The open-sourcing of DeepSeek V4 has sparked another wave of cost reduction at the model layer. It is a consensus that cheaper models lead to more frequent usage. However, the true industrial transformation lies in the latter part of the statement: the more frequent the usage, the more prominent the scarce value of vertical-specific tokens becomes. As top-tier open-source models become readily accessible, the core of competition has quietly shifted from "who can develop a better model" to "who can create more valuable tokens." Xunce Technology (03317.HK) is positioned at the forefront of this inflection point.
**First Half: Competing on Models; Second Half: Competing on Tokens** Over the past two years, the main theme of the large language model arms race has been parameter scale and the debate between open-source and proprietary models—whoever had the bigger, stronger, and cheaper model stood at the top of the food chain. The emergence of DeepSeek V4 essentially signals the end of this race: when top-tier open-source models are within easy reach, the scarcity of the models themselves disappears. The focus of competition swiftly shifts accordingly—it is no longer "which model do you use," but "how well do your tokens fit the specific scenario?" The logic is straightforward: the same token, used for casual conversation, might be worth $0.01 per million tokens; used for reviewing legal documents, it could be worth $1,000 per million tokens—a difference in value of a hundred thousand times. To use an analogy, if general-purpose large models are "power plants," then companies that refine data for vertical industries are "oil refineries"—and the pricing power for the refined vertical tokens ultimately lies in the scarcity of the application scenario. DeepSeek making "electricity" cheaper does not make refineries less valuable; on the contrary, lower electricity costs widen their profit margins. As the "refining layer" connecting large models to vertical scenarios, Xunce is in a pivotal position. Upstream model providers need Xunce's industry data for feedback, while downstream clients rely on its precise token services. This two-way dependency is expected to enhance Xunce's revenue share and pricing power.
**Xunce's ARR Skyrockets 300% QoQ as Token Price Increase Effect Accelerates** What DeepSeek V4 brings is a cheaper general-purpose computational foundation and more diverse application scenario entry points. More scenarios create more inelastic demand for vertical token calls; cheaper models encourage enterprises to deploy AI agents on a larger scale, and the consumption of high-quality industry data behind each agent is precisely Xunce's product. Xunce Technology uses AI Data Agent as its infrastructure to refine raw, private data from vertical industries like finance, telecommunications, and power into standardized vertical tokens that can be directly utilized by large models. These high-quality vertical data act as force multipliers for each token call, ensuring tokens are precisely consumed where they create the most value. Reports indicate that the call price for Xunce's vertical tokens has reached $10-$100 per million tokens, over ten times that of Anthropic's, and continues to grow based on specialized usage scenarios. Data is already confirming this trend: In 2025, the company achieved operating revenue of 1.285 billion yuan, a year-on-year increase of 103.28%. In April 2026, Xunce's Annual Recurring Revenue (ARR) related to token calls surged 300% quarter-over-quarter. The proportion of revenue from token-based payments has already exceeded 5%, with a full-year target to increase it to 20%-30%. This trend indicates that token call revenue is becoming a new, structural growth engine for Xunce and the industry, accelerating the company's transition towards an exponential growth model deeply tied to token call volume. The open-sourcing of DeepSeek V4 accelerates the commoditization of the model layer, thereby strengthening the scarcity premium of vertical tokens. As this understanding corrects itself at the market level, Xunce is poised to experience both performance delivery and valuation reassessment simultaneously—the classic conditions for triggering a Davis Double.
Comments