not API accessibility but the poor quality, fragmentation, and disorganization of their own data. Critical operational data is often siloed across dozens of incompatible systems, and vertical industry knowledge is buried in documents and processes, making it unusable for large models. This is akin to having crude oil but no refinery; even the best engine cannot run. XUNCE acts as the "refinery." It processes messy, raw data from sectors like finance, telecommunications, manufacturing, robotics, and commercial aerospace through a combination of real-time data infrastructure, Retrieval-Augmented Generation (RAG), industry knowledge graphs, and Agent orchestration, "refining" it into high-quality vertical Tokens that models can directly utilize. Where a general Token might need 100 calls, a XUNCE vertical Token often succeeds in one. Clients focus on the total cost, not the unit price. Is it ten times more expensive? Perhaps, but by saving 99% of无效 calls, the comprehensive cost actually decreases. This is the foundation of XUNCE's pricing power—it's not about charging more but about delivering greater value per dollar spent.
4. Repricing: From "Cost Center" to "ROI Engine" Thus, XUNCE is a provider of AI conversion efficiency. In the AI deployment chain, general-purpose models are responsible for "producing Tokens," while XUNCE specializes in "purifying Tokens"—distilling and enhancing generic Tokens with specialized data to convert them into high-value Tokens directly applicable to corporate decision-making. The efficiency of this step directly determines the overall ROI of a company's AI investment. The company's stock has surged 658% year-to-date, with its market capitalization exceeding HKD 100 billion. However, compared to its strategic position within the AI industrialization ecosystem, its current valuation might only represent the starting point of a long-term re-rating. One significant, yet unpriced, data point is that Chinese models already account for 61% of the total Token consumption among the top ten models on the OpenRouter platform. As the global number of AI Agents explodes (IDC predicts 22.16 billion by 2030), worldwide demand for Chinese vertical data Tokens is set to expand dramatically.
Conclusion: The Endgame of the Token Consumption War is a Battle of Conversion Efficiency In the war of Token consumption, the true victors will not be those offering the lowest unit price, but those enabling each Token to generate the highest business value. As the "Model-Data Resonance" policy mandates a data refinement cycle across over twenty industries and thousands of specialized datasets and agents come online, the critical success factor for AI industrialization is no longer "who has the compute" or "who has the model," but "who achieves the highest Token conversion efficiency." XUNCE's 300% quarter-over-quarter surge in Annual Recurring Revenue is the clearest signal yet:稀缺 assets in the age of efficiency are being repriced.
Comments